d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
9bf4c9c1b0
freebsd-dev/sys/dev/mpt/mpt_cam.c

5046 lines
136 KiB
C
Raw Normal View History

Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*-
* FreeBSD/CAM specific routines for LSI '909 FC adapters.
* FreeBSD Version.
*
* Copyright (c) 2000, 2001 by Greg Ansley
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
* 2. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
Role a microrev of the MPI Library in preparation for target mode work. Make my portions of the license clearer. Thank Chris Ellsworth for his support in getting a bunch of this done.
2006-02-25 07:45:54 +00:00
*/
/*-
* Copyright (c) 2002, 2006 by Matthew Jacob
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce at minimum a disclaimer
* substantially similar to the "NO WARRANTY" disclaimer below
* ("Disclaimer") and any redistribution must be conditioned upon including
* a substantially similar Disclaimer requirement for further binary
* redistribution.
* 3. Neither the names of the above listed copyright holders nor the names
* of any contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF THE COPYRIGHT
* OWNER OR CONTRIBUTOR IS ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*
Role a microrev of the MPI Library in preparation for target mode work. Make my portions of the license clearer. Thank Chris Ellsworth for his support in getting a bunch of this done.
2006-02-25 07:45:54 +00:00
* Support from Chris Ellsworth in order to make SAS adapters work
* is gratefully acknowledged.
Add acknowledgements to LSI-Logic for support
2006-05-29 20:34:28 +00:00
*
* Support from LSI-Logic has also gone a great deal toward making this a
* workable subsystem and is gratefully acknowledged.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
/*-
* Copyright (c) 2004, Avid Technology, Inc. and its contributors.
* Copyright (c) 2005, WHEEL Sp. z o.o.
* Copyright (c) 2004, 2005 Justin T. Gibbs
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce at minimum a disclaimer
* substantially similar to the "NO WARRANTY" disclaimer below
* ("Disclaimer") and any redistribution must be conditioned upon including
* a substantially similar Disclaimer requirement for further binary
* redistribution.
Correct attribution in clause three to address the correct copyright holders. The license that was approved for my changes to this driver originally came from LSI, but the changes to the driver core are not owned by LSI. MFC: 1 day
2005-08-03 14:08:41 +00:00
* 3. Neither the names of the above listed copyright holders nor the names
* of any contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF THE COPYRIGHT
* OWNER OR CONTRIBUTOR IS ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <dev/mpt/mpt.h>
#include <dev/mpt/mpt_cam.h>
#include <dev/mpt/mpt_raid.h>
#include "dev/mpt/mpilib/mpi_ioc.h" /* XXX Fix Event Handling!!! */
#include "dev/mpt/mpilib/mpi_init.h"
#include "dev/mpt/mpilib/mpi_targ.h"
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
#include "dev/mpt/mpilib/mpi_fc.h"
The poison pill of death: adding a target mode reply handler and target resources to a non-FC card killed us dead. Sorry for the breakage since last July 12.
2006-09-05 23:53:07 +00:00
#if __FreeBSD_version >= 500000
Add sysctl information about things like WWNN/WWPN. MFC after: 2 weeks
2006-07-16 06:05:44 +00:00
#include <sys/sysctl.h>
The poison pill of death: adding a target mode reply handler and target resources to a non-FC card killed us dead. Sorry for the breakage since last July 12.
2006-09-05 23:53:07 +00:00
#endif
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
#include <sys/callout.h>
#include <sys/kthread.h>
static void mpt_poll(struct cam_sim *);
static timeout_t mpt_timeout;
static void mpt_action(struct cam_sim *, union ccb *);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
static int
mpt_get_spi_settings(struct mpt_softc *, struct ccb_trans_settings *);
static void mpt_setwidth(struct mpt_softc *, int, int);
static void mpt_setsync(struct mpt_softc *, int, int, int);
static int mpt_update_spi_config(struct mpt_softc *, int);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static void mpt_calc_geometry(struct ccb_calc_geometry *ccg, int extended);
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static mpt_reply_handler_t mpt_scsi_reply_handler;
static mpt_reply_handler_t mpt_scsi_tmf_reply_handler;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static mpt_reply_handler_t mpt_fc_els_reply_handler;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
static int mpt_scsi_reply_frame_handler(struct mpt_softc *, request_t *,
MSG_DEFAULT_REPLY *);
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
static int mpt_bus_reset(struct mpt_softc *, target_id_t, lun_id_t, int);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static int mpt_fc_reset_link(struct mpt_softc *, int);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static int mpt_spawn_recovery_thread(struct mpt_softc *mpt);
static void mpt_terminate_recovery_thread(struct mpt_softc *mpt);
static void mpt_recovery_thread(void *arg);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
static void mpt_recover_commands(struct mpt_softc *mpt);
static int mpt_scsi_send_tmf(struct mpt_softc *, u_int, u_int, u_int,
u_int, u_int, u_int, int);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
static void mpt_fc_post_els(struct mpt_softc *mpt, request_t *, int);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static void mpt_post_target_command(struct mpt_softc *, request_t *, int);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
static int mpt_add_els_buffers(struct mpt_softc *mpt);
static int mpt_add_target_commands(struct mpt_softc *mpt);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static int mpt_enable_lun(struct mpt_softc *, target_id_t, lun_id_t);
static int mpt_disable_lun(struct mpt_softc *, target_id_t, lun_id_t);
static void mpt_target_start_io(struct mpt_softc *, union ccb *);
static cam_status mpt_abort_target_ccb(struct mpt_softc *, union ccb *);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
static int mpt_abort_target_cmd(struct mpt_softc *, request_t *);
static void mpt_scsi_tgt_status(struct mpt_softc *, union ccb *, request_t *,
uint8_t, uint8_t const *);
static void
mpt_scsi_tgt_tsk_mgmt(struct mpt_softc *, request_t *, mpt_task_mgmt_t,
tgt_resource_t *, int);
static void mpt_tgt_dump_tgt_state(struct mpt_softc *, request_t *);
static void mpt_tgt_dump_req_state(struct mpt_softc *, request_t *);
static mpt_reply_handler_t mpt_scsi_tgt_reply_handler;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static uint32_t scsi_io_handler_id = MPT_HANDLER_ID_NONE;
static uint32_t scsi_tmf_handler_id = MPT_HANDLER_ID_NONE;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static uint32_t fc_els_handler_id = MPT_HANDLER_ID_NONE;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static mpt_probe_handler_t mpt_cam_probe;
static mpt_attach_handler_t mpt_cam_attach;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static mpt_enable_handler_t mpt_cam_enable;
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
static mpt_ready_handler_t mpt_cam_ready;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static mpt_event_handler_t mpt_cam_event;
static mpt_reset_handler_t mpt_cam_ioc_reset;
static mpt_detach_handler_t mpt_cam_detach;
static struct mpt_personality mpt_cam_personality =
{
.name = "mpt_cam",
.probe = mpt_cam_probe,
.attach = mpt_cam_attach,
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
.enable = mpt_cam_enable,
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
.ready = mpt_cam_ready,
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
.event = mpt_cam_event,
.reset = mpt_cam_ioc_reset,
.detach = mpt_cam_detach,
};
DECLARE_MPT_PERSONALITY(mpt_cam, SI_ORDER_SECOND);
int
mpt_cam_probe(struct mpt_softc *mpt)
{
When probing to attach the CAM functionality, check against desired role configuration instead of existing role. This gets us out of the mess where we configured a role of NONE (or were LAN only, for example), but didn't continue to attach the CAM module (because we had neither initiator nor target role set). Unfortunately, the code that rewrites NVRAM to match actual to desired role only works if the CAM module attaches. MFC after: 2 weeks
2006-07-25 00:59:54 +00:00
int role;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
When probing to attach the CAM functionality, check against desired role configuration instead of existing role. This gets us out of the mess where we configured a role of NONE (or were LAN only, for example), but didn't continue to attach the CAM module (because we had neither initiator nor target role set). Unfortunately, the code that rewrites NVRAM to match actual to desired role only works if the CAM module attaches. MFC after: 2 weeks
2006-07-25 00:59:54 +00:00
* Only attach to nodes that support the initiator or target role
* (or want to) or have RAID physical devices that need CAM pass-thru
* support.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
When probing to attach the CAM functionality, check against desired role configuration instead of existing role. This gets us out of the mess where we configured a role of NONE (or were LAN only, for example), but didn't continue to attach the CAM module (because we had neither initiator nor target role set). Unfortunately, the code that rewrites NVRAM to match actual to desired role only works if the CAM module attaches. MFC after: 2 weeks
2006-07-25 00:59:54 +00:00
if (mpt->do_cfg_role) {
role = mpt->cfg_role;
} else {
role = mpt->role;
}
if ((role & (MPT_ROLE_TARGET|MPT_ROLE_INITIATOR)) != 0 ||
(mpt->ioc_page2 != NULL && mpt->ioc_page2->MaxPhysDisks != 0)) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (0);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (ENODEV);
}
int
mpt_cam_attach(struct mpt_softc *mpt)
{
struct cam_devq *devq;
mpt_handler_t handler;
int maxq;
int error;
TAILQ_INIT(&mpt->request_timeout_list);
maxq = (mpt->mpt_global_credits < MPT_MAX_REQUESTS(mpt))?
mpt->mpt_global_credits : MPT_MAX_REQUESTS(mpt);
handler.reply_handler = mpt_scsi_reply_handler;
error = mpt_register_handler(mpt, MPT_HANDLER_REPLY, handler,
&scsi_io_handler_id);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error != 0) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
handler.reply_handler = mpt_scsi_tmf_reply_handler;
error = mpt_register_handler(mpt, MPT_HANDLER_REPLY, handler,
&scsi_tmf_handler_id);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error != 0) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* If we're fibre channel and could support target mode, we register
* an ELS reply handler and give it resources.
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
if (mpt->is_fc && (mpt->role & MPT_ROLE_TARGET) != 0) {
handler.reply_handler = mpt_fc_els_reply_handler;
error = mpt_register_handler(mpt, MPT_HANDLER_REPLY, handler,
&fc_els_handler_id);
if (error != 0) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (mpt_add_els_buffers(mpt) == FALSE) {
error = ENOMEM;
goto cleanup0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
maxq -= mpt->els_cmds_allocated;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* If we support target mode, we register a reply handler for it,
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
* but don't add command resources until we actually enable target
* mode.
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
The poison pill of death: adding a target mode reply handler and target resources to a non-FC card killed us dead. Sorry for the breakage since last July 12.
2006-09-05 23:53:07 +00:00
if (mpt->is_fc && (mpt->role & MPT_ROLE_TARGET) != 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
handler.reply_handler = mpt_scsi_tgt_reply_handler;
error = mpt_register_handler(mpt, MPT_HANDLER_REPLY, handler,
&mpt->scsi_tgt_handler_id);
if (error != 0) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
* We keep one request reserved for timeout TMF requests.
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt->tmf_req = mpt_get_request(mpt, FALSE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
if (mpt->tmf_req == NULL) {
mpt_prt(mpt, "Unable to allocate dedicated TMF request!\n");
error = ENOMEM;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
/*
* Mark the request as free even though not on the free list.
* There is only one TMF request allowed to be outstanding at
* a time and the TMF routines perform their own allocation
* tracking using the standard state flags.
*/
mpt->tmf_req->state = REQ_STATE_FREE;
maxq--;
if (mpt_spawn_recovery_thread(mpt) != 0) {
mpt_prt(mpt, "Unable to spawn recovery thread!\n");
error = ENOMEM;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
goto cleanup0;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* The rest of this is CAM foo, for which we need to drop our lock
*/
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
* Create the device queue for our SIM(s).
*/
devq = cam_simq_alloc(maxq);
if (devq == NULL) {
mpt_prt(mpt, "Unable to allocate CAM SIMQ!\n");
error = ENOMEM;
goto cleanup;
}
/*
* Construct our SIM entry.
*/
mpt->sim = cam_sim_alloc(mpt_action, mpt_poll, "mpt", mpt,
mpt->unit, 1, maxq, devq);
if (mpt->sim == NULL) {
mpt_prt(mpt, "Unable to allocate CAM SIM!\n");
cam_simq_free(devq);
error = ENOMEM;
goto cleanup;
}
/*
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
* Register exactly this bus.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
if (xpt_bus_register(mpt->sim, 0) != CAM_SUCCESS) {
mpt_prt(mpt, "Bus registration Failed!\n");
error = ENOMEM;
goto cleanup;
}
if (xpt_create_path(&mpt->path, NULL, cam_sim_path(mpt->sim),
CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
mpt_prt(mpt, "Unable to allocate Path!\n");
error = ENOMEM;
goto cleanup;
}
/*
* Only register a second bus for RAID physical
* devices if the controller supports RAID.
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (mpt->ioc_page2 == NULL || mpt->ioc_page2->MaxPhysDisks == 0) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (0);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
* Create a "bus" to export all hidden disks to CAM.
*/
mpt->phydisk_sim = cam_sim_alloc(mpt_action, mpt_poll, "mpt", mpt,
mpt->unit, 1, maxq, devq);
if (mpt->phydisk_sim == NULL) {
mpt_prt(mpt, "Unable to allocate Physical Disk CAM SIM!\n");
error = ENOMEM;
goto cleanup;
}
/*
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
* Register this bus.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
if (xpt_bus_register(mpt->phydisk_sim, 1) != CAM_SUCCESS) {
mpt_prt(mpt, "Physical Disk Bus registration Failed!\n");
error = ENOMEM;
goto cleanup;
}
if (xpt_create_path(&mpt->phydisk_path, NULL,
cam_sim_path(mpt->phydisk_sim),
CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
mpt_prt(mpt, "Unable to allocate Physical Disk Path!\n");
error = ENOMEM;
goto cleanup;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
CAMLOCK_2_MPTLOCK(mpt);
When probing to attach the CAM functionality, check against desired role configuration instead of existing role. This gets us out of the mess where we configured a role of NONE (or were LAN only, for example), but didn't continue to attach the CAM module (because we had neither initiator nor target role set). Unfortunately, the code that rewrites NVRAM to match actual to desired role only works if the CAM module attaches. MFC after: 2 weeks
2006-07-25 00:59:54 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG, "attached cam\n");
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (0);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cleanup:
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
CAMLOCK_2_MPTLOCK(mpt);
cleanup0:
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_cam_detach(mpt);
return (error);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
* Read FC configuration information
*/
static int
mpt_read_config_info_fc(struct mpt_softc *mpt)
{
char *topology = NULL;
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
int rv;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_FC_PORT, 0,
0, &mpt->mpt_fcport_page0.Header, FALSE, 5000);
if (rv) {
return (-1);
}
mpt_lprt(mpt, MPT_PRT_DEBUG, "FC Port Page 0 Header: %x %x %x %x\n",
mpt->mpt_fcport_page0.Header.PageVersion,
mpt->mpt_fcport_page0.Header.PageLength,
mpt->mpt_fcport_page0.Header.PageNumber,
mpt->mpt_fcport_page0.Header.PageType);
rv = mpt_read_cur_cfg_page(mpt, 0, &mpt->mpt_fcport_page0.Header,
sizeof(mpt->mpt_fcport_page0), FALSE, 5000);
if (rv) {
mpt_prt(mpt, "failed to read FC Port Page 0\n");
return (-1);
}
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
mpt->mpt_fcport_speed = mpt->mpt_fcport_page0.CurrentSpeed;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
switch (mpt->mpt_fcport_page0.Flags &
MPI_FCPORTPAGE0_FLAGS_ATTACH_TYPE_MASK) {
case MPI_FCPORTPAGE0_FLAGS_ATTACH_NO_INIT:
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
mpt->mpt_fcport_speed = 0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
topology = "<NO LOOP>";
break;
case MPI_FCPORTPAGE0_FLAGS_ATTACH_POINT_TO_POINT:
topology = "N-Port";
break;
case MPI_FCPORTPAGE0_FLAGS_ATTACH_PRIVATE_LOOP:
topology = "NL-Port";
break;
case MPI_FCPORTPAGE0_FLAGS_ATTACH_FABRIC_DIRECT:
topology = "F-Port";
break;
case MPI_FCPORTPAGE0_FLAGS_ATTACH_PUBLIC_LOOP:
topology = "FL-Port";
break;
default:
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
mpt->mpt_fcport_speed = 0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
topology = "?";
break;
}
mpt_lprt(mpt, MPT_PRT_INFO,
"FC Port Page 0: Topology <%s> WWNN 0x%08x%08x WWPN 0x%08x%08x "
"Speed %u-Gbit\n", topology,
mpt->mpt_fcport_page0.WWNN.High,
mpt->mpt_fcport_page0.WWNN.Low,
mpt->mpt_fcport_page0.WWPN.High,
mpt->mpt_fcport_page0.WWPN.Low,
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
mpt->mpt_fcport_speed);
Add sysctl information about things like WWNN/WWPN. MFC after: 2 weeks
2006-07-16 06:05:44 +00:00
#if __FreeBSD_version >= 500000
{
struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(mpt->dev);
struct sysctl_oid *tree = device_get_sysctl_tree(mpt->dev);
snprintf(mpt->scinfo.fc.wwnn,
sizeof (mpt->scinfo.fc.wwnn), "0x%08x%08x",
mpt->mpt_fcport_page0.WWNN.High,
mpt->mpt_fcport_page0.WWNN.Low);
snprintf(mpt->scinfo.fc.wwpn,
sizeof (mpt->scinfo.fc.wwpn), "0x%08x%08x",
mpt->mpt_fcport_page0.WWPN.High,
mpt->mpt_fcport_page0.WWPN.Low);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Add sysctl information about things like WWNN/WWPN. MFC after: 2 weeks
2006-07-16 06:05:44 +00:00
SYSCTL_ADD_STRING(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
"wwnn", CTLFLAG_RD, mpt->scinfo.fc.wwnn, 0,
"World Wide Node Name");
SYSCTL_ADD_STRING(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
"wwpn", CTLFLAG_RD, mpt->scinfo.fc.wwpn, 0,
"World Wide Port Name");
}
#endif
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (0);
}
/*
* Set FC configuration information.
*/
static int
mpt_set_initial_config_fc(struct mpt_softc *mpt)
{
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
CONFIG_PAGE_FC_PORT_1 fc;
U32 fl;
int r, doit = 0;
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
int role;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
r = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_FC_PORT, 1, 0,
&fc.Header, FALSE, 5000);
if (r) {
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
mpt_prt(mpt, "failed to read FC page 1 header\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (mpt_fc_reset_link(mpt, 1));
}
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
r = mpt_read_cfg_page(mpt, MPI_CONFIG_ACTION_PAGE_READ_NVRAM, 0,
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
&fc.Header, sizeof (fc), FALSE, 5000);
if (r) {
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
mpt_prt(mpt, "failed to read FC page 1\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (mpt_fc_reset_link(mpt, 1));
}
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
/*
* Check our flags to make sure we support the role we want.
*/
doit = 0;
role = 0;
fl = le32toh(fc.Flags);;
if (fl & MPI_FCPORTPAGE1_FLAGS_PROT_FCP_INIT) {
role |= MPT_ROLE_INITIATOR;
}
if (fl & MPI_FCPORTPAGE1_FLAGS_PROT_FCP_TARG) {
role |= MPT_ROLE_TARGET;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
fl &= ~MPI_FCPORTPAGE1_FLAGS_PROT_MASK;
if (mpt->do_cfg_role == 0) {
role = mpt->cfg_role;
} else {
mpt->do_cfg_role = 0;
}
if (role != mpt->cfg_role) {
if (mpt->cfg_role & MPT_ROLE_INITIATOR) {
if ((role & MPT_ROLE_INITIATOR) == 0) {
mpt_prt(mpt, "adding initiator role\n");
fl |= MPI_FCPORTPAGE1_FLAGS_PROT_FCP_INIT;
doit++;
} else {
mpt_prt(mpt, "keeping initiator role\n");
}
} else if (role & MPT_ROLE_INITIATOR) {
mpt_prt(mpt, "removing initiator role\n");
doit++;
}
if (mpt->cfg_role & MPT_ROLE_TARGET) {
if ((role & MPT_ROLE_TARGET) == 0) {
mpt_prt(mpt, "adding target role\n");
fl |= MPI_FCPORTPAGE1_FLAGS_PROT_FCP_TARG;
doit++;
} else {
mpt_prt(mpt, "keeping target role\n");
}
} else if (role & MPT_ROLE_TARGET) {
mpt_prt(mpt, "removing target role\n");
doit++;
}
mpt->role = mpt->cfg_role;
}
if (fl & MPI_FCPORTPAGE1_FLAGS_PROT_FCP_TARG) {
if ((fl & MPI_FCPORTPAGE1_FLAGS_TARGET_MODE_OXID) == 0) {
mpt_prt(mpt, "adding OXID option\n");
fl |= MPI_FCPORTPAGE1_FLAGS_TARGET_MODE_OXID;
doit++;
}
}
if (doit) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
fc.Flags = htole32(fl);
r = mpt_write_cfg_page(mpt,
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
MPI_CONFIG_ACTION_PAGE_WRITE_NVRAM, 0, &fc.Header,
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
sizeof(fc), FALSE, 5000);
if (r != 0) {
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
mpt_prt(mpt, "failed to update NVRAM with changes\n");
return (0);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix config page writes to not strip out the attributes when you actually go write the config page. This fixes the long standing problem about updating NVRAM on Fibre Channel cards and seems so far to not break SPI config page writes. Put back role setting into mpt. That is, you can set a desired role for mpt as a hint. On the next reboot, it'll pick that up and redo the NVRAM settings appropriately and warn you that this won't take effect until the next reboot. This saves people the step of having to find a BIOS utilities disk to set target and/or initiator role for the MPT cards.
2006-07-12 07:48:50 +00:00
mpt_prt(mpt, "NOTE: NVRAM changes will not take "
"effect until next reboot or IOC reset\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
return (0);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
* Read SAS configuration information. Nothing to do yet.
*/
static int
mpt_read_config_info_sas(struct mpt_softc *mpt)
{
return (0);
}
/*
* Set SAS configuration information. Nothing to do yet.
*/
static int
mpt_set_initial_config_sas(struct mpt_softc *mpt)
{
return (0);
}
/*
* Read SCSI configuration information
*/
static int
mpt_read_config_info_spi(struct mpt_softc *mpt)
{
int rv, i;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_SCSI_PORT, 0, 0,
&mpt->mpt_port_page0.Header, FALSE, 5000);
if (rv) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
mpt_lprt(mpt, MPT_PRT_DEBUG, "SPI Port Page 0 Header: %x %x %x %x\n",
mpt->mpt_port_page0.Header.PageVersion,
mpt->mpt_port_page0.Header.PageLength,
mpt->mpt_port_page0.Header.PageNumber,
mpt->mpt_port_page0.Header.PageType);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_SCSI_PORT, 1, 0,
&mpt->mpt_port_page1.Header, FALSE, 5000);
if (rv) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
mpt_lprt(mpt, MPT_PRT_DEBUG, "SPI Port Page 1 Header: %x %x %x %x\n",
mpt->mpt_port_page1.Header.PageVersion,
mpt->mpt_port_page1.Header.PageLength,
mpt->mpt_port_page1.Header.PageNumber,
mpt->mpt_port_page1.Header.PageType);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_SCSI_PORT, 2, 0,
&mpt->mpt_port_page2.Header, FALSE, 5000);
if (rv) {
return (-1);
}
mpt_lprt(mpt, MPT_PRT_DEBUG, "SPI Port Page 2 Header: %x %x %x %x\n",
mpt->mpt_port_page2.Header.PageVersion,
mpt->mpt_port_page2.Header.PageLength,
mpt->mpt_port_page2.Header.PageNumber,
mpt->mpt_port_page2.Header.PageType);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
for (i = 0; i < 16; i++) {
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_SCSI_DEVICE,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
0, i, &mpt->mpt_dev_page0[i].Header, FALSE, 5000);
if (rv) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"SPI Target %d Device Page 0 Header: %x %x %x %x\n", i,
mpt->mpt_dev_page0[i].Header.PageVersion,
mpt->mpt_dev_page0[i].Header.PageLength,
mpt->mpt_dev_page0[i].Header.PageNumber,
mpt->mpt_dev_page0[i].Header.PageType);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
rv = mpt_read_cfg_header(mpt, MPI_CONFIG_PAGETYPE_SCSI_DEVICE,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
1, i, &mpt->mpt_dev_page1[i].Header, FALSE, 5000);
if (rv) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"SPI Target %d Device Page 1 Header: %x %x %x %x\n", i,
mpt->mpt_dev_page1[i].Header.PageVersion,
mpt->mpt_dev_page1[i].Header.PageLength,
mpt->mpt_dev_page1[i].Header.PageNumber,
mpt->mpt_dev_page1[i].Header.PageType);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
* At this point, we don't *have* to fail. As long as we have
* valid config header information, we can (barely) lurch
* along.
*/
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cur_cfg_page(mpt, 0, &mpt->mpt_port_page0.Header,
sizeof(mpt->mpt_port_page0), FALSE, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rv) {
mpt_prt(mpt, "failed to read SPI Port Page 0\n");
} else {
mpt_lprt(mpt, MPT_PRT_DEBUG,
"SPI Port Page 0: Capabilities %x PhysicalInterface %x\n",
mpt->mpt_port_page0.Capabilities,
mpt->mpt_port_page0.PhysicalInterface);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cur_cfg_page(mpt, 0, &mpt->mpt_port_page1.Header,
sizeof(mpt->mpt_port_page1), FALSE, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rv) {
mpt_prt(mpt, "failed to read SPI Port Page 1\n");
} else {
mpt_lprt(mpt, MPT_PRT_DEBUG,
"SPI Port Page 1: Configuration %x OnBusTimerValue %x\n",
mpt->mpt_port_page1.Configuration,
mpt->mpt_port_page1.OnBusTimerValue);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cur_cfg_page(mpt, 0, &mpt->mpt_port_page2.Header,
sizeof(mpt->mpt_port_page2), FALSE, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rv) {
mpt_prt(mpt, "failed to read SPI Port Page 2\n");
} else {
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
"Port Page 2: Flags %x Settings %x\n",
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt->mpt_port_page2.PortFlags,
mpt->mpt_port_page2.PortSettings);
for (i = 0; i < 16; i++) {
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
" Port Page 2 Tgt %d: timo %x SF %x Flags %x\n",
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
i, mpt->mpt_port_page2.DeviceSettings[i].Timeout,
mpt->mpt_port_page2.DeviceSettings[i].SyncFactor,
mpt->mpt_port_page2.DeviceSettings[i].DeviceFlags);
}
}
for (i = 0; i < 16; i++) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rv = mpt_read_cur_cfg_page(mpt, i,
&mpt->mpt_dev_page0[i].Header, sizeof(*mpt->mpt_dev_page0),
FALSE, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rv) {
mpt_prt(mpt,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"cannot read SPI Target %d Device Page 0\n", i);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
continue;
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
"target %d page 0: Negotiated Params %x Information %x\n",
i, mpt->mpt_dev_page0[i].NegotiatedParameters,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt->mpt_dev_page0[i].Information);
rv = mpt_read_cur_cfg_page(mpt, i,
&mpt->mpt_dev_page1[i].Header, sizeof(*mpt->mpt_dev_page1),
FALSE, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rv) {
mpt_prt(mpt,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"cannot read SPI Target %d Device Page 1\n", i);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
continue;
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
"target %d page 1: Requested Params %x Configuration %x\n",
i, mpt->mpt_dev_page1[i].RequestedParameters,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt->mpt_dev_page1[i].Configuration);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
return (0);
}
/*
* Validate SPI configuration information.
*
* In particular, validate SPI Port Page 1.
*/
static int
mpt_set_initial_config_spi(struct mpt_softc *mpt)
{
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
int i, j, pp1val = ((1 << mpt->mpt_ini_id) << 16) | mpt->mpt_ini_id;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int error;
mpt->mpt_disc_enable = 0xff;
mpt->mpt_tag_enable = 0;
if (mpt->mpt_port_page1.Configuration != pp1val) {
CONFIG_PAGE_SCSI_PORT_1 tmp;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt, "SPI Port Page 1 Config value bad (%x)- should "
"be %x\n", mpt->mpt_port_page1.Configuration, pp1val);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
tmp = mpt->mpt_port_page1;
tmp.Configuration = pp1val;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
error = mpt_write_cur_cfg_page(mpt, 0,
&tmp.Header, sizeof(tmp), FALSE, 5000);
if (error) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
error = mpt_read_cur_cfg_page(mpt, 0,
&tmp.Header, sizeof(tmp), FALSE, 5000);
if (error) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (-1);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (tmp.Configuration != pp1val) {
mpt_prt(mpt,
"failed to reset SPI Port Page 1 Config value\n");
return (-1);
}
mpt->mpt_port_page1 = tmp;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* The purpose of this exercise is to get
* all targets back to async/narrow.
*
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
* We skip this step if the BIOS has already negotiated
* speeds with the targets and does not require us to
* do Domain Validation.
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
*/
i = mpt->mpt_port_page2.PortSettings &
MPI_SCSIPORTPAGE2_PORT_MASK_NEGO_MASTER_SETTINGS;
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
j = mpt->mpt_port_page2.PortFlags &
MPI_SCSIPORTPAGE2_PORT_FLAGS_DV_MASK;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (i == MPI_SCSIPORTPAGE2_PORT_ALL_MASTER_SETTINGS /* &&
j == MPI_SCSIPORTPAGE2_PORT_FLAGS_OFF_DV */) {
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
"honoring BIOS transfer negotiations\n");
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
} else {
for (i = 0; i < 16; i++) {
mpt->mpt_dev_page1[i].RequestedParameters = 0;
mpt->mpt_dev_page1[i].Configuration = 0;
(void) mpt_update_spi_config(mpt, i);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
return (0);
}
int
mpt_cam_enable(struct mpt_softc *mpt)
{
if (mpt->is_fc) {
if (mpt_read_config_info_fc(mpt)) {
return (EIO);
}
if (mpt_set_initial_config_fc(mpt)) {
return (EIO);
}
} else if (mpt->is_sas) {
if (mpt_read_config_info_sas(mpt)) {
return (EIO);
}
if (mpt_set_initial_config_sas(mpt)) {
return (EIO);
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
} else if (mpt->is_spi) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (mpt_read_config_info_spi(mpt)) {
return (EIO);
}
if (mpt_set_initial_config_spi(mpt)) {
return (EIO);
}
}
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
return (0);
}
void
mpt_cam_ready(struct mpt_softc *mpt)
{
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
/*
* If we're in target mode, hang out resources now
* so we don't cause the world to hang talking to us.
*/
The poison pill of death: adding a target mode reply handler and target resources to a non-FC card killed us dead. Sorry for the breakage since last July 12.
2006-09-05 23:53:07 +00:00
if (mpt->is_fc && (mpt->role & MPT_ROLE_TARGET)) {
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
/*
* Try to add some target command resources
*/
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
MPT_LOCK(mpt);
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
if (mpt_add_target_commands(mpt) == FALSE) {
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
mpt_prt(mpt, "failed to add target commands\n");
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
}
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
MPT_UNLOCK(mpt);
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
void
mpt_cam_detach(struct mpt_softc *mpt)
{
mpt_handler_t handler;
mpt_terminate_recovery_thread(mpt);
handler.reply_handler = mpt_scsi_reply_handler;
mpt_deregister_handler(mpt, MPT_HANDLER_REPLY, handler,
scsi_io_handler_id);
handler.reply_handler = mpt_scsi_tmf_reply_handler;
mpt_deregister_handler(mpt, MPT_HANDLER_REPLY, handler,
scsi_tmf_handler_id);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
handler.reply_handler = mpt_fc_els_reply_handler;
mpt_deregister_handler(mpt, MPT_HANDLER_REPLY, handler,
fc_els_handler_id);
handler.reply_handler = mpt_scsi_tgt_reply_handler;
mpt_deregister_handler(mpt, MPT_HANDLER_REPLY, handler,
mpt->scsi_tgt_handler_id);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
if (mpt->tmf_req != NULL) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt->tmf_req->state = REQ_STATE_ALLOCATED;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_free_request(mpt, mpt->tmf_req);
mpt->tmf_req = NULL;
}
if (mpt->sim != NULL) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
xpt_free_path(mpt->path);
xpt_bus_deregister(cam_sim_path(mpt->sim));
cam_sim_free(mpt->sim, TRUE);
mpt->sim = NULL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
if (mpt->phydisk_sim != NULL) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
xpt_free_path(mpt->phydisk_path);
xpt_bus_deregister(cam_sim_path(mpt->phydisk_sim));
cam_sim_free(mpt->phydisk_sim, TRUE);
mpt->phydisk_sim = NULL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/* This routine is used after a system crash to dump core onto the swap device.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
static void
mpt_poll(struct cam_sim *sim)
{
struct mpt_softc *mpt;
mpt = (struct mpt_softc *)cam_sim_softc(sim);
MPT_LOCK(mpt);
mpt_intr(mpt);
MPT_UNLOCK(mpt);
}
/*
* Watchdog timeout routine for SCSI requests.
*/
static void
mpt_timeout(void *arg)
{
union ccb *ccb;
struct mpt_softc *mpt;
request_t *req;
ccb = (union ccb *)arg;
mpt = ccb->ccb_h.ccb_mpt_ptr;
MPT_LOCK(mpt);
req = ccb->ccb_h.ccb_req_ptr;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_prt(mpt, "request %p:%u timed out for ccb %p (req->ccb %p)\n", req,
req->serno, ccb, req->ccb);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* XXX: WHAT ARE WE TRYING TO DO HERE? */
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
if ((req->state & REQ_STATE_QUEUED) == REQ_STATE_QUEUED) {
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
TAILQ_INSERT_TAIL(&mpt->request_timeout_list, req, links);
req->state |= REQ_STATE_TIMEDOUT;
mpt_wakeup_recovery_thread(mpt);
}
MPT_UNLOCK(mpt);
}
/*
* Callback routine from "bus_dmamap_load" or, in simple cases, called directly.
*
* Takes a list of physical segments and builds the SGL for SCSI IO command
* and forwards the commard to the IOC after one last check that CAM has not
* aborted the transaction.
*/
static void
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_execute_req_a64(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
request_t *req, *trq;
char *mpt_off;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
union ccb *ccb;
struct mpt_softc *mpt;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
int seg, first_lim;
uint32_t flags, nxt_off;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
void *sglp = NULL;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MSG_REQUEST_HEADER *hdrp;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
SGE_SIMPLE64 *se;
SGE_CHAIN64 *ce;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
int istgt = 0;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
req = (request_t *)arg;
ccb = req->ccb;
mpt = ccb->ccb_h.ccb_mpt_ptr;
req = ccb->ccb_h.ccb_req_ptr;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
hdrp = req->req_vbuf;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
mpt_off = req->req_vbuf;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (error == 0 && ((uint32_t)nseg) >= mpt->max_seg_cnt) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
error = EFBIG;
}
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (error == 0) {
switch (hdrp->Function) {
case MPI_FUNCTION_SCSI_IO_REQUEST:
case MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH:
istgt = 0;
sglp = &((PTR_MSG_SCSI_IO_REQUEST)hdrp)->SGL;
break;
case MPI_FUNCTION_TARGET_ASSIST:
istgt = 1;
sglp = &((PTR_MSG_TARGET_ASSIST_REQUEST)hdrp)->SGL;
break;
default:
mpt_prt(mpt, "bad fct 0x%x in mpt_execute_req_a64\n",
hdrp->Function);
error = EINVAL;
break;
}
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (error == 0 && ((uint32_t)nseg) >= mpt->max_seg_cnt) {
error = EFBIG;
mpt_prt(mpt, "segment count %d too large (max %u)\n",
nseg, mpt->max_seg_cnt);
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
bad:
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
if (error != 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error != EFBIG && error != ENOMEM) {
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_prt(mpt, "mpt_execute_req_a64: err %d\n", error);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_INPROG) {
cam_status status;
mpt_freeze_ccb(ccb);
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (error == EFBIG) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
status = CAM_REQ_TOO_BIG;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
} else if (error == ENOMEM) {
if (mpt->outofbeer == 0) {
mpt->outofbeer = 1;
xpt_freeze_simq(mpt->sim, 1);
mpt_lprt(mpt, MPT_PRT_DEBUG,
"FREEZEQ\n");
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
status = CAM_REQUEUE_REQ;
} else {
status = CAM_REQ_CMP_ERR;
}
mpt_set_ccb_status(ccb, status);
}
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req =
MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
MPT_TGT_STATE(mpt, cmd_req)->state = TGT_STATE_IN_CAM;
MPT_TGT_STATE(mpt, cmd_req)->ccb = NULL;
MPT_TGT_STATE(mpt, cmd_req)->req = NULL;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
mpt_free_request(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
return;
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* No data to transfer?
* Just make a single simple SGL with zero length.
*/
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (mpt->verbose >= MPT_PRT_DEBUG) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int tidx = ((char *)sglp) - mpt_off;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
memset(&mpt_off[tidx], 0xff, MPT_REQUEST_AREA - tidx);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (nseg == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
SGE_SIMPLE32 *se1 = (SGE_SIMPLE32 *) sglp;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
MPI_pSGE_SET_FLAGS(se1,
(MPI_SGE_FLAGS_LAST_ELEMENT | MPI_SGE_FLAGS_END_OF_BUFFER |
MPI_SGE_FLAGS_SIMPLE_ELEMENT | MPI_SGE_FLAGS_END_OF_LIST));
goto out;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
flags = MPI_SGE_FLAGS_SIMPLE_ELEMENT | MPI_SGE_FLAGS_64_BIT_ADDRESSING;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (istgt == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
flags |= MPI_SGE_FLAGS_HOST_TO_IOC;
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
flags |= MPI_SGE_FLAGS_HOST_TO_IOC;
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (!(ccb->ccb_h.flags & (CAM_SG_LIST_PHYS|CAM_DATA_PHYS))) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
bus_dmasync_op_t op;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (istgt == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
op = BUS_DMASYNC_PREREAD;
} else {
op = BUS_DMASYNC_PREWRITE;
}
} else {
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
op = BUS_DMASYNC_PREWRITE;
} else {
op = BUS_DMASYNC_PREREAD;
}
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
bus_dmamap_sync(mpt->buffer_dmat, req->dmap, op);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* Okay, fill in what we can at the end of the command frame.
* If we have up to MPT_NSGL_FIRST, we can fit them all into
* the command frame.
*
* Otherwise, we fill up through MPT_NSGL_FIRST less one
* SIMPLE64 pointers and start doing CHAIN64 entries after
* that.
*/
if (nseg < MPT_NSGL_FIRST(mpt)) {
first_lim = nseg;
} else {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
* Leave room for CHAIN element
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
first_lim = MPT_NSGL_FIRST(mpt) - 1;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
se = (SGE_SIMPLE64 *) sglp;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
for (seg = 0; seg < first_lim; seg++, se++, dm_segs++) {
uint32_t tf;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(se, 0, sizeof (*se));
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
se->Address.Low = dm_segs->ds_addr;
if (sizeof(bus_addr_t) > 4) {
se->Address.High = ((uint64_t) dm_segs->ds_addr) >> 32;
}
MPI_pSGE_SET_LENGTH(se, dm_segs->ds_len);
tf = flags;
if (seg == first_lim - 1) {
tf |= MPI_SGE_FLAGS_LAST_ELEMENT;
}
if (seg == nseg - 1) {
tf |= MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
}
MPI_pSGE_SET_FLAGS(se, tf);
}
if (seg == nseg) {
goto out;
}
/*
* Tell the IOC where to find the first chain element.
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
hdrp->ChainOffset = ((char *)se - (char *)hdrp) >> 2;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
nxt_off = MPT_RQSL(mpt);
trq = req;
/*
* Make up the rest of the data segments out of a chain element
* (contiained in the current request frame) which points to
* SIMPLE64 elements in the next request frame, possibly ending
* with *another* chain element (if there's more).
*/
while (seg < nseg) {
int this_seg_lim;
uint32_t tf, cur_off;
bus_addr_t chain_list_addr;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
* Point to the chain descriptor. Note that the chain
* descriptor is at the end of the *previous* list (whether
* chain or simple).
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
ce = (SGE_CHAIN64 *) se;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* Before we change our current pointer, make sure we won't
* overflow the request area with this frame. Note that we
* test against 'greater than' here as it's okay in this case
* to have next offset be just outside the request area.
*/
if ((nxt_off + MPT_RQSL(mpt)) > MPT_REQUEST_AREA) {
nxt_off = MPT_REQUEST_AREA;
goto next_chain;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* Set our SGE element pointer to the beginning of the chain
* list and update our next chain list offset.
*/
se = (SGE_SIMPLE64 *) &mpt_off[nxt_off];
cur_off = nxt_off;
nxt_off += MPT_RQSL(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* Now initialized the chain descriptor.
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(ce, 0, sizeof (*ce));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* Get the physical address of the chain list.
*/
chain_list_addr = trq->req_pbuf;
chain_list_addr += cur_off;
if (sizeof (bus_addr_t) > 4) {
ce->Address.High =
(uint32_t) ((uint64_t)chain_list_addr >> 32);
}
ce->Address.Low = (uint32_t) chain_list_addr;
ce->Flags = MPI_SGE_FLAGS_CHAIN_ELEMENT |
MPI_SGE_FLAGS_64_BIT_ADDRESSING;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
/*
* If we have more than a frame's worth of segments left,
* set up the chain list to have the last element be another
* chain descriptor.
*/
if ((nseg - seg) > MPT_NSGL(mpt)) {
this_seg_lim = seg + MPT_NSGL(mpt) - 1;
/*
* The length of the chain is the length in bytes of the
* number of segments plus the next chain element.
*
* The next chain descriptor offset is the length,
* in words, of the number of segments.
*/
ce->Length = (this_seg_lim - seg) *
sizeof (SGE_SIMPLE64);
ce->NextChainOffset = ce->Length >> 2;
ce->Length += sizeof (SGE_CHAIN64);
} else {
this_seg_lim = nseg;
ce->Length = (this_seg_lim - seg) *
sizeof (SGE_SIMPLE64);
}
/*
* Fill in the chain list SGE elements with our segment data.
*
* If we're the last element in this chain list, set the last
* element flag. If we're the completely last element period,
* set the end of list and end of buffer flags.
*/
while (seg < this_seg_lim) {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(se, 0, sizeof (*se));
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
se->Address.Low = dm_segs->ds_addr;
if (sizeof (bus_addr_t) > 4) {
se->Address.High =
((uint64_t)dm_segs->ds_addr) >> 32;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
MPI_pSGE_SET_LENGTH(se, dm_segs->ds_len);
tf = flags;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (seg == this_seg_lim - 1) {
tf |= MPI_SGE_FLAGS_LAST_ELEMENT;
}
if (seg == nseg - 1) {
tf |= MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
MPI_pSGE_SET_FLAGS(se, tf);
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
se++;
seg++;
dm_segs++;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
next_chain:
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
* If we have more segments to do and we've used up all of
* the space in a request area, go allocate another one
* and chain to that.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (seg < nseg && nxt_off >= MPT_REQUEST_AREA) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
request_t *nrq;
CAMLOCK_2_MPTLOCK(mpt);
nrq = mpt_get_request(mpt, FALSE);
MPTLOCK_2_CAMLOCK(mpt);
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
if (nrq == NULL) {
error = ENOMEM;
goto bad;
}
/*
* Append the new request area on the tail of our list.
*/
if ((trq = req->chain) == NULL) {
req->chain = nrq;
} else {
while (trq->chain != NULL) {
trq = trq->chain;
}
trq->chain = nrq;
}
trq = nrq;
mpt_off = trq->req_vbuf;
if (mpt->verbose >= MPT_PRT_DEBUG) {
memset(mpt_off, 0xff, MPT_REQUEST_AREA);
}
nxt_off = 0;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
out:
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
* Last time we need to check if this CCB needs to be aborted.
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_INPROG) {
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req =
MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
MPT_TGT_STATE(mpt, cmd_req)->state = TGT_STATE_IN_CAM;
MPT_TGT_STATE(mpt, cmd_req)->ccb = NULL;
MPT_TGT_STATE(mpt, cmd_req)->req = NULL;
}
mpt_prt(mpt,
"mpt_execute_req_a64: I/O cancelled (status 0x%x)\n",
ccb->ccb_h.status & CAM_STATUS_MASK);
if (nseg && (ccb->ccb_h.flags & CAM_SG_LIST_PHYS) == 0) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
bus_dmamap_unload(mpt->buffer_dmat, req->dmap);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
xpt_done(ccb);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
CAMLOCK_2_MPTLOCK(mpt);
mpt_free_request(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
return;
}
ccb->ccb_h.status |= CAM_SIM_QUEUED;
if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
ccb->ccb_h.timeout_ch =
timeout(mpt_timeout, (caddr_t)ccb,
(ccb->ccb_h.timeout * hz) / 1000);
} else {
callout_handle_init(&ccb->ccb_h.timeout_ch);
}
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
if (mpt->verbose > MPT_PRT_DEBUG) {
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
int nc = 0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_print_request(req->req_vbuf);
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
for (trq = req->chain; trq; trq = trq->chain) {
printf(" Additional Chain Area %d\n", nc++);
mpt_dump_sgl(trq->req_vbuf, 0);
}
}
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req = MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, cmd_req);
#ifdef WE_TRUST_AUTO_GOOD_STATUS
if ((ccb->ccb_h.flags & CAM_SEND_STATUS) &&
csio->scsi_status == SCSI_STATUS_OK && tgt->resid == 0) {
tgt->state = TGT_STATE_MOVING_DATA_AND_STATUS;
} else {
tgt->state = TGT_STATE_MOVING_DATA;
}
#else
tgt->state = TGT_STATE_MOVING_DATA;
#endif
}
CAMLOCK_2_MPTLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_send_cmd(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
}
static void
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_execute_req(void *arg, bus_dma_segment_t *dm_segs, int nseg, int error)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
request_t *req, *trq;
char *mpt_off;
union ccb *ccb;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
struct mpt_softc *mpt;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int seg, first_lim;
uint32_t flags, nxt_off;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
void *sglp = NULL;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MSG_REQUEST_HEADER *hdrp;
SGE_SIMPLE32 *se;
SGE_CHAIN32 *ce;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
int istgt = 0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
req = (request_t *)arg;
ccb = req->ccb;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt = ccb->ccb_h.ccb_mpt_ptr;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
req = ccb->ccb_h.ccb_req_ptr;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
hdrp = req->req_vbuf;
mpt_off = req->req_vbuf;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (error == 0 && ((uint32_t)nseg) >= mpt->max_seg_cnt) {
error = EFBIG;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (error == 0) {
switch (hdrp->Function) {
case MPI_FUNCTION_SCSI_IO_REQUEST:
case MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH:
sglp = &((PTR_MSG_SCSI_IO_REQUEST)hdrp)->SGL;
break;
case MPI_FUNCTION_TARGET_ASSIST:
istgt = 1;
sglp = &((PTR_MSG_TARGET_ASSIST_REQUEST)hdrp)->SGL;
break;
default:
mpt_prt(mpt, "bad fct 0x%x in mpt_execute_req\n",
hdrp->Function);
error = EINVAL;
break;
}
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error == 0 && ((uint32_t)nseg) >= mpt->max_seg_cnt) {
error = EFBIG;
mpt_prt(mpt, "segment count %d too large (max %u)\n",
nseg, mpt->max_seg_cnt);
}
bad:
if (error != 0) {
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (error != EFBIG && error != ENOMEM) {
mpt_prt(mpt, "mpt_execute_req: err %d\n", error);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_INPROG) {
cam_status status;
mpt_freeze_ccb(ccb);
if (error == EFBIG) {
status = CAM_REQ_TOO_BIG;
} else if (error == ENOMEM) {
if (mpt->outofbeer == 0) {
mpt->outofbeer = 1;
xpt_freeze_simq(mpt->sim, 1);
mpt_lprt(mpt, MPT_PRT_DEBUG,
"FREEZEQ\n");
}
status = CAM_REQUEUE_REQ;
} else {
status = CAM_REQ_CMP_ERR;
}
mpt_set_ccb_status(ccb, status);
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req =
MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
MPT_TGT_STATE(mpt, cmd_req)->state = TGT_STATE_IN_CAM;
MPT_TGT_STATE(mpt, cmd_req)->ccb = NULL;
MPT_TGT_STATE(mpt, cmd_req)->req = NULL;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
mpt_free_request(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
return;
}
/*
* No data to transfer?
* Just make a single simple SGL with zero length.
*/
if (mpt->verbose >= MPT_PRT_DEBUG) {
int tidx = ((char *)sglp) - mpt_off;
memset(&mpt_off[tidx], 0xff, MPT_REQUEST_AREA - tidx);
}
if (nseg == 0) {
SGE_SIMPLE32 *se1 = (SGE_SIMPLE32 *) sglp;
MPI_pSGE_SET_FLAGS(se1,
(MPI_SGE_FLAGS_LAST_ELEMENT | MPI_SGE_FLAGS_END_OF_BUFFER |
MPI_SGE_FLAGS_SIMPLE_ELEMENT | MPI_SGE_FLAGS_END_OF_LIST));
goto out;
}
flags = MPI_SGE_FLAGS_SIMPLE_ELEMENT;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (istgt == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
flags |= MPI_SGE_FLAGS_HOST_TO_IOC;
}
} else {
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
flags |= MPI_SGE_FLAGS_HOST_TO_IOC;
}
}
if (!(ccb->ccb_h.flags & (CAM_SG_LIST_PHYS|CAM_DATA_PHYS))) {
bus_dmasync_op_t op;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (istgt) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
op = BUS_DMASYNC_PREREAD;
} else {
op = BUS_DMASYNC_PREWRITE;
}
} else {
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
op = BUS_DMASYNC_PREWRITE;
} else {
op = BUS_DMASYNC_PREREAD;
}
}
bus_dmamap_sync(mpt->buffer_dmat, req->dmap, op);
}
/*
* Okay, fill in what we can at the end of the command frame.
* If we have up to MPT_NSGL_FIRST, we can fit them all into
* the command frame.
*
* Otherwise, we fill up through MPT_NSGL_FIRST less one
* SIMPLE32 pointers and start doing CHAIN32 entries after
* that.
*/
if (nseg < MPT_NSGL_FIRST(mpt)) {
first_lim = nseg;
} else {
/*
* Leave room for CHAIN element
*/
first_lim = MPT_NSGL_FIRST(mpt) - 1;
}
se = (SGE_SIMPLE32 *) sglp;
for (seg = 0; seg < first_lim; seg++, se++, dm_segs++) {
uint32_t tf;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(se, 0,sizeof (*se));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
se->Address = dm_segs->ds_addr;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MPI_pSGE_SET_LENGTH(se, dm_segs->ds_len);
tf = flags;
if (seg == first_lim - 1) {
tf |= MPI_SGE_FLAGS_LAST_ELEMENT;
}
if (seg == nseg - 1) {
tf |= MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
}
MPI_pSGE_SET_FLAGS(se, tf);
}
if (seg == nseg) {
goto out;
}
/*
* Tell the IOC where to find the first chain element.
*/
hdrp->ChainOffset = ((char *)se - (char *)hdrp) >> 2;
nxt_off = MPT_RQSL(mpt);
trq = req;
/*
* Make up the rest of the data segments out of a chain element
* (contiained in the current request frame) which points to
* SIMPLE32 elements in the next request frame, possibly ending
* with *another* chain element (if there's more).
*/
while (seg < nseg) {
int this_seg_lim;
uint32_t tf, cur_off;
bus_addr_t chain_list_addr;
/*
* Point to the chain descriptor. Note that the chain
* descriptor is at the end of the *previous* list (whether
* chain or simple).
*/
ce = (SGE_CHAIN32 *) se;
/*
* Before we change our current pointer, make sure we won't
* overflow the request area with this frame. Note that we
* test against 'greater than' here as it's okay in this case
* to have next offset be just outside the request area.
*/
if ((nxt_off + MPT_RQSL(mpt)) > MPT_REQUEST_AREA) {
nxt_off = MPT_REQUEST_AREA;
goto next_chain;
}
/*
* Set our SGE element pointer to the beginning of the chain
* list and update our next chain list offset.
*/
se = (SGE_SIMPLE32 *) &mpt_off[nxt_off];
cur_off = nxt_off;
nxt_off += MPT_RQSL(mpt);
/*
* Now initialized the chain descriptor.
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(ce, 0, sizeof (*ce));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
* Get the physical address of the chain list.
*/
chain_list_addr = trq->req_pbuf;
chain_list_addr += cur_off;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ce->Address = chain_list_addr;
ce->Flags = MPI_SGE_FLAGS_CHAIN_ELEMENT;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
* If we have more than a frame's worth of segments left,
* set up the chain list to have the last element be another
* chain descriptor.
*/
if ((nseg - seg) > MPT_NSGL(mpt)) {
this_seg_lim = seg + MPT_NSGL(mpt) - 1;
/*
* The length of the chain is the length in bytes of the
* number of segments plus the next chain element.
*
* The next chain descriptor offset is the length,
* in words, of the number of segments.
*/
ce->Length = (this_seg_lim - seg) *
sizeof (SGE_SIMPLE32);
ce->NextChainOffset = ce->Length >> 2;
ce->Length += sizeof (SGE_CHAIN32);
} else {
this_seg_lim = nseg;
ce->Length = (this_seg_lim - seg) *
sizeof (SGE_SIMPLE32);
}
/*
* Fill in the chain list SGE elements with our segment data.
*
* If we're the last element in this chain list, set the last
* element flag. If we're the completely last element period,
* set the end of list and end of buffer flags.
*/
while (seg < this_seg_lim) {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(se, 0, sizeof (*se));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
se->Address = dm_segs->ds_addr;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MPI_pSGE_SET_LENGTH(se, dm_segs->ds_len);
tf = flags;
if (seg == this_seg_lim - 1) {
tf |= MPI_SGE_FLAGS_LAST_ELEMENT;
}
if (seg == nseg - 1) {
tf |= MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
}
MPI_pSGE_SET_FLAGS(se, tf);
se++;
seg++;
dm_segs++;
}
next_chain:
/*
* If we have more segments to do and we've used up all of
* the space in a request area, go allocate another one
* and chain to that.
*/
if (seg < nseg && nxt_off >= MPT_REQUEST_AREA) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
request_t *nrq;
CAMLOCK_2_MPTLOCK(mpt);
nrq = mpt_get_request(mpt, FALSE);
MPTLOCK_2_CAMLOCK(mpt);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (nrq == NULL) {
error = ENOMEM;
goto bad;
}
/*
* Append the new request area on the tail of our list.
*/
if ((trq = req->chain) == NULL) {
req->chain = nrq;
} else {
while (trq->chain != NULL) {
trq = trq->chain;
}
trq->chain = nrq;
}
trq = nrq;
mpt_off = trq->req_vbuf;
if (mpt->verbose >= MPT_PRT_DEBUG) {
memset(mpt_off, 0xff, MPT_REQUEST_AREA);
}
nxt_off = 0;
}
}
out:
/*
* Last time we need to check if this CCB needs to be aborted.
*/
if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_INPROG) {
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req =
MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
MPT_TGT_STATE(mpt, cmd_req)->state = TGT_STATE_IN_CAM;
MPT_TGT_STATE(mpt, cmd_req)->ccb = NULL;
MPT_TGT_STATE(mpt, cmd_req)->req = NULL;
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_prt(mpt,
"mpt_execute_req: I/O cancelled (status 0x%x)\n",
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ccb->ccb_h.status & CAM_STATUS_MASK);
if (nseg && (ccb->ccb_h.flags & CAM_SG_LIST_PHYS) == 0) {
bus_dmamap_unload(mpt->buffer_dmat, req->dmap);
}
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
mpt_free_request(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
return;
}
ccb->ccb_h.status |= CAM_SIM_QUEUED;
if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
ccb->ccb_h.timeout_ch =
timeout(mpt_timeout, (caddr_t)ccb,
(ccb->ccb_h.timeout * hz) / 1000);
} else {
callout_handle_init(&ccb->ccb_h.timeout_ch);
}
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
if (mpt->verbose > MPT_PRT_DEBUG) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int nc = 0;
mpt_print_request(req->req_vbuf);
for (trq = req->chain; trq; trq = trq->chain) {
printf(" Additional Chain Area %d\n", nc++);
mpt_dump_sgl(trq->req_vbuf, 0);
}
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (hdrp->Function == MPI_FUNCTION_TARGET_ASSIST) {
request_t *cmd_req = MPT_TAG_2_REQ(mpt, ccb->csio.tag_id);
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, cmd_req);
#ifdef WE_TRUST_AUTO_GOOD_STATUS
if ((ccb->ccb_h.flags & CAM_SEND_STATUS) &&
csio->scsi_status == SCSI_STATUS_OK && tgt->resid == 0) {
tgt->state = TGT_STATE_MOVING_DATA_AND_STATUS;
} else {
tgt->state = TGT_STATE_MOVING_DATA;
}
#else
tgt->state = TGT_STATE_MOVING_DATA;
#endif
}
CAMLOCK_2_MPTLOCK(mpt);
mpt_send_cmd(mpt, req);
MPTLOCK_2_CAMLOCK(mpt);
}
static void
mpt_start(struct cam_sim *sim, union ccb *ccb)
{
request_t *req;
struct mpt_softc *mpt;
MSG_SCSI_IO_REQUEST *mpt_req;
struct ccb_scsiio *csio = &ccb->csio;
struct ccb_hdr *ccbh = &ccb->ccb_h;
bus_dmamap_callback_t *cb;
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
target_id_t tgt;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int raid_passthru;
/* Get the pointer for the physical addapter */
mpt = ccb->ccb_h.ccb_mpt_ptr;
raid_passthru = (sim == mpt->phydisk_sim);
CAMLOCK_2_MPTLOCK(mpt);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((req = mpt_get_request(mpt, FALSE)) == NULL) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (mpt->outofbeer == 0) {
mpt->outofbeer = 1;
xpt_freeze_simq(mpt->sim, 1);
mpt_lprt(mpt, MPT_PRT_DEBUG, "FREEZEQ\n");
}
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_REQUEUE_REQ);
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
return;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_not_spcl(mpt, req, "mpt_start", __LINE__);
#endif
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MPTLOCK_2_CAMLOCK(mpt);
if (sizeof (bus_addr_t) > 4) {
cb = mpt_execute_req_a64;
} else {
cb = mpt_execute_req;
}
/*
* Link the ccb and the request structure so we can find
* the other knowing either the request or the ccb
*/
req->ccb = ccb;
ccb->ccb_h.ccb_req_ptr = req;
/* Now we build the command for the IOC */
mpt_req = req->req_vbuf;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(mpt_req, 0, sizeof (MSG_SCSI_IO_REQUEST));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->Function = MPI_FUNCTION_SCSI_IO_REQUEST;
if (raid_passthru) {
mpt_req->Function = MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH;
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
CAMLOCK_2_MPTLOCK(mpt);
if (mpt_map_physdisk(mpt, ccb, &tgt) != 0) {
MPTLOCK_2_CAMLOCK(mpt);
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_DEV_NOT_THERE);
xpt_done(ccb);
return;
}
MPTLOCK_2_CAMLOCK(mpt);
mpt_req->Bus = 0; /* we never set bus here */
} else {
tgt = ccb->ccb_h.target_id;
mpt_req->Bus = 0; /* XXX */
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
mpt_req->SenseBufferLength =
(csio->sense_len < MPT_SENSE_SIZE) ?
csio->sense_len : MPT_SENSE_SIZE;
/*
* We use the message context to find the request structure when we
* Get the command completion interrupt from the IOC.
*/
mpt_req->MsgContext = htole32(req->index | scsi_io_handler_id);
/* Which physical device to do the I/O on */
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
mpt_req->TargetID = tgt;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/* We assume a single level LUN type */
if (ccb->ccb_h.target_lun >= 256) {
mpt_req->LUN[0] = 0x40 | ((ccb->ccb_h.target_lun >> 8) & 0x3f);
mpt_req->LUN[1] = ccb->ccb_h.target_lun & 0xff;
} else {
mpt_req->LUN[1] = ccb->ccb_h.target_lun;
}
/* Set the direction of the transfer */
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->Control = MPI_SCSIIO_CONTROL_READ;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_OUT) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->Control = MPI_SCSIIO_CONTROL_WRITE;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->Control = MPI_SCSIIO_CONTROL_NODATATRANSFER;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_TAG_ACTION_VALID) != 0) {
switch(ccb->csio.tag_action) {
case MSG_HEAD_OF_Q_TAG:
mpt_req->Control |= MPI_SCSIIO_CONTROL_HEADOFQ;
break;
case MSG_ACA_TASK:
mpt_req->Control |= MPI_SCSIIO_CONTROL_ACAQ;
break;
case MSG_ORDERED_Q_TAG:
mpt_req->Control |= MPI_SCSIIO_CONTROL_ORDEREDQ;
break;
case MSG_SIMPLE_Q_TAG:
default:
mpt_req->Control |= MPI_SCSIIO_CONTROL_SIMPLEQ;
break;
}
} else {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (mpt->is_fc || mpt->is_sas) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->Control |= MPI_SCSIIO_CONTROL_SIMPLEQ;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/* XXX No such thing for a target doing packetized. */
mpt_req->Control |= MPI_SCSIIO_CONTROL_UNTAGGED;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (mpt->is_spi) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (ccb->ccb_h.flags & CAM_DIS_DISCONNECT) {
mpt_req->Control |= MPI_SCSIIO_CONTROL_NO_DISCONNECT;
}
}
/* Copy the scsi command block into place */
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((ccb->ccb_h.flags & CAM_CDB_POINTER) != 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
bcopy(csio->cdb_io.cdb_ptr, mpt_req->CDB, csio->cdb_len);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
bcopy(csio->cdb_io.cdb_bytes, mpt_req->CDB, csio->cdb_len);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_req->CDBLength = csio->cdb_len;
mpt_req->DataLength = csio->dxfer_len;
mpt_req->SenseBufferLowAddr = req->sense_pbuf;
/*
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
* Do a *short* print here if we're set to MPT_PRT_DEBUG
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
if (mpt->verbose == MPT_PRT_DEBUG) {
mpt_prt(mpt, "mpt_start: %s op 0x%x ",
(mpt_req->Function == MPI_FUNCTION_SCSI_IO_REQUEST)?
"SCSI_IO_REQUEST" : "SCSI_IO_PASSTHRU", mpt_req->CDB[0]);
if (mpt_req->Control != MPI_SCSIIO_CONTROL_NODATATRANSFER) {
mpt_prtc(mpt, "(%s %u byte%s ",
(mpt_req->Control == MPI_SCSIIO_CONTROL_READ)?
"read" : "write", csio->dxfer_len,
(csio->dxfer_len == 1)? ")" : "s)");
}
mpt_prtc(mpt, "tgt %u lun %u req %p:%u\n", tgt,
ccb->ccb_h.target_lun, req, req->serno);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
/*
* If we have any data to send with this command map it into bus space.
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccbh->flags & CAM_DIR_MASK) != CAM_DIR_NONE) {
if ((ccbh->flags & CAM_SCATTER_VALID) == 0) {
/*
* We've been given a pointer to a single buffer.
*/
if ((ccbh->flags & CAM_DATA_PHYS) == 0) {
/*
* Virtual address that needs to translated into
* one or more physical address ranges.
*/
int error;
int s = splsoftvm();
error = bus_dmamap_load(mpt->buffer_dmat,
req->dmap, csio->data_ptr, csio->dxfer_len,
cb, req, 0);
splx(s);
if (error == EINPROGRESS) {
/*
* So as to maintain ordering,
* freeze the controller queue
* until our mapping is
* returned.
*/
xpt_freeze_simq(mpt->sim, 1);
ccbh->status |= CAM_RELEASE_SIMQ;
}
} else {
/*
* We have been given a pointer to single
* physical buffer.
*/
struct bus_dma_segment seg;
seg.ds_addr =
(bus_addr_t)(vm_offset_t)csio->data_ptr;
seg.ds_len = csio->dxfer_len;
(*cb)(req, &seg, 1, 0);
}
} else {
/*
* We have been given a list of addresses.
* This case could be easily supported but they are not
* currently generated by the CAM subsystem so there
* is no point in wasting the time right now.
*/
struct bus_dma_segment *segs;
if ((ccbh->flags & CAM_SG_LIST_PHYS) == 0) {
(*cb)(req, NULL, 0, EFAULT);
} else {
/* Just use the segments provided */
segs = (struct bus_dma_segment *)csio->data_ptr;
(*cb)(req, segs, csio->sglist_cnt, 0);
}
}
} else {
(*cb)(req, NULL, 0, 0);
}
}
static int
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
mpt_bus_reset(struct mpt_softc *mpt, target_id_t tgt, lun_id_t lun,
int sleep_ok)
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
{
int error;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
uint16_t status;
uint8_t response;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
error = mpt_scsi_send_tmf(mpt,
(tgt != CAM_TARGET_WILDCARD || lun != CAM_LUN_WILDCARD) ?
MPI_SCSITASKMGMT_TASKTYPE_TARGET_RESET :
MPI_SCSITASKMGMT_TASKTYPE_RESET_BUS,
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt->is_fc ? MPI_SCSITASKMGMT_MSGFLAGS_LIP_RESET_OPTION : 0,
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
0, /* XXX How do I get the channel ID? */
tgt != CAM_TARGET_WILDCARD ? tgt : 0,
lun != CAM_LUN_WILDCARD ? lun : 0,
0, sleep_ok);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error != 0) {
/*
* mpt_scsi_send_tmf hard resets on failure, so no
* need to do so here.
*/
mpt_prt(mpt,
"mpt_bus_reset: mpt_scsi_send_tmf returned %d\n", error);
return (EIO);
}
/* Wait for bus reset to be processed by the IOC. */
error = mpt_wait_req(mpt, mpt->tmf_req, REQ_STATE_DONE,
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
REQ_STATE_DONE, sleep_ok, 5000);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
status = mpt->tmf_req->IOCStatus;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
response = mpt->tmf_req->ResponseCode;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt->tmf_req->state = REQ_STATE_FREE;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt, "mpt_bus_reset: Reset timed-out. "
"Resetting controller.\n");
mpt_reset(mpt, TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (ETIMEDOUT);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
if ((status & MPI_IOCSTATUS_MASK) != MPI_IOCSTATUS_SUCCESS) {
mpt_prt(mpt, "mpt_bus_reset: TMF IOC Status 0x%x. "
"Resetting controller.\n", status);
mpt_reset(mpt, TRUE);
return (EIO);
}
if (response != MPI_SCSITASKMGMT_RSP_TM_SUCCEEDED &&
response != MPI_SCSITASKMGMT_RSP_TM_COMPLETE) {
mpt_prt(mpt, "mpt_bus_reset: TMF Response 0x%x. "
"Resetting controller.\n", response);
mpt_reset(mpt, TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (EIO);
}
return (0);
}
static int
mpt_fc_reset_link(struct mpt_softc *mpt, int dowait)
{
int r = 0;
request_t *req;
PTR_MSG_FC_PRIMITIVE_SEND_REQUEST fc;
req = mpt_get_request(mpt, FALSE);
if (req == NULL) {
return (ENOMEM);
}
fc = req->req_vbuf;
memset(fc, 0, sizeof(*fc));
fc->SendFlags = MPI_FC_PRIM_SEND_FLAGS_RESET_LINK;
fc->Function = MPI_FUNCTION_FC_PRIMITIVE_SEND;
fc->MsgContext = htole32(req->index | fc_els_handler_id);
mpt_send_cmd(mpt, req);
if (dowait) {
r = mpt_wait_req(mpt, req, REQ_STATE_DONE,
REQ_STATE_DONE, FALSE, 60 * 1000);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (r == 0) {
mpt_free_request(mpt, req);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
return (r);
}
static int
mpt_cam_event(struct mpt_softc *mpt, request_t *req,
MSG_EVENT_NOTIFY_REPLY *msg)
{
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
switch(msg->Event & 0xFF) {
case MPI_EVENT_UNIT_ATTENTION:
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
mpt_prt(mpt, "UNIT ATTENTION: Bus: 0x%02x TargetID: 0x%02x\n",
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
(msg->Data[0] >> 8) & 0xff, msg->Data[0] & 0xff);
break;
case MPI_EVENT_IOC_BUS_RESET:
/* We generated a bus reset */
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
mpt_prt(mpt, "IOC Generated Bus Reset Port: %d\n",
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
(msg->Data[0] >> 8) & 0xff);
xpt_async(AC_BUS_RESET, mpt->path, NULL);
break;
case MPI_EVENT_EXT_BUS_RESET:
/* Someone else generated a bus reset */
mpt_prt(mpt, "External Bus Reset Detected\n");
/*
* These replies don't return EventData like the MPI
* spec says they do
*/
xpt_async(AC_BUS_RESET, mpt->path, NULL);
break;
case MPI_EVENT_RESCAN:
/*
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* In general this means a device has been added to the loop.
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
mpt_prt(mpt, "Rescan Port: %d\n", (msg->Data[0] >> 8) & 0xff);
/* xpt_async(AC_FOUND_DEVICE, path, NULL); */
break;
case MPI_EVENT_LINK_STATUS_CHANGE:
mpt_prt(mpt, "Port %d: LinkState: %s\n",
(msg->Data[1] >> 8) & 0xff,
((msg->Data[0] & 0xff) == 0)? "Failed" : "Active");
break;
case MPI_EVENT_LOOP_STATE_CHANGE:
switch ((msg->Data[0] >> 16) & 0xff) {
case 0x01:
mpt_prt(mpt,
"Port 0x%x: FC LinkEvent: LIP(%02x,%02x) "
"(Loop Initialization)\n",
(msg->Data[1] >> 8) & 0xff,
(msg->Data[0] >> 8) & 0xff,
(msg->Data[0] ) & 0xff);
switch ((msg->Data[0] >> 8) & 0xff) {
case 0xF7:
if ((msg->Data[0] & 0xff) == 0xF7) {
mpt_prt(mpt, "Device needs AL_PA\n");
} else {
mpt_prt(mpt, "Device %02x doesn't like "
"FC performance\n",
msg->Data[0] & 0xFF);
}
break;
case 0xF8:
if ((msg->Data[0] & 0xff) == 0xF7) {
mpt_prt(mpt, "Device had loop failure "
"at its receiver prior to acquiring"
" AL_PA\n");
} else {
mpt_prt(mpt, "Device %02x detected loop"
" failure at its receiver\n",
msg->Data[0] & 0xFF);
}
break;
default:
mpt_prt(mpt, "Device %02x requests that device "
"%02x reset itself\n",
msg->Data[0] & 0xFF,
(msg->Data[0] >> 8) & 0xFF);
break;
}
break;
case 0x02:
mpt_prt(mpt, "Port 0x%x: FC LinkEvent: "
"LPE(%02x,%02x) (Loop Port Enable)\n",
(msg->Data[1] >> 8) & 0xff, /* Port */
(msg->Data[0] >> 8) & 0xff, /* Character 3 */
(msg->Data[0] ) & 0xff /* Character 4 */);
break;
case 0x03:
mpt_prt(mpt, "Port 0x%x: FC LinkEvent: "
"LPB(%02x,%02x) (Loop Port Bypass)\n",
(msg->Data[1] >> 8) & 0xff, /* Port */
(msg->Data[0] >> 8) & 0xff, /* Character 3 */
(msg->Data[0] ) & 0xff /* Character 4 */);
break;
default:
mpt_prt(mpt, "Port 0x%x: FC LinkEvent: Unknown "
"FC event (%02x %02x %02x)\n",
(msg->Data[1] >> 8) & 0xff, /* Port */
(msg->Data[0] >> 16) & 0xff, /* Event */
(msg->Data[0] >> 8) & 0xff, /* Character 3 */
(msg->Data[0] ) & 0xff /* Character 4 */);
}
break;
case MPI_EVENT_LOGOUT:
mpt_prt(mpt, "FC Logout Port: %d N_PortID: %02x\n",
(msg->Data[1] >> 8) & 0xff, msg->Data[0]);
break;
case MPI_EVENT_EVENT_CHANGE:
mpt_lprt(mpt, MPT_PRT_DEBUG,
"mpt_cam_event: MPI_EVENT_EVENT_CHANGE\n");
break;
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
case MPI_EVENT_QUEUE_FULL:
{
Connect up a QUEUE FULL event with CAM and adjust openings. Unfortunately, the QUEUE FULL event only tells you Bus && Target. It doesn't tell you lun. In order for the XPT_REL_SIMQ action to work, we have to have a real lun. But which one? For now, just iterate over MPT_MAX_LUNS. Practically speaking, this is only going to be happening for lower quality SAS or SATA drives behind the SAS controller, which means only lun 0, so it's not so bad. Helpful Reminder Nagging from: John Baldwin, Fred Whiteside MFC after: 5 days
2006-09-21 20:35:12 +00:00
struct cam_sim *sim;
struct cam_path *tmppath;
struct ccb_relsim crs;
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
PTR_EVENT_DATA_QUEUE_FULL pqf =
(PTR_EVENT_DATA_QUEUE_FULL) msg->Data;
Connect up a QUEUE FULL event with CAM and adjust openings. Unfortunately, the QUEUE FULL event only tells you Bus && Target. It doesn't tell you lun. In order for the XPT_REL_SIMQ action to work, we have to have a real lun. But which one? For now, just iterate over MPT_MAX_LUNS. Practically speaking, this is only going to be happening for lower quality SAS or SATA drives behind the SAS controller, which means only lun 0, so it's not so bad. Helpful Reminder Nagging from: John Baldwin, Fred Whiteside MFC after: 5 days
2006-09-21 20:35:12 +00:00
lun_id_t lun_id;
mpt_prt(mpt, "QUEUE FULL EVENT: Bus 0x%02x Target 0x%02x Depth "
"%d\n", pqf->Bus, pqf->TargetID, pqf->CurrentDepth);
if (mpt->phydisk_sim) {
sim = mpt->phydisk_sim;
} else {
sim = mpt->sim;
}
MPTLOCK_2_CAMLOCK(mpt);
for (lun_id = 0; lun_id < MPT_MAX_LUNS; lun_id++) {
if (xpt_create_path(&tmppath, NULL, cam_sim_path(sim),
pqf->TargetID, lun_id) != CAM_REQ_CMP) {
mpt_prt(mpt, "unable to create a path to send "
"XPT_REL_SIMQ");
CAMLOCK_2_MPTLOCK(mpt);
break;
}
xpt_setup_ccb(&crs.ccb_h, tmppath, 5);
crs.ccb_h.func_code = XPT_REL_SIMQ;
crs.release_flags = RELSIM_ADJUST_OPENINGS;
crs.openings = pqf->CurrentDepth - 1;
xpt_action((union ccb *)&crs);
if (crs.ccb_h.status != CAM_REQ_CMP) {
mpt_prt(mpt, "XPT_REL_SIMQ failed\n");
}
xpt_free_path(tmppath);
}
CAMLOCK_2_MPTLOCK(mpt);
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
break;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
case MPI_EVENT_SAS_DEVICE_STATUS_CHANGE:
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
{
mpt_lprt(mpt, MPT_PRT_DEBUG,
"mpt_cam_event: SAS_DEVICE_STATUS_CHANGE\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
break;
Create a 'ready' handler for each personality. The purpose of this handler is to able to be called after *all* attach and enable events are done. We establish a SYSINIT hook to call this handler. The current usage for it is to add scsi target resources *after* all enables are done. There seems to be some dependencies between different halves of a dual-port with respect to target mode. Put in more meaningful event messages for some events- in particular QUEUE FULL events so we can see what the queue depth was when the IOC sent us this message. MFC after: 1 week
2006-09-07 23:08:21 +00:00
}
case MPI_EVENT_SAS_SES:
{
mpt_lprt(mpt, MPT_PRT_DEBUG,
"mpt_cam_event: MPI_EVENT_SAS_SES\n");
break;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
default:
mpt_lprt(mpt, MPT_PRT_WARN, "mpt_cam_event: 0x%x\n",
msg->Event & 0xFF);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return (0);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return (1);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
* Reply path for all SCSI I/O requests, called from our
* interrupt handler by extracting our handler index from
* the MsgContext field of the reply from the IOC.
*
* This routine is optimized for the common case of a
* completion without error. All exception handling is
* offloaded to non-inlined helper routines to minimize
* cache footprint.
*/
static int
mpt_scsi_reply_handler(struct mpt_softc *mpt, request_t *req,
uint32_t reply_desc, MSG_DEFAULT_REPLY *reply_frame)
{
MSG_SCSI_IO_REQUEST *scsi_req;
union ccb *ccb;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
target_id_t tgt;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (req->state == REQ_STATE_FREE) {
mpt_prt(mpt, "mpt_scsi_reply_handler: req already free\n");
return (TRUE);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
scsi_req = (MSG_SCSI_IO_REQUEST *)req->req_vbuf;
ccb = req->ccb;
if (ccb == NULL) {
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
mpt_prt(mpt, "mpt_scsi_reply_handler: req %p:%u with no ccb\n",
req, req->serno);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return (TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
tgt = scsi_req->TargetID;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
untimeout(mpt_timeout, ccb, ccb->ccb_h.timeout_ch);
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE) {
bus_dmasync_op_t op;
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN)
op = BUS_DMASYNC_POSTREAD;
else
op = BUS_DMASYNC_POSTWRITE;
bus_dmamap_sync(mpt->buffer_dmat, req->dmap, op);
bus_dmamap_unload(mpt->buffer_dmat, req->dmap);
}
if (reply_frame == NULL) {
/*
* Context only reply, completion without error status.
*/
ccb->csio.resid = 0;
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
ccb->csio.scsi_status = SCSI_STATUS_OK;
} else {
mpt_scsi_reply_frame_handler(mpt, req, reply_frame);
}
if (mpt->outofbeer) {
ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
mpt->outofbeer = 0;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG, "THAWQ\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (scsi_req->CDB[0] == INQUIRY && (scsi_req->CDB[1] & SI_EVPD) == 0) {
struct scsi_inquiry_data *iq =
(struct scsi_inquiry_data *)ccb->csio.data_ptr;
if (scsi_req->Function ==
MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH) {
/*
* Fake out the device type so that only the
* pass-thru device will attach.
*/
iq->device &= ~0x1F;
iq->device |= T_NODEVICE;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
if (mpt->verbose == MPT_PRT_DEBUG) {
mpt_prt(mpt, "mpt_scsi_reply_handler: %p:%u complete\n",
req, req->serno);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
if ((req->state & REQ_STATE_TIMEDOUT) == 0) {
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
} else {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt, "completing timedout/aborted req %p:%u\n",
req, req->serno);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
TAILQ_REMOVE(&mpt->request_timeout_list, req, links);
}
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
KASSERT((req->state & REQ_STATE_NEED_WAKEUP) == 0,
("CCB req needed wakeup"));
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_req_not_spcl(mpt, req, "mpt_scsi_reply_handler", __LINE__);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#endif
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_free_request(mpt, req);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
return (TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
static int
mpt_scsi_tmf_reply_handler(struct mpt_softc *mpt, request_t *req,
uint32_t reply_desc, MSG_DEFAULT_REPLY *reply_frame)
{
MSG_SCSI_TASK_MGMT_REPLY *tmf_reply;
KASSERT(req == mpt->tmf_req, ("TMF Reply not using mpt->tmf_req"));
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_not_spcl(mpt, req, "mpt_scsi_tmf_reply_handler", __LINE__);
#endif
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
tmf_reply = (MSG_SCSI_TASK_MGMT_REPLY *)reply_frame;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* Record IOC Status and Response Code of TMF for any waiters. */
req->IOCStatus = le16toh(tmf_reply->IOCStatus);
req->ResponseCode = tmf_reply->ResponseCode;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_lprt(mpt, MPT_PRT_INFO, "TMF complete: req %p:%u status 0x%x\n",
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
req, req->serno, le16toh(tmf_reply->IOCStatus));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
if ((req->state & REQ_STATE_NEED_WAKEUP) != 0) {
req->state |= REQ_STATE_DONE;
wakeup(req);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt->tmf_req->state = REQ_STATE_FREE;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
return (TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
/*
* XXX: Move to definitions file
*/
#define ELS 0x22
#define FC4LS 0x32
#define ABTS 0x81
#define BA_ACC 0x84
#define LS_RJT 0x01
#define LS_ACC 0x02
#define PLOGI 0x03
#define LOGO 0x05
#define SRR 0x14
#define PRLI 0x20
#define PRLO 0x21
#define ADISC 0x52
#define RSCN 0x61
static void
mpt_fc_els_send_response(struct mpt_softc *mpt, request_t *req,
PTR_MSG_LINK_SERVICE_BUFFER_POST_REPLY rp, U8 length)
{
MSG_LINK_SERVICE_RSP_REQUEST tmp;
PTR_MSG_LINK_SERVICE_RSP_REQUEST rsp;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
* We are going to reuse the ELS request to send this response back.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
rsp = &tmp;
memset(rsp, 0, sizeof(*rsp));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
#ifdef USE_IMMEDIATE_LINK_DATA
/*
* Apparently the IMMEDIATE stuff doesn't seem to work.
*/
rsp->RspFlags = LINK_SERVICE_RSP_FLAGS_IMMEDIATE;
#endif
rsp->RspLength = length;
rsp->Function = MPI_FUNCTION_FC_LINK_SRVC_RSP;
rsp->MsgContext = htole32(req->index | fc_els_handler_id);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
* Copy over information from the original reply frame to
* it's correct place in the response.
*/
memcpy((U8 *)rsp + 0x0c, (U8 *)rp + 0x1c, 24);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
* And now copy back the temporary area to the original frame.
*/
memcpy(req->req_vbuf, rsp, sizeof (MSG_LINK_SERVICE_RSP_REQUEST));
rsp = req->req_vbuf;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
#ifdef USE_IMMEDIATE_LINK_DATA
memcpy((U8 *)&rsp->SGL, &((U8 *)req->req_vbuf)[MPT_RQSL(mpt)], length);
#else
{
PTR_SGE_SIMPLE32 se = (PTR_SGE_SIMPLE32) &rsp->SGL;
bus_addr_t paddr = req->req_pbuf;
paddr += MPT_RQSL(mpt);
se->FlagsLength =
MPI_SGE_FLAGS_HOST_TO_IOC |
MPI_SGE_FLAGS_SIMPLE_ELEMENT |
MPI_SGE_FLAGS_LAST_ELEMENT |
MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
se->FlagsLength <<= MPI_SGE_FLAGS_SHIFT;
se->FlagsLength |= (length);
se->Address = (uint32_t) paddr;
}
#endif
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
* Send it on...
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_send_cmd(mpt, req);
}
static int
mpt_fc_els_reply_handler(struct mpt_softc *mpt, request_t *req,
uint32_t reply_desc, MSG_DEFAULT_REPLY *reply_frame)
{
PTR_MSG_LINK_SERVICE_BUFFER_POST_REPLY rp =
(PTR_MSG_LINK_SERVICE_BUFFER_POST_REPLY) reply_frame;
U8 rctl;
U8 type;
U8 cmd;
U16 status = le16toh(reply_frame->IOCStatus);
U32 *elsbuf;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
int ioindex;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
int do_refresh = TRUE;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
KASSERT(mpt_req_on_free_list(mpt, req) == 0,
("fc_els_reply_handler: req %p:%u for function %x on freelist!",
req, req->serno, rp->Function));
if (rp->Function != MPI_FUNCTION_FC_PRIMITIVE_SEND) {
mpt_req_spcl(mpt, req, "fc_els_reply_handler", __LINE__);
} else {
mpt_req_not_spcl(mpt, req, "fc_els_reply_handler", __LINE__);
}
#endif
Remove MPT_PRT_INVARIANT- it was a silly idea.
2006-05-04 02:34:18 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"FC_ELS Complete: req %p:%u, reply %p function %x\n",
req, req->serno, reply_frame, reply_frame->Function);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (status != MPI_IOCSTATUS_SUCCESS) {
mpt_prt(mpt, "ELS REPLY STATUS 0x%x for Function %x\n",
status, reply_frame->Function);
if (status == MPI_IOCSTATUS_INVALID_STATE) {
/*
* XXX: to get around shutdown issue
*/
mpt->disabled = 1;
return (TRUE);
}
return (TRUE);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
* If the function of a link service response, we recycle the
* response to be a refresh for a new link service request.
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
*
* The request pointer is bogus in this case and we have to fetch
* it based upon the TransactionContext.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rp->Function == MPI_FUNCTION_FC_LINK_SRVC_RSP) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* Freddie Uncle Charlie Katie */
/* We don't get the IOINDEX as part of the Link Svc Rsp */
for (ioindex = 0; ioindex < mpt->els_cmds_allocated; ioindex++)
if (mpt->els_cmd_ptrs[ioindex] == req) {
break;
}
KASSERT(ioindex < mpt->els_cmds_allocated,
("can't find my mommie!"));
/* remove from active list as we're going to re-post it */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
mpt_fc_post_els(mpt, req, ioindex);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (TRUE);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rp->Function == MPI_FUNCTION_FC_PRIMITIVE_SEND) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* remove from active list as we're done */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
If we're in mpt_wait_req and the command times out, mark it as timed out. Don't try and free the config request for read_cfg_header that times out because it's still active. Put in code for the config reply handler that will then free up timed out requests. Fix the FC_PRIMITIVE_SEND completion to not try and free a command twice. Dunno how this possibly could have been working for awhile. MFC after: 2 weeks
2006-07-16 03:34:55 +00:00
if (req->state & REQ_STATE_TIMEDOUT) {
mpt_lprt(mpt, MPT_PRT_DEBUG,
"Sync Primitive Send Completed After Timeout\n");
mpt_free_request(mpt, req);
} else if ((req->state & REQ_STATE_NEED_WAKEUP) == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
"Async Primitive Send Complete\n");
mpt_free_request(mpt, req);
} else {
mpt_lprt(mpt, MPT_PRT_DEBUG,
If we're in mpt_wait_req and the command times out, mark it as timed out. Don't try and free the config request for read_cfg_header that times out because it's still active. Put in code for the config reply handler that will then free up timed out requests. Fix the FC_PRIMITIVE_SEND completion to not try and free a command twice. Dunno how this possibly could have been working for awhile. MFC after: 2 weeks
2006-07-16 03:34:55 +00:00
"Sync Primitive Send Complete- Waking Waiter\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
wakeup(req);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (TRUE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rp->Function != MPI_FUNCTION_FC_LINK_SRVC_BUF_POST) {
mpt_prt(mpt, "unexpected ELS_REPLY: Function 0x%x Flags %x "
"Length %d Message Flags %x\n", rp->Function, rp->Flags,
rp->MsgLength, rp->MsgFlags);
return (TRUE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rp->MsgLength <= 5) {
/*
* This is just a ack of an original ELS buffer post
*/
Remove MPT_PRT_INVARIANT- it was a silly idea.
2006-05-04 02:34:18 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"RECV'd ACK of FC_ELS buf post %p:%u\n", req, req->serno);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
return (TRUE);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
rctl = (le32toh(rp->Rctl_Did) & MPI_FC_RCTL_MASK) >> MPI_FC_RCTL_SHIFT;
type = (le32toh(rp->Type_Fctl) & MPI_FC_TYPE_MASK) >> MPI_FC_TYPE_SHIFT;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
elsbuf = &((U32 *)req->req_vbuf)[MPT_RQSL(mpt)/sizeof (U32)];
cmd = be32toh(elsbuf[0]) >> 24;
if (rp->Flags & MPI_LS_BUF_POST_REPLY_FLAG_NO_RSP_NEEDED) {
mpt_lprt(mpt, MPT_PRT_ALWAYS, "ELS_REPLY: response unneeded\n");
return (TRUE);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
ioindex = le32toh(rp->TransactionContext);
req = mpt->els_cmd_ptrs[ioindex];
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (rctl == ELS && type == 1) {
switch (cmd) {
case PRLI:
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
* Send back a PRLI ACC
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_prt(mpt, "PRLI from 0x%08x%08x\n",
le32toh(rp->Wwn.PortNameHigh),
le32toh(rp->Wwn.PortNameLow));
elsbuf[0] = htobe32(0x02100014);
elsbuf[1] |= htobe32(0x00000100);
elsbuf[4] = htobe32(0x00000002);
if (mpt->role & MPT_ROLE_TARGET)
elsbuf[4] |= htobe32(0x00000010);
if (mpt->role & MPT_ROLE_INITIATOR)
elsbuf[4] |= htobe32(0x00000020);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* remove from active list as we're done */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_fc_els_send_response(mpt, req, rp, 20);
do_refresh = FALSE;
break;
case PRLO:
memset(elsbuf, 0, 5 * (sizeof (U32)));
elsbuf[0] = htobe32(0x02100014);
elsbuf[1] = htobe32(0x08000100);
mpt_prt(mpt, "PRLO from 0x%08x%08x\n",
le32toh(rp->Wwn.PortNameHigh),
le32toh(rp->Wwn.PortNameLow));
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* remove from active list as we're done */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_fc_els_send_response(mpt, req, rp, 20);
do_refresh = FALSE;
break;
default:
mpt_prt(mpt, "ELS TYPE 1 COMMAND: %x\n", cmd);
break;
}
} else if (rctl == ABTS && type == 0) {
uint16_t rx_id = le16toh(rp->Rxid);
uint16_t ox_id = le16toh(rp->Oxid);
request_t *tgt_req = NULL;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_prt(mpt,
"ELS: ABTS OX_ID 0x%x RX_ID 0x%x from 0x%08x%08x\n",
ox_id, rx_id, le32toh(rp->Wwn.PortNameHigh),
le32toh(rp->Wwn.PortNameLow));
if (rx_id >= mpt->mpt_max_tgtcmds) {
mpt_prt(mpt, "Bad RX_ID 0x%x\n", rx_id);
} else if (mpt->tgt_cmd_ptrs == NULL) {
mpt_prt(mpt, "No TGT CMD PTRS\n");
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
tgt_req = mpt->tgt_cmd_ptrs[rx_id];
}
if (tgt_req) {
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, tgt_req);
uint8_t *vbuf;
union ccb *ccb = tgt->ccb;
uint32_t ct_id;
vbuf = tgt_req->req_vbuf;
vbuf += MPT_RQSL(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
* Check to make sure we have the correct command
* The reply descriptor in the target state should
* should contain an IoIndex that should match the
* RX_ID.
*
* It'd be nice to have OX_ID to crosscheck with
* as well.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ct_id = GET_IO_INDEX(tgt->reply_desc);
if (ct_id != rx_id) {
mpt_lprt(mpt, MPT_PRT_ERROR, "ABORT Mismatch: "
"RX_ID received=0x%x; RX_ID in cmd=0x%x\n",
rx_id, ct_id);
goto skip;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ccb = tgt->ccb;
if (ccb) {
mpt_prt(mpt,
"CCB (%p): lun %u flags %x status %x\n",
ccb, ccb->ccb_h.target_lun,
ccb->ccb_h.flags, ccb->ccb_h.status);
}
mpt_prt(mpt, "target state 0x%x resid %u xfrd %u rpwrd "
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
"%x nxfers %x\n", tgt->state,
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
tgt->resid, tgt->bytes_xfered, tgt->reply_desc,
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->nxfers);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
skip:
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (mpt_abort_target_cmd(mpt, tgt_req)) {
mpt_prt(mpt, "unable to start TargetAbort\n");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
} else {
mpt_prt(mpt, "no back pointer for RX_ID 0x%x\n", rx_id);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
memset(elsbuf, 0, 5 * (sizeof (U32)));
elsbuf[0] = htobe32(0);
elsbuf[1] = htobe32((ox_id << 16) | rx_id);
elsbuf[2] = htobe32(0x000ffff);
/*
* Dork with the reply frame so that the reponse to it
* will be correct.
*/
rp->Rctl_Did += ((BA_ACC - ABTS) << MPI_FC_RCTL_SHIFT);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* remove from active list as we're done */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_fc_els_send_response(mpt, req, rp, 12);
do_refresh = FALSE;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
} else {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_prt(mpt, "ELS: RCTL %x TYPE %x CMD %x\n", rctl, type, cmd);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (do_refresh == TRUE) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/* remove from active list as we're done */
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
req->state &= ~REQ_STATE_QUEUED;
req->state |= REQ_STATE_DONE;
mpt_fc_post_els(mpt, req, ioindex);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
return (TRUE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
* Clean up all SCSI Initiator personality state in response
* to a controller reset.
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
static void
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_cam_ioc_reset(struct mpt_softc *mpt, int type)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
/*
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
* The pending list is already run down by
* the generic handler. Perform the same
* operation on the timed out request list.
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_complete_request_chain(mpt, &mpt->request_timeout_list,
MPI_IOCSTATUS_INVALID_STATE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* XXX: We need to repost ELS and Target Command Buffers?
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Inform the XPT that a bus reset has occurred.
*/
xpt_async(AC_BUS_RESET, mpt->path, NULL);
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Parse additional completion information in the reply
* frame for SCSI I/O requests.
*/
static int
mpt_scsi_reply_frame_handler(struct mpt_softc *mpt, request_t *req,
MSG_DEFAULT_REPLY *reply_frame)
{
union ccb *ccb;
MSG_SCSI_IO_REPLY *scsi_io_reply;
u_int ioc_status;
u_int sstate;
u_int loginfo;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
MPT_DUMP_REPLY_FRAME(mpt, reply_frame);
KASSERT(reply_frame->Function == MPI_FUNCTION_SCSI_IO_REQUEST
|| reply_frame->Function == MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH,
("MPT SCSI I/O Handler called with incorrect reply type"));
KASSERT((reply_frame->MsgFlags & MPI_MSGFLAGS_CONTINUATION_REPLY) == 0,
("MPT SCSI I/O Handler called with continuation reply"));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
scsi_io_reply = (MSG_SCSI_IO_REPLY *)reply_frame;
ioc_status = le16toh(scsi_io_reply->IOCStatus);
loginfo = ioc_status & MPI_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE;
ioc_status &= MPI_IOCSTATUS_MASK;
sstate = scsi_io_reply->SCSIState;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
ccb = req->ccb;
ccb->csio.resid =
ccb->csio.dxfer_len - le32toh(scsi_io_reply->TransferCount);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((sstate & MPI_SCSI_STATE_AUTOSENSE_VALID) != 0
&& (ccb->ccb_h.flags & (CAM_SENSE_PHYS | CAM_SENSE_PTR)) == 0) {
ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
ccb->csio.sense_resid =
ccb->csio.sense_len - scsi_io_reply->SenseCount;
bcopy(req->sense_vbuf, &ccb->csio.sense_data,
min(ccb->csio.sense_len, scsi_io_reply->SenseCount));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((sstate & MPI_SCSI_STATE_QUEUE_TAG_REJECTED) != 0) {
/*
* Tag messages rejected, but non-tagged retry
* was successful.
XXXX
mpt_set_tags(mpt, devinfo, MPT_QUEUE_NONE);
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
switch(ioc_status) {
case MPI_IOCSTATUS_SCSI_RESIDUAL_MISMATCH:
/*
* XXX
* Linux driver indicates that a zero
* transfer length with this error code
* indicates a CRC error.
*
* No need to swap the bytes for checking
* against zero.
*/
if (scsi_io_reply->TransferCount == 0) {
mpt_set_ccb_status(ccb, CAM_UNCOR_PARITY);
break;
}
/* FALLTHROUGH */
case MPI_IOCSTATUS_SCSI_DATA_UNDERRUN:
case MPI_IOCSTATUS_SUCCESS:
case MPI_IOCSTATUS_SCSI_RECOVERED_ERROR:
if ((sstate & MPI_SCSI_STATE_NO_SCSI_STATUS) != 0) {
/*
* Status was never returned for this transaction.
*/
mpt_set_ccb_status(ccb, CAM_UNEXP_BUSFREE);
} else if (scsi_io_reply->SCSIStatus != SCSI_STATUS_OK) {
ccb->csio.scsi_status = scsi_io_reply->SCSIStatus;
mpt_set_ccb_status(ccb, CAM_SCSI_STATUS_ERROR);
if ((sstate & MPI_SCSI_STATE_AUTOSENSE_FAILED) != 0)
mpt_set_ccb_status(ccb, CAM_AUTOSENSE_FAIL);
} else if ((sstate & MPI_SCSI_STATE_RESPONSE_INFO_VALID) != 0) {
/* XXX Handle SPI-Packet and FCP-2 reponse info. */
mpt_set_ccb_status(ccb, CAM_REQ_CMP_ERR);
} else
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case MPI_IOCSTATUS_SCSI_DATA_OVERRUN:
mpt_set_ccb_status(ccb, CAM_DATA_RUN_ERR);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case MPI_IOCSTATUS_SCSI_IO_DATA_ERROR:
mpt_set_ccb_status(ccb, CAM_UNCOR_PARITY);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case MPI_IOCSTATUS_SCSI_DEVICE_NOT_THERE:
/*
* Since selection timeouts and "device really not
* there" are grouped into this error code, report
* selection timeout. Selection timeouts are
* typically retried before giving up on the device
* whereas "device not there" errors are considered
* unretryable.
*/
mpt_set_ccb_status(ccb, CAM_SEL_TIMEOUT);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case MPI_IOCSTATUS_SCSI_PROTOCOL_ERROR:
mpt_set_ccb_status(ccb, CAM_SEQUENCE_FAIL);
break;
case MPI_IOCSTATUS_SCSI_INVALID_BUS:
mpt_set_ccb_status(ccb, CAM_PATH_INVALID);
break;
case MPI_IOCSTATUS_SCSI_INVALID_TARGETID:
mpt_set_ccb_status(ccb, CAM_TID_INVALID);
break;
case MPI_IOCSTATUS_SCSI_TASK_MGMT_FAILED:
ccb->ccb_h.status = CAM_UA_TERMIO;
break;
case MPI_IOCSTATUS_INVALID_STATE:
/*
* The IOC has been reset. Emulate a bus reset.
*/
/* FALLTHROUGH */
case MPI_IOCSTATUS_SCSI_EXT_TERMINATED:
ccb->ccb_h.status = CAM_SCSI_BUS_RESET;
break;
case MPI_IOCSTATUS_SCSI_TASK_TERMINATED:
case MPI_IOCSTATUS_SCSI_IOC_TERMINATED:
/*
* Don't clobber any timeout status that has
* already been set for this transaction. We
* want the SCSI layer to be able to differentiate
* between the command we aborted due to timeout
* and any innocent bystanders.
*/
if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_INPROG)
break;
mpt_set_ccb_status(ccb, CAM_REQ_TERMIO);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case MPI_IOCSTATUS_INSUFFICIENT_RESOURCES:
mpt_set_ccb_status(ccb, CAM_RESRC_UNAVAIL);
break;
case MPI_IOCSTATUS_BUSY:
mpt_set_ccb_status(ccb, CAM_BUSY);
break;
case MPI_IOCSTATUS_INVALID_FUNCTION:
case MPI_IOCSTATUS_INVALID_SGL:
case MPI_IOCSTATUS_INTERNAL_ERROR:
case MPI_IOCSTATUS_INVALID_FIELD:
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
default:
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/* XXX
* Some of the above may need to kick
* of a recovery action!!!!
*/
ccb->ccb_h.status = CAM_UNREC_HBA_ERROR;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) {
mpt_freeze_ccb(ccb);
}
return (TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static void
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_action(struct cam_sim *sim, union ccb *ccb)
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
{
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
struct mpt_softc *mpt;
struct ccb_trans_settings *cts;
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
target_id_t tgt;
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
lun_id_t lun;
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
int raid_passthru;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE, ("mpt_action\n"));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt = (struct mpt_softc *)cam_sim_softc(sim);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
KASSERT(MPT_OWNED(mpt) == 0, ("mpt owned on entrance to mpt_action"));
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
raid_passthru = (sim == mpt->phydisk_sim);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt = ccb->ccb_h.target_id;
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
lun = ccb->ccb_h.target_lun;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (raid_passthru && ccb->ccb_h.func_code != XPT_PATH_INQ &&
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
ccb->ccb_h.func_code != XPT_RESET_BUS &&
ccb->ccb_h.func_code != XPT_RESET_DEV) {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
CAMLOCK_2_MPTLOCK(mpt);
if (mpt_map_physdisk(mpt, ccb, &tgt) != 0) {
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_DEV_NOT_THERE);
xpt_done(ccb);
return;
}
MPTLOCK_2_CAMLOCK(mpt);
}
ccb->ccb_h.ccb_mpt_ptr = mpt;
switch (ccb->ccb_h.func_code) {
case XPT_SCSI_IO: /* Execute the requested I/O operation */
/*
* Do a couple of preliminary checks...
*/
if ((ccb->ccb_h.flags & CAM_CDB_POINTER) != 0) {
if ((ccb->ccb_h.flags & CAM_CDB_PHYS) != 0) {
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_REQ_INVALID);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
/* Max supported CDB length is 16 bytes */
/* XXX Unless we implement the new 32byte message type */
if (ccb->csio.cdb_len >
sizeof (((PTR_MSG_SCSI_IO_REQUEST)0)->CDB)) {
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_REQ_INVALID);
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
break;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
ccb->csio.scsi_status = SCSI_STATUS_OK;
mpt_start(sim, ccb);
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
return;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case XPT_RESET_BUS:
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
case XPT_RESET_DEV:
mpt_lprt(mpt, MPT_PRT_DEBUG,
ccb->ccb_h.func_code == XPT_RESET_BUS ?
"XPT_RESET_BUS\n" : "XPT_RESET_DEV\n");
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
(void) mpt_bus_reset(mpt, tgt, lun, FALSE);
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
* mpt_bus_reset is always successful in that it
* will fall back to a hard reset should a bus
* reset attempt fail.
*/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
break;
case XPT_ABORT:
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
{
union ccb *accb = ccb->cab.abort_ccb;
CAMLOCK_2_MPTLOCK(mpt);
switch (accb->ccb_h.func_code) {
case XPT_ACCEPT_TARGET_IO:
case XPT_IMMED_NOTIFY:
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
ccb->ccb_h.status = mpt_abort_target_ccb(mpt, ccb);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
break;
case XPT_CONT_TARGET_IO:
mpt_prt(mpt, "cannot abort active CTIOs yet\n");
ccb->ccb_h.status = CAM_UA_ABORT;
break;
case XPT_SCSI_IO:
ccb->ccb_h.status = CAM_UA_ABORT;
break;
default:
ccb->ccb_h.status = CAM_REQ_INVALID;
break;
}
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
#define IS_CURRENT_SETTINGS(c) ((c)->type == CTS_TYPE_CURRENT_SETTINGS)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
#define DP_DISC_ENABLE 0x1
#define DP_DISC_DISABL 0x2
#define DP_DISC (DP_DISC_ENABLE|DP_DISC_DISABL)
#define DP_TQING_ENABLE 0x4
#define DP_TQING_DISABL 0x8
#define DP_TQING (DP_TQING_ENABLE|DP_TQING_DISABL)
#define DP_WIDE 0x10
#define DP_NARROW 0x20
#define DP_WIDTH (DP_WIDE|DP_NARROW)
#define DP_SYNC 0x40
case XPT_SET_TRAN_SETTINGS: /* Nexus Settings */
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
{
struct ccb_trans_settings_scsi *scsi;
struct ccb_trans_settings_spi *spi;
uint8_t dval;
u_int period;
u_int offset;
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
int i, j;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cts = &ccb->cts;
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (mpt->is_fc || mpt->is_sas) {
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
break;
}
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
/*
* Skip attempting settings on RAID volume disks.
* Other devices on the bus get the normal treatment.
*/
if (mpt->phydisk_sim && raid_passthru == 0 &&
mpt_is_raid_volume(mpt, tgt) != 0) {
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
"skipping transfer settings for RAID volumes\n");
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
break;
}
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
i = mpt->mpt_port_page2.PortSettings &
MPI_SCSIPORTPAGE2_PORT_MASK_NEGO_MASTER_SETTINGS;
j = mpt->mpt_port_page2.PortFlags &
MPI_SCSIPORTPAGE2_PORT_FLAGS_DV_MASK;
if (i == MPI_SCSIPORTPAGE2_PORT_ALL_MASTER_SETTINGS &&
j == MPI_SCSIPORTPAGE2_PORT_FLAGS_OFF_DV) {
mpt_lprt(mpt, MPT_PRT_ALWAYS,
"honoring BIOS transfer negotiations\n");
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
break;
}
dval = 0;
period = 0;
offset = 0;
scsi = &cts->proto_specific.scsi;
spi = &cts->xport_specific.spi;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if ((spi->valid & CTS_SPI_VALID_DISC) != 0) {
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
dval |= ((spi->flags & CTS_SPI_FLAGS_DISC_ENB) != 0) ?
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
DP_DISC_ENABLE : DP_DISC_DISABL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if ((scsi->valid & CTS_SCSI_VALID_TQ) != 0) {
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
dval |= ((scsi->flags & CTS_SCSI_FLAGS_TAG_ENB) != 0) ?
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
DP_TQING_ENABLE : DP_TQING_DISABL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if ((spi->valid & CTS_SPI_VALID_BUS_WIDTH) != 0) {
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
dval |= (spi->bus_width == MSG_EXT_WDTR_BUS_16_BIT) ?
DP_WIDE : DP_NARROW;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Fix some negotiation issues (like not being able to negotiate async)
2006-11-02 23:19:00 +00:00
if (spi->valid & CTS_SPI_VALID_SYNC_OFFSET) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
dval |= DP_SYNC;
offset = spi->sync_offset;
Fix some negotiation issues (like not being able to negotiate async)
2006-11-02 23:19:00 +00:00
} else {
PTR_CONFIG_PAGE_SCSI_DEVICE_1 ptr =
&mpt->mpt_dev_page1[tgt];
offset = ptr->RequestedParameters;
offset &= MPI_SCSIDEVPAGE1_RP_MAX_SYNC_OFFSET_MASK;
offset >>= MPI_SCSIDEVPAGE1_RP_SHIFT_MAX_SYNC_OFFSET;
}
if (spi->valid & CTS_SPI_VALID_SYNC_RATE) {
dval |= DP_SYNC;
period = spi->sync_period;
} else {
PTR_CONFIG_PAGE_SCSI_DEVICE_1 ptr =
&mpt->mpt_dev_page1[tgt];
period = ptr->RequestedParameters;
period &= MPI_SCSIDEVPAGE1_RP_MIN_SYNC_PERIOD_MASK;
period >>= MPI_SCSIDEVPAGE1_RP_SHIFT_MIN_SYNC_PERIOD;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
CAMLOCK_2_MPTLOCK(mpt);
if (dval & DP_DISC_ENABLE) {
mpt->mpt_disc_enable |= (1 << tgt);
} else if (dval & DP_DISC_DISABL) {
mpt->mpt_disc_enable &= ~(1 << tgt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (dval & DP_TQING_ENABLE) {
mpt->mpt_tag_enable |= (1 << tgt);
} else if (dval & DP_TQING_DISABL) {
mpt->mpt_tag_enable &= ~(1 << tgt);
}
if (dval & DP_WIDTH) {
mpt_setwidth(mpt, tgt, 1);
}
if (dval & DP_SYNC) {
mpt_setsync(mpt, tgt, period, offset);
}
Fix some negotiation issues (like not being able to negotiate async)
2006-11-02 23:19:00 +00:00
if (dval == 0) {
MPTLOCK_2_CAMLOCK(mpt);
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
break;
}
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
"Set Settings[%d]: 0x%x period 0x%x offset %d\n", tgt,
dval, period , offset);
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
if (mpt_update_spi_config(mpt, tgt)) {
mpt_set_ccb_status(ccb, CAM_REQ_CMP_ERR);
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
} else {
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
case XPT_GET_TRAN_SETTINGS:
cts = &ccb->cts;
if (mpt->is_fc) {
struct ccb_trans_settings_fc *fc =
&cts->xport_specific.fc;
cts->protocol = PROTO_SCSI;
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cts->protocol_version = SCSI_REV_SPC;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cts->transport = XPORT_FC;
cts->transport_version = 0;
fc->valid = CTS_FC_VALID_SPEED;
fc->bitrate = 100000; /* XXX: Need for 2Gb/s */
/* XXX: need a port database for each target */
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
} else if (mpt->is_sas) {
struct ccb_trans_settings_sas *sas =
&cts->xport_specific.sas;
cts->protocol = PROTO_SCSI;
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cts->protocol_version = SCSI_REV_SPC2;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
cts->transport = XPORT_SAS;
cts->transport_version = 0;
sas->valid = CTS_SAS_VALID_SPEED;
sas->bitrate = 300000; /* XXX: Default 3Gbps */
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
} else if (mpt_get_spi_settings(mpt, cts) != 0) {
mpt_set_ccb_status(ccb, CAM_REQ_CMP_ERR);
break;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
case XPT_CALC_GEOMETRY:
{
struct ccb_calc_geometry *ccg;
ccg = &ccb->ccg;
if (ccg->block_size == 0) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_REQ_INVALID);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
}
mpt_calc_geometry(ccg, /*extended*/1);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
KASSERT(ccb->ccb_h.status, ("zero ccb sts at %d\n", __LINE__));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
}
case XPT_PATH_INQ: /* Path routing inquiry */
{
struct ccb_pathinq *cpi = &ccb->cpi;
cpi->version_num = 1;
cpi->target_sprt = 0;
cpi->hba_eng_cnt = 0;
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
cpi->max_target = mpt->mpt_max_devices - 1;
/*
* XXX: FC cards report MAX_DEVICES of 512- but we
* XXX: seem to hang when going higher than 255.
*/
if (cpi->max_target > 255)
cpi->max_target = 255;
VMWare ESX reports > 16 targets for the LSI-Logic U320 model it emulates. Then it crashes and burns when you probe that high.
2006-06-26 05:44:18 +00:00
/*
* XXX: VMware ESX reports > 16 devices and then dies
* XXX: when we probe.
*/
if (mpt->is_spi && cpi->max_target > 15)
cpi->max_target = 15;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cpi->max_lun = 7;
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
cpi->initiator_id = mpt->mpt_ini_id;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cpi->bus_id = cam_sim_bus(sim);
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
/*
* Actual speed for each device varies.
*
* The base speed is the speed of the underlying connection.
* This is strictly determined for SPI (async, narrow). If
* link is up for Fibre Channel, then speed can be gotten
* from that.
*/
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (mpt->is_fc) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cpi->hba_misc = PIM_NOBUSRESET;
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
cpi->base_transfer_speed =
mpt->mpt_fcport_speed * 100000;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cpi->hba_inquiry = PI_TAG_ABLE;
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cpi->transport = XPORT_FC;
cpi->transport_version = 0;
Do initial cut of SAS HBA support. These controllers (106X) seem to support automatically both SATA and SAS drives. The async SAS event handling we catch but ignore at present (so automagic attach/detach isn't hooked up yet). Do 64 bit PCI support- we can now work on systems with > 4GB of memory. Do large transfer support- we now can support up to reported chain depth, or the length of our request area. We simply allocate additional request elements when we would run out of room for chain lists. Tested on Ultra320, FC and SAS controllers on AMD64 and i386 platforms. There were no RAID cards available for me to regression test. The error recovery for this driver still is pretty bad.
2006-02-11 01:35:29 +00:00
} else if (mpt->is_sas) {
cpi->hba_misc = PIM_NOBUSRESET;
cpi->base_transfer_speed = 300000;
cpi->hba_inquiry = PI_TAG_ABLE;
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cpi->transport = XPORT_SAS;
cpi->transport_version = 0;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
} else {
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
cpi->hba_misc = PIM_SEQSCAN;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
cpi->base_transfer_speed = 3300;
cpi->hba_inquiry = PI_SDTR_ABLE|PI_TAG_ABLE|PI_WIDE_16;
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cpi->transport = XPORT_SPI;
cpi->transport_version = 2;
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
}
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
The first of 3 major steps to move the CAM layer forward to using the CAM_NEW_TRAN_CODE that has been in the tree for some years now. This first step consists solely of adding to or correcting CAM_NEW_TRAN_CODE pieces in the kernel source tree such that a both a GENERIC (at least on i386) and a LINT build with CAM_NEW_TRAN_CODE as an option will compile correctly and run (at least with some the h/w I have). After a short settle time, the other pieces (making CAM_NEW_TRAN_CODE the default and updating libcam and camcontrol) will be brought in. This will be an incompatible change in that the size of structures related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change in both size and content. However, basic system operation and basic system utilities work well enough with this change. Reviewed by: freebsd-scsi and specific stakeholders
2006-10-31 05:53:29 +00:00
cpi->protocol = PROTO_SCSI;
cpi->protocol_version = SCSI_REV_2;
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
/*
* We give our fake RAID passhtru bus a width that is MaxVolumes
* wide, restrict it to one lun and have it *not* be a bus
* that can have a SCSI bus reset.
*/
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
if (raid_passthru) {
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
cpi->max_target = mpt->ioc_page2->MaxPhysDisks - 1;
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
cpi->initiator_id = cpi->max_target + 1;
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
cpi->max_lun = 0;
Do some source && comment cleanup. Clean out the abortive start to homegrown, per-mpt, Domain Validation. This should really be done at a higher level. Use the PIM_SEQSCAN flag for U320- this seems to correct cases of being unable to consistently negotiate U320 in the cases where I'd seen this before. Between this and other recent checkins, this driver is pretty close to being ready for MFC. Reviewed by: scottl, ken, scsi@ MFC after: 1 week
2006-06-05 22:25:49 +00:00
cpi->hba_misc |= PIM_NOBUSRESET;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if ((mpt->role & MPT_ROLE_INITIATOR) == 0) {
cpi->hba_misc |= PIM_NOINITIATOR;
}
The poison pill of death: adding a target mode reply handler and target resources to a non-FC card killed us dead. Sorry for the breakage since last July 12.
2006-09-05 23:53:07 +00:00
if (mpt->is_fc && (mpt->role & MPT_ROLE_TARGET)) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
cpi->target_sprt =
PIT_PROCESSOR | PIT_DISCONNECT | PIT_TERM_IO;
} else {
cpi->target_sprt = 0;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
strncpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
strncpy(cpi->hba_vid, "LSI", HBA_IDLEN);
strncpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
cpi->unit_number = cam_sim_unit(sim);
cpi->ccb_h.status = CAM_REQ_CMP;
break;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
case XPT_EN_LUN: /* Enable LUN as a target */
{
int result;
CAMLOCK_2_MPTLOCK(mpt);
if (ccb->cel.enable)
result = mpt_enable_lun(mpt,
ccb->ccb_h.target_id, ccb->ccb_h.target_lun);
else
result = mpt_disable_lun(mpt,
ccb->ccb_h.target_id, ccb->ccb_h.target_lun);
MPTLOCK_2_CAMLOCK(mpt);
if (result == 0) {
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
} else {
mpt_set_ccb_status(ccb, CAM_REQ_CMP_ERR);
}
break;
}
case XPT_NOTIFY_ACK: /* recycle notify ack */
case XPT_IMMED_NOTIFY: /* Add Immediate Notify Resource */
case XPT_ACCEPT_TARGET_IO: /* Add Accept Target IO Resource */
{
tgt_resource_t *trtp;
lun_id_t lun = ccb->ccb_h.target_lun;
ccb->ccb_h.sim_priv.entries[0].field = 0;
ccb->ccb_h.sim_priv.entries[1].ptr = mpt;
ccb->ccb_h.flags = 0;
if (lun == CAM_LUN_WILDCARD) {
if (ccb->ccb_h.target_id != CAM_TARGET_WILDCARD) {
mpt_set_ccb_status(ccb, CAM_REQ_INVALID);
break;
}
trtp = &mpt->trt_wildcard;
} else if (lun >= MPT_MAX_LUNS) {
mpt_set_ccb_status(ccb, CAM_REQ_INVALID);
break;
} else {
trtp = &mpt->trt[lun];
}
CAMLOCK_2_MPTLOCK(mpt);
if (ccb->ccb_h.func_code == XPT_ACCEPT_TARGET_IO) {
mpt_lprt(mpt, MPT_PRT_DEBUG1,
"Put FREE ATIO %p lun %d\n", ccb, lun);
STAILQ_INSERT_TAIL(&trtp->atios, &ccb->ccb_h,
sim_links.stqe);
} else if (ccb->ccb_h.func_code == XPT_IMMED_NOTIFY) {
mpt_lprt(mpt, MPT_PRT_DEBUG1,
"Put FREE INOT lun %d\n", lun);
STAILQ_INSERT_TAIL(&trtp->inots, &ccb->ccb_h,
sim_links.stqe);
} else {
mpt_lprt(mpt, MPT_PRT_ALWAYS, "Got Notify ACK\n");
}
mpt_set_ccb_status(ccb, CAM_REQ_INPROG);
MPTLOCK_2_CAMLOCK(mpt);
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
return;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
case XPT_CONT_TARGET_IO:
CAMLOCK_2_MPTLOCK(mpt);
mpt_target_start_io(mpt, ccb);
MPTLOCK_2_CAMLOCK(mpt);
Major Fixes: Don't enable/disable I/O space except for SAS adapters. This fixes a problem with VMware 4.5 Workstation. Fix an egregious bug introduced to target mode so it actually will not panic when you first enable a lun. Minor fixes: Take more infor from port facts and configuration pages. MFC after: 1 week
2006-06-25 04:23:26 +00:00
return;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
default:
ccb->ccb_h.status = CAM_REQ_INVALID;
break;
}
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
xpt_done(ccb);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
static int
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_get_spi_settings(struct mpt_softc *mpt, struct ccb_trans_settings *cts)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
struct ccb_trans_settings_scsi *scsi = &cts->proto_specific.scsi;
struct ccb_trans_settings_spi *spi = &cts->xport_specific.spi;
target_id_t tgt;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
uint8_t dval, pval, oval;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
int rv;
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
if (xpt_path_sim(cts->ccb_h.path) == mpt->phydisk_sim) {
if (mpt_map_physdisk(mpt, (union ccb *)cts, &tgt)) {
return (-1);
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
}
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
} else {
tgt = cts->ccb_h.target_id;
Work in progress toward fixing IM checked in after having lost one set to a peninsula power failure last night. After this, I can see both submembers and the raid volumes again, but speed negotiation is still broken. Add a mpt_raid_free_mem function to centralize the resource reclaim and fixed a small memory leak. Remove restriction on number of targets for systems with IM enabled- you can have setups that have both IM volumes as well as other devices. Fix target id selection for passthru and nonpastrhu cases. Move complete command dumpt to MPT_PRT_DEBUG1 level so that just setting debug level gets mostly informative albeit less verbose dumping.
2006-05-27 17:26:57 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
* XXX: We aren't looking Port Page 2 BIOS settings here.
* XXX: For goal settings, we pick the max from port page 0
*
* For current settings we read the current settings out from
* device page 0 for that target.
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
*/
if (IS_CURRENT_SETTINGS(cts)) {
CONFIG_PAGE_SCSI_DEVICE_0 tmp;
dval = 0;
CAMLOCK_2_MPTLOCK(mpt);
tmp = mpt->mpt_dev_page0[tgt];
rv = mpt_read_cur_cfg_page(mpt, tgt, &tmp.Header,
sizeof(tmp), FALSE, 5000);
if (rv) {
MPTLOCK_2_CAMLOCK(mpt);
mpt_prt(mpt, "can't get tgt %d config page 0\n", tgt);
return (rv);
}
MPTLOCK_2_CAMLOCK(mpt);
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
dval |= (tmp.NegotiatedParameters & MPI_SCSIDEVPAGE0_NP_WIDE) ?
DP_WIDE : DP_NARROW;
dval |= (mpt->mpt_disc_enable & (1 << tgt)) ?
DP_DISC_ENABLE : DP_DISC_DISABL;
dval |= (mpt->mpt_tag_enable & (1 << tgt)) ?
DP_TQING_ENABLE : DP_TQING_DISABL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
oval = (tmp.NegotiatedParameters >> 16) & 0xff;
pval = (tmp.NegotiatedParameters >> 8) & 0xff;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt->mpt_dev_page0[tgt] = tmp;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
} else {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
* XXX: Just make theoretical maximum.
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
*/
Add ability to reset individual devices and fix SCSI speed negotiation. Reviewed by: mjacob (initial version)
2006-06-09 23:11:43 +00:00
dval = DP_WIDE|DP_DISC_ENABLE|DP_TQING_ENABLE;
oval = (mpt->mpt_port_page0.Capabilities >> 16) & 0xff;
pval = (mpt->mpt_port_page0.Capabilities >> 8) & 0xff;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
cts->protocol = PROTO_SCSI;
cts->protocol_version = SCSI_REV_2;
cts->transport = XPORT_SPI;
cts->transport_version = 2;
scsi->flags &= ~CTS_SCSI_FLAGS_TAG_ENB;
spi->flags &= ~CTS_SPI_FLAGS_DISC_ENB;
if (dval & DP_DISC_ENABLE) {
spi->flags |= CTS_SPI_FLAGS_DISC_ENB;
}
if (dval & DP_TQING_ENABLE) {
scsi->flags |= CTS_SCSI_FLAGS_TAG_ENB;
}
if (oval && pval) {
spi->sync_offset = oval;
spi->sync_period = pval;
spi->valid |= CTS_SPI_VALID_SYNC_OFFSET;
spi->valid |= CTS_SPI_VALID_SYNC_RATE;
}
spi->valid |= CTS_SPI_VALID_BUS_WIDTH;
if (dval & DP_WIDE) {
spi->bus_width = MSG_EXT_WDTR_BUS_16_BIT;
} else {
spi->bus_width = MSG_EXT_WDTR_BUS_8_BIT;
}
if (cts->ccb_h.target_lun != CAM_LUN_WILDCARD) {
scsi->valid = CTS_SCSI_VALID_TQ;
spi->valid |= CTS_SPI_VALID_DISC;
} else {
scsi->valid = 0;
}
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
"mpt_get_spi_settings[%d]: %s 0x%x period 0x%x offset %d\n", tgt,
IS_CURRENT_SETTINGS(cts)? "ACTIVE" : "NVRAM ", dval, pval, oval);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (0);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
static void
mpt_setwidth(struct mpt_softc *mpt, int tgt, int onoff)
{
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
PTR_CONFIG_PAGE_SCSI_DEVICE_1 ptr;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
ptr = &mpt->mpt_dev_page1[tgt];
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (onoff) {
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
ptr->RequestedParameters |= MPI_SCSIDEVPAGE1_RP_WIDE;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
} else {
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_WIDE;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
}
static void
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_setsync(struct mpt_softc *mpt, int tgt, int period, int offset)
{
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
PTR_CONFIG_PAGE_SCSI_DEVICE_1 ptr;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
+ Change some debug messages to MPT_PRT_NEGOTIATE level (so we can see the results of SPI negotiation w/o being overwhelmed with other crap). + For U320 devices, check against both Settings *and* DV flags before deciding whether we need to skip actual SPI settings for a device. + Go back to creating a 'physical disk' side of a raid/passthru bus that is limited to the number of maximum physical disks. Actually, this isn't probably *quite* right yet for one RAID volume, and if we ever end up with finding a device that supports more than one RAID volume (not likely), it probably won't quite be right either. The problem here is that the creating of this 'physical' passthru sim is just a cheap way to leverage off the CAM midlayer to do our negotiation for us on the subentities that make up a RAID volume. It almost causes more trouble than it is worth because we have to remember which side we're talking to in terms of forming commands and which target ids are real and so on. Bleah. + Skip trying to actually do SPI settings for the RAID volumes on the real side of the raid/passthru bus pair- this just confuses the issue. The underlying real physical devices will have the negotiation performed and the Raid volume will inherit the resultant settings. At the sime time, non-RAID devices can be on the same real bus, so *do* perform negotiations with them. + At the end of doing all of the settings twiddling, *ahem*, remember to go update the settings on the card itself (dunno how this got nuked). At this point, negotiations *seem* to be being done (again) correctly for both RAID volumes and their subentities. And they seem to be *mostly* now right for other non-RAID entities on the same bus (I ended up with 3 out of 8 other disks still at narror/async- haven't the slightest idea why yes). Finally, negotiations on a normal bus seem to work (again). There's still more work coming into this area, but we're in the final stretch.
2006-05-29 20:30:40 +00:00
ptr = &mpt->mpt_dev_page1[tgt];
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_MIN_SYNC_PERIOD_MASK;
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_MAX_SYNC_OFFSET_MASK;
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_DT;
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_QAS;
ptr->RequestedParameters &= ~MPI_SCSIDEVPAGE1_RP_IU;
Fix some negotiation issues (like not being able to negotiate async)
2006-11-02 23:19:00 +00:00
if (period == 0) {
return;
}
ptr->RequestedParameters |=
period << MPI_SCSIDEVPAGE1_RP_SHIFT_MIN_SYNC_PERIOD;
ptr->RequestedParameters |=
offset << MPI_SCSIDEVPAGE1_RP_SHIFT_MAX_SYNC_OFFSET;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
if (period < 0xa) {
ptr->RequestedParameters |= MPI_SCSIDEVPAGE1_RP_DT;
}
if (period < 0x9) {
ptr->RequestedParameters |= MPI_SCSIDEVPAGE1_RP_QAS;
ptr->RequestedParameters |= MPI_SCSIDEVPAGE1_RP_IU;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
static int
mpt_update_spi_config(struct mpt_softc *mpt, int tgt)
{
CONFIG_PAGE_SCSI_DEVICE_1 tmp;
int rv;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
mpt_lprt(mpt, MPT_PRT_NEGOTIATION,
"mpt_update_spi_config[%d].page1: Requested Params 0x%08x\n",
tgt, mpt->mpt_dev_page1[tgt].RequestedParameters);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
tmp = mpt->mpt_dev_page1[tgt];
rv = mpt_write_cur_cfg_page(mpt, tgt,
&tmp.Header, sizeof(tmp), FALSE, 5000);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
if (rv) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt, "mpt_update_spi_config: write cur page failed\n");
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (-1);
}
return (0);
}
static void
mpt_calc_geometry(struct ccb_calc_geometry *ccg, int extended)
{
#if __FreeBSD_version >= 500000
cam_calc_geometry(ccg, extended);
#else
uint32_t size_mb;
uint32_t secs_per_cylinder;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (ccg->block_size == 0) {
ccg->ccb_h.status = CAM_REQ_INVALID;
return;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
size_mb = ccg->volume_size / ((1024L * 1024L) / ccg->block_size);
if (size_mb > 1024 && extended) {
ccg->heads = 255;
ccg->secs_per_track = 63;
} else {
ccg->heads = 64;
ccg->secs_per_track = 32;
}
secs_per_cylinder = ccg->heads * ccg->secs_per_track;
ccg->cylinders = ccg->volume_size / secs_per_cylinder;
ccg->ccb_h.status = CAM_REQ_CMP;
#endif
}
/****************************** Timeout Recovery ******************************/
static int
mpt_spawn_recovery_thread(struct mpt_softc *mpt)
{
int error;
error = mpt_kthread_create(mpt_recovery_thread, mpt,
&mpt->recovery_thread, /*flags*/0,
/*altstack*/0, "mpt_recovery%d", mpt->unit);
return (error);
}
static void
mpt_terminate_recovery_thread(struct mpt_softc *mpt)
{
if (mpt->recovery_thread == NULL) {
return;
}
mpt->shutdwn_recovery = 1;
wakeup(mpt);
/*
* Sleep on a slightly different location
* for this interlock just for added safety.
*/
mpt_sleep(mpt, &mpt->recovery_thread, PUSER, "thtrm", 0);
}
static void
mpt_recovery_thread(void *arg)
{
struct mpt_softc *mpt;
#if __FreeBSD_version >= 500000
mtx_lock(&Giant);
#endif
mpt = (struct mpt_softc *)arg;
MPT_LOCK(mpt);
for (;;) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (TAILQ_EMPTY(&mpt->request_timeout_list) != 0) {
if (mpt->shutdwn_recovery == 0) {
mpt_sleep(mpt, mpt, PUSER, "idle", 0);
}
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (mpt->shutdwn_recovery != 0) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
break;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_recover_commands(mpt);
}
mpt->recovery_thread = NULL;
wakeup(&mpt->recovery_thread);
MPT_UNLOCK(mpt);
#if __FreeBSD_version >= 500000
mtx_unlock(&Giant);
#endif
kthread_exit(0);
}
static int
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_scsi_send_tmf(struct mpt_softc *mpt, u_int type, u_int flags,
u_int channel, u_int target, u_int lun, u_int abort_ctx, int sleep_ok)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
MSG_SCSI_TASK_MGMT *tmf_req;
int error;
/*
* Wait for any current TMF request to complete.
* We're only allowed to issue one TMF at a time.
*/
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
error = mpt_wait_req(mpt, mpt->tmf_req, REQ_STATE_FREE, REQ_STATE_FREE,
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
sleep_ok, MPT_TMF_MAX_TIMEOUT);
if (error != 0) {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_reset(mpt, TRUE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
return (ETIMEDOUT);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_assign_serno(mpt, mpt->tmf_req);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt->tmf_req->state = REQ_STATE_ALLOCATED|REQ_STATE_QUEUED;
tmf_req = (MSG_SCSI_TASK_MGMT *)mpt->tmf_req->req_vbuf;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(tmf_req, 0, sizeof(*tmf_req));
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
tmf_req->TargetID = target;
tmf_req->Bus = channel;
tmf_req->ChainOffset = 0;
tmf_req->Function = MPI_FUNCTION_SCSI_TASK_MGMT;
tmf_req->Reserved = 0;
tmf_req->TaskType = type;
tmf_req->Reserved1 = 0;
tmf_req->MsgFlags = flags;
tmf_req->MsgContext =
htole32(mpt->tmf_req->index | scsi_tmf_handler_id);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(&tmf_req->LUN, 0,
sizeof(tmf_req->LUN) + sizeof(tmf_req->Reserved2));
if (lun > 256) {
tmf_req->LUN[0] = 0x40 | ((lun >> 8) & 0x3f);
tmf_req->LUN[1] = lun & 0xff;
} else {
tmf_req->LUN[1] = lun;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
tmf_req->TaskMsgContext = abort_ctx;
If we actually succeed in the Task Management Function where we are aborting timed out commands, pull the request off the TAILQ.
2006-03-17 04:54:06 +00:00
mpt_lprt(mpt, MPT_PRT_INFO,
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
"Issuing TMF %p:%u with MsgContext of 0x%x\n", mpt->tmf_req,
mpt->tmf_req->serno, tmf_req->MsgContext);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (mpt->verbose > MPT_PRT_DEBUG) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
mpt_print_request(tmf_req);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
KASSERT(mpt_req_on_pending_list(mpt, mpt->tmf_req) == 0,
("mpt_scsi_send_tmf: tmf_req already on pending list"));
TAILQ_INSERT_HEAD(&mpt->request_pending_list, mpt->tmf_req, links);
error = mpt_send_handshake_cmd(mpt, sizeof(*tmf_req), tmf_req);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (error != MPT_OK) {
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_reset(mpt, TRUE);
}
return (error);
}
/*
* When a command times out, it is placed on the requeust_timeout_list
* and we wake our recovery thread. The MPT-Fusion architecture supports
* only a single TMF operation at a time, so we serially abort/bdr, etc,
* the timedout transactions. The next TMF is issued either by the
* completion handler of the current TMF waking our recovery thread,
* or the TMF timeout handler causing a hard reset sequence.
*/
static void
mpt_recover_commands(struct mpt_softc *mpt)
{
request_t *req;
union ccb *ccb;
int error;
if (TAILQ_EMPTY(&mpt->request_timeout_list) != 0) {
/*
* No work to do- leave.
*/
mpt_prt(mpt, "mpt_recover_commands: no requests.\n");
return;
}
/*
* Flush any commands whose completion coincides with their timeout.
*/
mpt_intr(mpt);
if (TAILQ_EMPTY(&mpt->request_timeout_list) != 0) {
/*
* The timedout commands have already
* completed. This typically means
* that either the timeout value was on
* the hairy edge of what the device
* requires or - more likely - interrupts
* are not happening.
*/
mpt_prt(mpt, "Timedout requests already complete. "
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
"Interrupts may not be functioning.\n");
mpt_enable_ints(mpt);
return;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
/*
* We have no visibility into the current state of the
* controller, so attempt to abort the commands in the
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* order they timed-out. For initiator commands, we
* depend on the reply handler pulling requests off
* the timeout list.
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
*/
while ((req = TAILQ_FIRST(&mpt->request_timeout_list)) != NULL) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
uint16_t status;
uint8_t response;
MSG_REQUEST_HEADER *hdrp = req->req_vbuf;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt, "attempting to abort req %p:%u function %x\n",
req, req->serno, hdrp->Function);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
ccb = req->ccb;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (ccb == NULL) {
mpt_prt(mpt, "null ccb in timed out request. "
"Resetting Controller.\n");
mpt_reset(mpt, TRUE);
continue;
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_set_ccb_status(ccb, CAM_CMD_TIMEOUT);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* Check to see if this is not an initiator command and
* deal with it differently if it is.
*/
switch (hdrp->Function) {
case MPI_FUNCTION_SCSI_IO_REQUEST:
Get most of the way back to having Integrated Mirroring work again- the addition of target mode support broke it massively.
2006-05-26 05:54:21 +00:00
case MPI_FUNCTION_RAID_SCSI_IO_PASSTHROUGH:
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
break;
default:
/*
* XXX: FIX ME: need to abort target assists...
*/
mpt_prt(mpt, "just putting it back on the pend q\n");
TAILQ_REMOVE(&mpt->request_timeout_list, req, links);
TAILQ_INSERT_HEAD(&mpt->request_pending_list, req,
links);
continue;
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
error = mpt_scsi_send_tmf(mpt,
MPI_SCSITASKMGMT_TASKTYPE_ABORT_TASK,
0, 0, ccb->ccb_h.target_id, ccb->ccb_h.target_lun,
htole32(req->index | scsi_io_handler_id), TRUE);
if (error != 0) {
/*
* mpt_scsi_send_tmf hard resets on failure, so no
* need to do so here. Our queue should be emptied
* by the hard reset.
*/
continue;
}
error = mpt_wait_req(mpt, mpt->tmf_req, REQ_STATE_DONE,
REQ_STATE_DONE, TRUE, 500);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
status = mpt->tmf_req->IOCStatus;
response = mpt->tmf_req->ResponseCode;
mpt->tmf_req->state = REQ_STATE_FREE;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (error != 0) {
/*
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* If we've errored out,, reset the controller.
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
*/
mpt_prt(mpt, "mpt_recover_commands: abort timed-out. "
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"Resetting controller\n");
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_reset(mpt, TRUE);
continue;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if ((status & MPI_IOCSTATUS_MASK) != MPI_IOCSTATUS_SUCCESS) {
mpt_prt(mpt, "mpt_recover_commands: IOC Status 0x%x. "
"Resetting controller.\n", status);
mpt_reset(mpt, TRUE);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
continue;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (response != MPI_SCSITASKMGMT_RSP_TM_SUCCEEDED &&
response != MPI_SCSITASKMGMT_RSP_TM_COMPLETE) {
mpt_prt(mpt, "mpt_recover_commands: TMF Response 0x%x. "
"Resetting controller.\n", response);
mpt_reset(mpt, TRUE);
continue;
}
mpt_prt(mpt, "abort of req %p:%u completed\n", req, req->serno);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/************************ Target Mode Support ****************************/
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
static void
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_fc_post_els(struct mpt_softc *mpt, request_t *req, int ioindex)
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
{
MSG_LINK_SERVICE_BUFFER_POST_REQUEST *fc;
PTR_SGE_TRANSACTION32 tep;
PTR_SGE_SIMPLE32 se;
bus_addr_t paddr;
paddr = req->req_pbuf;
paddr += MPT_RQSL(mpt);
fc = req->req_vbuf;
memset(fc, 0, MPT_REQUEST_AREA);
fc->BufferCount = 1;
fc->Function = MPI_FUNCTION_FC_LINK_SRVC_BUF_POST;
fc->MsgContext = htole32(req->index | fc_els_handler_id);
/*
* Okay, set up ELS buffer pointers. ELS buffer pointers
* consist of a TE SGL element (with details length of zero)
* followe by a SIMPLE SGL element which holds the address
* of the buffer.
*/
tep = (PTR_SGE_TRANSACTION32) &fc->SGL;
tep->ContextSize = 4;
tep->Flags = 0;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
tep->TransactionContext[0] = htole32(ioindex);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
se = (PTR_SGE_SIMPLE32) &tep->TransactionDetails[0];
se->FlagsLength =
MPI_SGE_FLAGS_HOST_TO_IOC |
MPI_SGE_FLAGS_SIMPLE_ELEMENT |
MPI_SGE_FLAGS_LAST_ELEMENT |
MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
se->FlagsLength <<= MPI_SGE_FLAGS_SHIFT;
se->FlagsLength |= (MPT_NRFM(mpt) - MPT_RQSL(mpt));
se->Address = (uint32_t) paddr;
Remove MPT_PRT_INVARIANT- it was a silly idea.
2006-05-04 02:34:18 +00:00
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"add ELS index %d ioindex %d for %p:%u\n",
req->index, ioindex, req, req->serno);
KASSERT(((req->state & REQ_STATE_LOCKED) != 0),
("mpt_fc_post_els: request not locked"));
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_send_cmd(mpt, req);
}
static void
mpt_post_target_command(struct mpt_softc *mpt, request_t *req, int ioindex)
{
PTR_MSG_TARGET_CMD_BUFFER_POST_REQUEST fc;
PTR_CMD_BUFFER_DESCRIPTOR cb;
bus_addr_t paddr;
paddr = req->req_pbuf;
paddr += MPT_RQSL(mpt);
memset(req->req_vbuf, 0, MPT_REQUEST_AREA);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
MPT_TGT_STATE(mpt, req)->state = TGT_STATE_LOADING;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
fc = req->req_vbuf;
fc->BufferCount = 1;
fc->Function = MPI_FUNCTION_TARGET_CMD_BUFFER_POST;
fc->MsgContext = htole32(req->index | mpt->scsi_tgt_handler_id);
cb = &fc->Buffer[0];
cb->IoIndex = htole16(ioindex);
cb->u.PhysicalAddress32 = (U32) paddr;
mpt_check_doorbell(mpt);
mpt_send_cmd(mpt, req);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
static int
mpt_add_els_buffers(struct mpt_softc *mpt)
{
int i;
if (mpt->is_fc == 0) {
return (TRUE);
}
if (mpt->els_cmds_allocated) {
return (TRUE);
}
mpt->els_cmd_ptrs = malloc(MPT_MAX_ELS * sizeof (request_t *),
M_DEVBUF, M_NOWAIT | M_ZERO);
if (mpt->els_cmd_ptrs == NULL) {
return (FALSE);
}
/*
* Feed the chip some ELS buffer resources
*/
for (i = 0; i < MPT_MAX_ELS; i++) {
request_t *req = mpt_get_request(mpt, FALSE);
if (req == NULL) {
break;
}
req->state |= REQ_STATE_LOCKED;
mpt->els_cmd_ptrs[i] = req;
mpt_fc_post_els(mpt, req, i);
}
if (i == 0) {
mpt_prt(mpt, "unable to add ELS buffer resources\n");
free(mpt->els_cmd_ptrs, M_DEVBUF);
mpt->els_cmd_ptrs = NULL;
return (FALSE);
}
if (i != MPT_MAX_ELS) {
mpt_lprt(mpt, MPT_PRT_INFO,
"only added %d of %d ELS buffers\n", i, MPT_MAX_ELS);
}
mpt->els_cmds_allocated = i;
return(TRUE);
}
static int
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_add_target_commands(struct mpt_softc *mpt)
{
int i, max;
if (mpt->tgt_cmd_ptrs) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
return (TRUE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
max = MPT_MAX_REQUESTS(mpt) >> 1;
if (max > mpt->mpt_max_tgtcmds) {
max = mpt->mpt_max_tgtcmds;
}
mpt->tgt_cmd_ptrs =
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
malloc(max * sizeof (request_t *), M_DEVBUF, M_NOWAIT | M_ZERO);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (mpt->tgt_cmd_ptrs == NULL) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_prt(mpt,
"mpt_add_target_commands: could not allocate cmd ptrs\n");
return (FALSE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
for (i = 0; i < max; i++) {
request_t *req;
req = mpt_get_request(mpt, FALSE);
if (req == NULL) {
break;
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
req->state |= REQ_STATE_LOCKED;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt->tgt_cmd_ptrs[i] = req;
mpt_post_target_command(mpt, req, i);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
if (i == 0) {
mpt_lprt(mpt, MPT_PRT_ERROR, "could not add any target bufs\n");
free(mpt->tgt_cmd_ptrs, M_DEVBUF);
mpt->tgt_cmd_ptrs = NULL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
return (FALSE);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt->tgt_cmds_allocated = i;
if (i < max) {
mpt_lprt(mpt, MPT_PRT_INFO,
"added %d of %d target bufs\n", i, max);
}
return (i);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
static int
mpt_enable_lun(struct mpt_softc *mpt, target_id_t tgt, lun_id_t lun)
{
if (tgt == CAM_TARGET_WILDCARD && lun == CAM_LUN_WILDCARD) {
mpt->twildcard = 1;
} else if (lun >= MPT_MAX_LUNS) {
return (EINVAL);
} else if (tgt != CAM_TARGET_WILDCARD && tgt != 0) {
return (EINVAL);
}
if (mpt->tenabled == 0) {
if (mpt->is_fc) {
(void) mpt_fc_reset_link(mpt, 0);
}
mpt->tenabled = 1;
}
if (lun == CAM_LUN_WILDCARD) {
mpt->trt_wildcard.enabled = 1;
} else {
mpt->trt[lun].enabled = 1;
}
return (0);
}
static int
mpt_disable_lun(struct mpt_softc *mpt, target_id_t tgt, lun_id_t lun)
{
int i;
if (tgt == CAM_TARGET_WILDCARD && lun == CAM_LUN_WILDCARD) {
mpt->twildcard = 0;
} else if (lun >= MPT_MAX_LUNS) {
return (EINVAL);
} else if (tgt != CAM_TARGET_WILDCARD && tgt != 0) {
return (EINVAL);
}
if (lun == CAM_LUN_WILDCARD) {
mpt->trt_wildcard.enabled = 0;
} else {
mpt->trt[lun].enabled = 0;
}
for (i = 0; i < MPT_MAX_LUNS; i++) {
if (mpt->trt[lun].enabled) {
break;
}
}
if (i == MPT_MAX_LUNS && mpt->twildcard == 0) {
if (mpt->is_fc) {
(void) mpt_fc_reset_link(mpt, 0);
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt->tenabled = 0;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
return (0);
}
/*
* Called with MPT lock held
*/
static void
mpt_target_start_io(struct mpt_softc *mpt, union ccb *ccb)
{
struct ccb_scsiio *csio = &ccb->csio;
request_t *cmd_req = MPT_TAG_2_REQ(mpt, csio->tag_id);
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, cmd_req);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
switch (tgt->state) {
case TGT_STATE_IN_CAM:
break;
case TGT_STATE_MOVING_DATA:
mpt_set_ccb_status(ccb, CAM_REQUEUE_REQ);
xpt_freeze_simq(mpt->sim, 1);
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
tgt->ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
return;
default:
mpt_prt(mpt, "ccb %p flags 0x%x tag 0x%08x had bad request "
"starting I/O\n", ccb, csio->ccb_h.flags, csio->tag_id);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_tgt_dump_req_state(mpt, cmd_req);
mpt_set_ccb_status(ccb, CAM_REQ_CMP_ERR);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
return;
}
if (csio->dxfer_len) {
bus_dmamap_callback_t *cb;
PTR_MSG_TARGET_ASSIST_REQUEST ta;
request_t *req;
KASSERT((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE,
("dxfer_len %u but direction is NONE\n", csio->dxfer_len));
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if ((req = mpt_get_request(mpt, FALSE)) == NULL) {
if (mpt->outofbeer == 0) {
mpt->outofbeer = 1;
xpt_freeze_simq(mpt->sim, 1);
mpt_lprt(mpt, MPT_PRT_DEBUG, "FREEZEQ\n");
}
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
mpt_set_ccb_status(ccb, CAM_REQUEUE_REQ);
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
return;
}
ccb->ccb_h.status = CAM_SIM_QUEUED | CAM_REQ_INPROG;
if (sizeof (bus_addr_t) > 4) {
cb = mpt_execute_req_a64;
} else {
cb = mpt_execute_req;
}
req->ccb = ccb;
ccb->ccb_h.ccb_req_ptr = req;
/*
* Record the currently active ccb and the
* request for it in our target state area.
*/
tgt->ccb = ccb;
tgt->req = req;
memset(req->req_vbuf, 0, MPT_RQSL(mpt));
ta = req->req_vbuf;
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
if (mpt->is_sas) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
PTR_MPI_TARGET_SSP_CMD_BUFFER ssp =
cmd_req->req_vbuf;
ta->QueueTag = ssp->InitiatorTag;
More checkpointing on the way toward really (finally) fixing speed negotiation. Also fix the mpt_execute_req function to actually match mpt_execute_req_a64. This may explain why i386 users were having more grief.
2006-06-02 18:50:39 +00:00
} else if (mpt->is_spi) {
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER sp =
cmd_req->req_vbuf;
ta->QueueTag = sp->Tag;
}
ta->Function = MPI_FUNCTION_TARGET_ASSIST;
ta->MsgContext = htole32(req->index | mpt->scsi_tgt_handler_id);
ta->ReplyWord = htole32(tgt->reply_desc);
if (csio->ccb_h.target_lun > 256) {
ta->LUN[0] =
0x40 | ((csio->ccb_h.target_lun >> 8) & 0x3f);
ta->LUN[1] = csio->ccb_h.target_lun & 0xff;
} else {
ta->LUN[1] = csio->ccb_h.target_lun;
}
ta->RelativeOffset = tgt->bytes_xfered;
ta->DataLength = ccb->csio.dxfer_len;
if (ta->DataLength > tgt->resid) {
ta->DataLength = tgt->resid;
}
/*
* XXX Should be done after data transfer completes?
*/
tgt->resid -= csio->dxfer_len;
tgt->bytes_xfered += csio->dxfer_len;
if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_IN) {
ta->TargetAssistFlags |=
TARGET_ASSIST_FLAGS_DATA_DIRECTION;
}
#ifdef WE_TRUST_AUTO_GOOD_STATUS
if ((ccb->ccb_h.flags & CAM_SEND_STATUS) &&
csio->scsi_status == SCSI_STATUS_OK && tgt->resid == 0) {
ta->TargetAssistFlags |=
TARGET_ASSIST_FLAGS_AUTO_STATUS;
}
#endif
tgt->state = TGT_STATE_SETTING_UP_FOR_DATA;
mpt_lprt(mpt, MPT_PRT_DEBUG,
"DATA_CCB %p tag %x %u bytes %u resid flg %x req %p:%u "
"nxtstate=%d\n", csio, csio->tag_id, csio->dxfer_len,
tgt->resid, ccb->ccb_h.flags, req, req->serno, tgt->state);
MPTLOCK_2_CAMLOCK(mpt);
if ((ccb->ccb_h.flags & CAM_SCATTER_VALID) == 0) {
if ((ccb->ccb_h.flags & CAM_DATA_PHYS) == 0) {
int error;
int s = splsoftvm();
error = bus_dmamap_load(mpt->buffer_dmat,
req->dmap, csio->data_ptr, csio->dxfer_len,
cb, req, 0);
splx(s);
if (error == EINPROGRESS) {
xpt_freeze_simq(mpt->sim, 1);
ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
}
} else {
/*
* We have been given a pointer to single
* physical buffer.
*/
struct bus_dma_segment seg;
seg.ds_addr = (bus_addr_t)
(vm_offset_t)csio->data_ptr;
seg.ds_len = csio->dxfer_len;
(*cb)(req, &seg, 1, 0);
}
} else {
/*
* We have been given a list of addresses.
* This case could be easily supported but they are not
* currently generated by the CAM subsystem so there
* is no point in wasting the time right now.
*/
struct bus_dma_segment *sgs;
if ((ccb->ccb_h.flags & CAM_SG_LIST_PHYS) == 0) {
(*cb)(req, NULL, 0, EFAULT);
} else {
/* Just use the segments provided */
sgs = (struct bus_dma_segment *)csio->data_ptr;
(*cb)(req, sgs, csio->sglist_cnt, 0);
}
}
CAMLOCK_2_MPTLOCK(mpt);
} else {
uint8_t *sp = NULL, sense[MPT_SENSE_SIZE];
/*
* XXX: I don't know why this seems to happen, but
* XXX: completing the CCB seems to make things happy.
* XXX: This seems to happen if the initiator requests
* XXX: enough data that we have to do multiple CTIOs.
*/
if ((ccb->ccb_h.flags & CAM_SEND_STATUS) == 0) {
mpt_lprt(mpt, MPT_PRT_DEBUG,
"Meaningless STATUS CCB (%p): flags %x status %x "
"resid %d bytes_xfered %u\n", ccb, ccb->ccb_h.flags,
ccb->ccb_h.status, tgt->resid, tgt->bytes_xfered);
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
return;
}
if (ccb->ccb_h.flags & CAM_SEND_SENSE) {
sp = sense;
memcpy(sp, &csio->sense_data,
min(csio->sense_len, MPT_SENSE_SIZE));
}
mpt_scsi_tgt_status(mpt, ccb, cmd_req, csio->scsi_status, sp);
}
}
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
static void
mpt_scsi_tgt_local(struct mpt_softc *mpt, request_t *cmd_req,
uint32_t lun, int send, uint8_t *data, size_t length)
{
mpt_tgt_state_t *tgt;
PTR_MSG_TARGET_ASSIST_REQUEST ta;
SGE_SIMPLE32 *se;
uint32_t flags;
uint8_t *dptr;
bus_addr_t pptr;
request_t *req;
if (length == 0) {
mpt_scsi_tgt_status(mpt, NULL, cmd_req, 0, NULL);
return;
}
tgt = MPT_TGT_STATE(mpt, cmd_req);
if ((req = mpt_get_request(mpt, FALSE)) == NULL) {
mpt_prt(mpt, "out of resources- dropping local response\n");
return;
}
tgt->is_local = 1;
memset(req->req_vbuf, 0, MPT_RQSL(mpt));
ta = req->req_vbuf;
if (mpt->is_sas) {
PTR_MPI_TARGET_SSP_CMD_BUFFER ssp = cmd_req->req_vbuf;
ta->QueueTag = ssp->InitiatorTag;
} else if (mpt->is_spi) {
PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER sp = cmd_req->req_vbuf;
ta->QueueTag = sp->Tag;
}
ta->Function = MPI_FUNCTION_TARGET_ASSIST;
ta->MsgContext = htole32(req->index | mpt->scsi_tgt_handler_id);
ta->ReplyWord = htole32(tgt->reply_desc);
if (lun > 256) {
ta->LUN[0] = 0x40 | ((lun >> 8) & 0x3f);
ta->LUN[1] = lun & 0xff;
} else {
ta->LUN[1] = lun;
}
ta->RelativeOffset = 0;
ta->DataLength = length;
dptr = req->req_vbuf;
dptr += MPT_RQSL(mpt);
pptr = req->req_pbuf;
pptr += MPT_RQSL(mpt);
memcpy(dptr, data, min(length, MPT_RQSL(mpt)));
se = (SGE_SIMPLE32 *) &ta->SGL[0];
memset(se, 0,sizeof (*se));
flags = MPI_SGE_FLAGS_SIMPLE_ELEMENT;
if (send) {
ta->TargetAssistFlags |= TARGET_ASSIST_FLAGS_DATA_DIRECTION;
flags |= MPI_SGE_FLAGS_HOST_TO_IOC;
}
se->Address = pptr;
MPI_pSGE_SET_LENGTH(se, length);
flags |= MPI_SGE_FLAGS_LAST_ELEMENT;
flags |= MPI_SGE_FLAGS_END_OF_LIST | MPI_SGE_FLAGS_END_OF_BUFFER;
MPI_pSGE_SET_FLAGS(se, flags);
tgt->ccb = NULL;
tgt->req = req;
tgt->resid = 0;
tgt->bytes_xfered = length;
#ifdef WE_TRUST_AUTO_GOOD_STATUS
tgt->state = TGT_STATE_MOVING_DATA_AND_STATUS;
#else
tgt->state = TGT_STATE_MOVING_DATA;
#endif
mpt_send_cmd(mpt, req);
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Abort queued up CCBs
*/
static cam_status
mpt_abort_target_ccb(struct mpt_softc *mpt, union ccb *ccb)
{
struct mpt_hdr_stailq *lp;
struct ccb_hdr *srch;
int found = 0;
union ccb *accb = ccb->cab.abort_ccb;
tgt_resource_t *trtp;
mpt_lprt(mpt, MPT_PRT_DEBUG, "aborting ccb %p\n", accb);
if (ccb->ccb_h.target_lun == CAM_LUN_WILDCARD) {
trtp = &mpt->trt_wildcard;
} else {
trtp = &mpt->trt[ccb->ccb_h.target_lun];
}
if (accb->ccb_h.func_code == XPT_ACCEPT_TARGET_IO) {
lp = &trtp->atios;
} else if (accb->ccb_h.func_code == XPT_IMMED_NOTIFY) {
lp = &trtp->inots;
} else {
return (CAM_REQ_INVALID);
}
STAILQ_FOREACH(srch, lp, sim_links.stqe) {
if (srch == &accb->ccb_h) {
found = 1;
STAILQ_REMOVE(lp, srch, ccb_hdr, sim_links.stqe);
break;
}
}
if (found) {
accb->ccb_h.status = CAM_REQ_ABORTED;
xpt_done(accb);
return (CAM_REQ_CMP);
}
mpt_prt(mpt, "mpt_abort_tgt_ccb: CCB %p not found\n", ccb);
return (CAM_PATH_INVALID);
}
/*
* Ask the MPT to abort the current target command
*/
static int
mpt_abort_target_cmd(struct mpt_softc *mpt, request_t *cmd_req)
{
int error;
request_t *req;
PTR_MSG_TARGET_MODE_ABORT abtp;
req = mpt_get_request(mpt, FALSE);
if (req == NULL) {
return (-1);
}
abtp = req->req_vbuf;
memset(abtp, 0, sizeof (*abtp));
abtp->MsgContext = htole32(req->index | mpt->scsi_tgt_handler_id);
abtp->AbortType = TARGET_MODE_ABORT_TYPE_EXACT_IO;
abtp->Function = MPI_FUNCTION_TARGET_MODE_ABORT;
abtp->ReplyWord = htole32(MPT_TGT_STATE(mpt, cmd_req)->reply_desc);
error = 0;
if (mpt->is_fc || mpt->is_sas) {
mpt_send_cmd(mpt, req);
} else {
error = mpt_send_handshake_cmd(mpt, sizeof(*req), req);
}
return (error);
}
/*
* WE_TRUST_AUTO_GOOD_STATUS- I've found that setting
* TARGET_STATUS_SEND_FLAGS_AUTO_GOOD_STATUS leads the
* FC929 to set bogus FC_RSP fields (nonzero residuals
* but w/o RESID fields set). This causes QLogic initiators
* to think maybe that a frame was lost.
*
* WE_CAN_USE_AUTO_REPOST- we can't use AUTO_REPOST because
* we use allocated requests to do TARGET_ASSIST and we
* need to know when to release them.
*/
static void
mpt_scsi_tgt_status(struct mpt_softc *mpt, union ccb *ccb, request_t *cmd_req,
uint8_t status, uint8_t const *sense_data)
{
uint8_t *cmd_vbuf;
mpt_tgt_state_t *tgt;
PTR_MSG_TARGET_STATUS_SEND_REQUEST tp;
request_t *req;
bus_addr_t paddr;
int resplen = 0;
cmd_vbuf = cmd_req->req_vbuf;
cmd_vbuf += MPT_RQSL(mpt);
tgt = MPT_TGT_STATE(mpt, cmd_req);
if ((req = mpt_get_request(mpt, FALSE)) == NULL) {
if (mpt->outofbeer == 0) {
mpt->outofbeer = 1;
xpt_freeze_simq(mpt->sim, 1);
mpt_lprt(mpt, MPT_PRT_DEBUG, "FREEZEQ\n");
}
if (ccb) {
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
mpt_set_ccb_status(ccb, CAM_REQUEUE_REQ);
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
} else {
mpt_prt(mpt,
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
"could not allocate status request- dropping\n");
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
return;
}
req->ccb = ccb;
if (ccb) {
ccb->ccb_h.ccb_mpt_ptr = mpt;
ccb->ccb_h.ccb_req_ptr = req;
}
/*
* Record the currently active ccb, if any, and the
* request for it in our target state area.
*/
tgt->ccb = ccb;
tgt->req = req;
tgt->state = TGT_STATE_SENDING_STATUS;
tp = req->req_vbuf;
paddr = req->req_pbuf;
paddr += MPT_RQSL(mpt);
memset(tp, 0, sizeof (*tp));
tp->Function = MPI_FUNCTION_TARGET_STATUS_SEND;
if (mpt->is_fc) {
PTR_MPI_TARGET_FCP_CMD_BUFFER fc =
(PTR_MPI_TARGET_FCP_CMD_BUFFER) cmd_vbuf;
uint8_t *sts_vbuf;
uint32_t *rsp;
sts_vbuf = req->req_vbuf;
sts_vbuf += MPT_RQSL(mpt);
rsp = (uint32_t *) sts_vbuf;
memcpy(tp->LUN, fc->FcpLun, sizeof (tp->LUN));
/*
* The MPI_TARGET_FCP_RSP_BUFFER define is unfortunate.
* It has to be big-endian in memory and is organized
* in 32 bit words, which are much easier to deal with
* as words which are swizzled as needed.
*
* All we're filling here is the FC_RSP payload.
* We may just have the chip synthesize it if
* we have no residual and an OK status.
*
*/
memset(rsp, 0, sizeof (MPI_TARGET_FCP_RSP_BUFFER));
rsp[2] = status;
if (tgt->resid) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rsp[2] |= 0x800; /* XXXX NEED MNEMONIC!!!! */
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
rsp[3] = htobe32(tgt->resid);
#ifdef WE_TRUST_AUTO_GOOD_STATUS
resplen = sizeof (MPI_TARGET_FCP_RSP_BUFFER);
#endif
}
if (status == SCSI_STATUS_CHECK_COND) {
int i;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
rsp[2] |= 0x200; /* XXXX NEED MNEMONIC!!!! */
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
rsp[4] = htobe32(MPT_SENSE_SIZE);
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
if (sense_data) {
memcpy(&rsp[8], sense_data, MPT_SENSE_SIZE);
} else {
mpt_prt(mpt, "mpt_scsi_tgt_status: CHECK CONDI"
"TION but no sense data?\n");
memset(&rsp, 0, MPT_SENSE_SIZE);
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
for (i = 8; i < (8 + (MPT_SENSE_SIZE >> 2)); i++) {
rsp[i] = htobe32(rsp[i]);
}
#ifdef WE_TRUST_AUTO_GOOD_STATUS
resplen = sizeof (MPI_TARGET_FCP_RSP_BUFFER);
#endif
}
#ifndef WE_TRUST_AUTO_GOOD_STATUS
resplen = sizeof (MPI_TARGET_FCP_RSP_BUFFER);
#endif
rsp[2] = htobe32(rsp[2]);
} else if (mpt->is_sas) {
PTR_MPI_TARGET_SSP_CMD_BUFFER ssp =
(PTR_MPI_TARGET_SSP_CMD_BUFFER) cmd_vbuf;
memcpy(tp->LUN, ssp->LogicalUnitNumber, sizeof (tp->LUN));
} else {
PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER sp =
(PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER) cmd_vbuf;
tp->StatusCode = status;
tp->QueueTag = htole16(sp->Tag);
memcpy(tp->LUN, sp->LogicalUnitNumber, sizeof (tp->LUN));
}
tp->ReplyWord = htole32(tgt->reply_desc);
tp->MsgContext = htole32(req->index | mpt->scsi_tgt_handler_id);
#ifdef WE_CAN_USE_AUTO_REPOST
tp->MsgFlags = TARGET_STATUS_SEND_FLAGS_REPOST_CMD_BUFFER;
#endif
if (status == SCSI_STATUS_OK && resplen == 0) {
tp->MsgFlags |= TARGET_STATUS_SEND_FLAGS_AUTO_GOOD_STATUS;
} else {
tp->StatusDataSGE.u.Address32 = (uint32_t) paddr;
tp->StatusDataSGE.FlagsLength =
MPI_SGE_FLAGS_HOST_TO_IOC |
MPI_SGE_FLAGS_SIMPLE_ELEMENT |
MPI_SGE_FLAGS_LAST_ELEMENT |
MPI_SGE_FLAGS_END_OF_LIST |
MPI_SGE_FLAGS_END_OF_BUFFER;
tp->StatusDataSGE.FlagsLength <<= MPI_SGE_FLAGS_SHIFT;
tp->StatusDataSGE.FlagsLength |= resplen;
}
mpt_lprt(mpt, MPT_PRT_DEBUG,
"STATUS_CCB %p (wit%s sense) tag %x req %p:%u resid %u\n",
ccb, sense_data?"h" : "hout", ccb? ccb->csio.tag_id : -1, req,
req->serno, tgt->resid);
if (ccb) {
ccb->ccb_h.status = CAM_SIM_QUEUED | CAM_REQ_INPROG;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
ccb->ccb_h.timeout_ch = timeout(mpt_timeout, ccb, 60 * hz);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
mpt_send_cmd(mpt, req);
}
static void
mpt_scsi_tgt_tsk_mgmt(struct mpt_softc *mpt, request_t *req, mpt_task_mgmt_t fc,
tgt_resource_t *trtp, int init_id)
{
struct ccb_immed_notify *inot;
mpt_tgt_state_t *tgt;
tgt = MPT_TGT_STATE(mpt, req);
inot = (struct ccb_immed_notify *) STAILQ_FIRST(&trtp->inots);
if (inot == NULL) {
mpt_lprt(mpt, MPT_PRT_WARN, "no INOTSs- sending back BSY\n");
mpt_scsi_tgt_status(mpt, NULL, req, SCSI_STATUS_BUSY, NULL);
return;
}
STAILQ_REMOVE_HEAD(&trtp->inots, sim_links.stqe);
mpt_lprt(mpt, MPT_PRT_DEBUG1,
"Get FREE INOT %p lun %d\n", inot, inot->ccb_h.target_lun);
memset(&inot->sense_data, 0, sizeof (inot->sense_data));
inot->sense_len = 0;
memset(inot->message_args, 0, sizeof (inot->message_args));
inot->initiator_id = init_id; /* XXX */
/*
* This is a somewhat grotesque attempt to map from task management
* to old style SCSI messages. God help us all.
*/
switch (fc) {
case MPT_ABORT_TASK_SET:
inot->message_args[0] = MSG_ABORT_TAG;
break;
case MPT_CLEAR_TASK_SET:
inot->message_args[0] = MSG_CLEAR_TASK_SET;
break;
case MPT_TARGET_RESET:
inot->message_args[0] = MSG_TARGET_RESET;
break;
case MPT_CLEAR_ACA:
inot->message_args[0] = MSG_CLEAR_ACA;
break;
case MPT_TERMINATE_TASK:
inot->message_args[0] = MSG_ABORT_TAG;
break;
default:
inot->message_args[0] = MSG_NOOP;
break;
}
tgt->ccb = (union ccb *) inot;
inot->ccb_h.status = CAM_MESSAGE_RECV|CAM_DEV_QFRZN;
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
xpt_done((union ccb *)inot);
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
static void
mpt_scsi_tgt_atio(struct mpt_softc *mpt, request_t *req, uint32_t reply_desc)
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
{
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
struct ccb_accept_tio *atiop;
lun_id_t lun;
int tag_action = 0;
mpt_tgt_state_t *tgt;
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
tgt_resource_t *trtp = NULL;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
U8 *lunptr;
U8 *vbuf;
U16 itag;
U16 ioindex;
mpt_task_mgmt_t fct = MPT_NIL_TMT_VALUE;
uint8_t *cdbp;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
* First, DMA sync the received command-
* which is in the *request* * phys area.
*
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
* XXX: We could optimize this for a range
*/
bus_dmamap_sync(mpt->request_dmat, mpt->request_dmap,
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
BUS_DMASYNC_POSTREAD);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Stash info for the current command where we can get at it later.
*/
vbuf = req->req_vbuf;
vbuf += MPT_RQSL(mpt);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Get our state pointer set up.
*/
tgt = MPT_TGT_STATE(mpt, req);
if (tgt->state != TGT_STATE_LOADED) {
mpt_tgt_dump_req_state(mpt, req);
panic("bad target state in mpt_scsi_tgt_atio");
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
memset(tgt, 0, sizeof (mpt_tgt_state_t));
tgt->state = TGT_STATE_IN_CAM;
tgt->reply_desc = reply_desc;
ioindex = GET_IO_INDEX(reply_desc);
Fix some negotiation issues (like not being able to negotiate async)
2006-11-02 23:19:00 +00:00
if (mpt->verbose >= MPT_PRT_DEBUG) {
mpt_dump_data(mpt, "mpt_scsi_tgt_atio response", vbuf,
max(sizeof (MPI_TARGET_FCP_CMD_BUFFER),
max(sizeof (MPI_TARGET_SSP_CMD_BUFFER),
sizeof (MPI_TARGET_SCSI_SPI_CMD_BUFFER))));
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (mpt->is_fc) {
PTR_MPI_TARGET_FCP_CMD_BUFFER fc;
fc = (PTR_MPI_TARGET_FCP_CMD_BUFFER) vbuf;
if (fc->FcpCntl[2]) {
/*
* Task Management Request
*/
switch (fc->FcpCntl[2]) {
case 0x2:
fct = MPT_ABORT_TASK_SET;
break;
case 0x4:
fct = MPT_CLEAR_TASK_SET;
break;
case 0x20:
fct = MPT_TARGET_RESET;
break;
case 0x40:
fct = MPT_CLEAR_ACA;
break;
case 0x80:
fct = MPT_TERMINATE_TASK;
break;
default:
mpt_prt(mpt, "CORRUPTED TASK MGMT BITS: 0x%x\n",
fc->FcpCntl[2]);
mpt_scsi_tgt_status(mpt, 0, req,
SCSI_STATUS_OK, 0);
return;
}
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
} else {
switch (fc->FcpCntl[1]) {
case 0:
tag_action = MSG_SIMPLE_Q_TAG;
break;
case 1:
tag_action = MSG_HEAD_OF_Q_TAG;
break;
case 2:
tag_action = MSG_ORDERED_Q_TAG;
break;
default:
/*
* Bah. Ignore Untagged Queing and ACA
*/
tag_action = MSG_SIMPLE_Q_TAG;
break;
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->resid = be32toh(fc->FcpDl);
cdbp = fc->FcpCdb;
lunptr = fc->FcpLun;
itag = be16toh(fc->OptionalOxid);
} else if (mpt->is_sas) {
PTR_MPI_TARGET_SSP_CMD_BUFFER ssp;
ssp = (PTR_MPI_TARGET_SSP_CMD_BUFFER) vbuf;
cdbp = ssp->CDB;
lunptr = ssp->LogicalUnitNumber;
itag = ssp->InitiatorTag;
} else {
PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER sp;
sp = (PTR_MPI_TARGET_SCSI_SPI_CMD_BUFFER) vbuf;
cdbp = sp->CDB;
lunptr = sp->LogicalUnitNumber;
itag = sp->Tag;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Generate a simple lun
*/
switch (lunptr[0] & 0xc0) {
case 0x40:
lun = ((lunptr[0] & 0x3f) << 8) | lunptr[1];
break;
case 0:
lun = lunptr[1];
break;
default:
mpt_lprt(mpt, MPT_PRT_ERROR, "cannot handle this type lun\n");
lun = 0xffff;
break;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Deal with non-enabled or bad luns here.
*/
if (lun >= MPT_MAX_LUNS || mpt->tenabled == 0 ||
mpt->trt[lun].enabled == 0) {
if (mpt->twildcard) {
trtp = &mpt->trt_wildcard;
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
} else if (fct == MPT_NIL_TMT_VALUE) {
/*
* In this case, we haven't got an upstream listener
* for either a specific lun or wildcard luns. We
* have to make some sensible response. For regular
* inquiry, just return some NOT HERE inquiry data.
* For VPD inquiry, report illegal field in cdb.
* For REQUEST SENSE, just return NO SENSE data.
* REPORT LUNS gets illegal command.
* All other commands get 'no such device'.
*/
uint8_t *sp, cond, buf[MPT_SENSE_SIZE];
mpt_prt(mpt, "CMD 0x%x to unmanaged lun %u\n",
cdbp[0], lun);
memset(buf, 0, MPT_SENSE_SIZE);
cond = SCSI_STATUS_CHECK_COND;
buf[0] = 0xf0;
buf[2] = 0x5;
buf[7] = 0x8;
sp = buf;
tgt->tag_id = MPT_MAKE_TAGID(mpt, req, ioindex);
switch (cdbp[0]) {
case INQUIRY:
{
static uint8_t iqd[8] = {
0x7f, 0x0, 0x4, 0x12, 0x0
};
if (cdbp[1] != 0) {
buf[12] = 0x26;
buf[13] = 0x01;
break;
}
mpt_prt(mpt, "local inquiry\n");
mpt_scsi_tgt_local(mpt, req, lun, 1,
iqd, sizeof (iqd));
return;
}
case REQUEST_SENSE:
{
buf[2] = 0x0;
mpt_prt(mpt, "local request sense\n");
mpt_scsi_tgt_local(mpt, req, lun, 1,
buf, sizeof (buf));
return;
}
case REPORT_LUNS:
buf[12] = 0x26;
break;
default:
buf[12] = 0x25;
break;
}
mpt_scsi_tgt_status(mpt, NULL, req, cond, sp);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return;
}
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
/* otherwise, leave trtp NULL */
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
} else {
trtp = &mpt->trt[lun];
}
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
/*
* Deal with any task management
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (fct != MPT_NIL_TMT_VALUE) {
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
if (trtp == NULL) {
mpt_prt(mpt, "task mgmt function %x but no listener\n",
fct);
mpt_scsi_tgt_status(mpt, 0, req,
SCSI_STATUS_OK, 0);
} else {
mpt_scsi_tgt_tsk_mgmt(mpt, req, fct, trtp,
GET_INITIATOR_INDEX(reply_desc));
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return;
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Change some order of the way we do some target mode ops. Found by Coverity.
2006-04-21 18:31:21 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
atiop = (struct ccb_accept_tio *) STAILQ_FIRST(&trtp->atios);
if (atiop == NULL) {
mpt_lprt(mpt, MPT_PRT_WARN,
"no ATIOs for lun %u- sending back %s\n", lun,
mpt->tenabled? "QUEUE FULL" : "BUSY");
mpt_scsi_tgt_status(mpt, NULL, req,
mpt->tenabled? SCSI_STATUS_QUEUE_FULL : SCSI_STATUS_BUSY,
NULL);
return;
}
STAILQ_REMOVE_HEAD(&trtp->atios, sim_links.stqe);
mpt_lprt(mpt, MPT_PRT_DEBUG1,
"Get FREE ATIO %p lun %d\n", atiop, atiop->ccb_h.target_lun);
atiop->ccb_h.ccb_mpt_ptr = mpt;
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
atiop->ccb_h.status = CAM_CDB_RECVD;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
atiop->ccb_h.target_lun = lun;
atiop->sense_len = 0;
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
atiop->init_id = GET_INITIATOR_INDEX(reply_desc);
atiop->cdb_len = mpt_cdblen(cdbp[0], 16);
memcpy(atiop->cdb_io.cdb_bytes, cdbp, atiop->cdb_len);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* The tag we construct here allows us to find the
* original request that the command came in with.
*
* This way we don't have to depend on anything but the
* tag to find things when CCBs show back up from CAM.
*/
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
atiop->tag_id = MPT_MAKE_TAGID(mpt, req, ioindex);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->tag_id = atiop->tag_id;
if (tag_action) {
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
atiop->tag_action = tag_action;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
atiop->ccb_h.flags = CAM_TAG_ACTION_VALID;
}
if (mpt->verbose >= MPT_PRT_DEBUG) {
int i;
mpt_prt(mpt, "START_CCB %p for lun %u CDB=<", atiop,
atiop->ccb_h.target_lun);
for (i = 0; i < atiop->cdb_len; i++) {
mpt_prtc(mpt, "%02x%c", cdbp[i] & 0xff,
(i == (atiop->cdb_len - 1))? '>' : ' ');
}
mpt_prtc(mpt, " itag %x tag %x rdesc %x dl=%u\n",
itag, atiop->tag_id, tgt->reply_desc, tgt->resid);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
MPTLOCK_2_CAMLOCK(mpt);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
xpt_done((union ccb *)atiop);
In receiving a new ATIO, don't record the associated CCB in the target state structure. This field is only for CCBs that are associated with actions that are occurring on the HBA (i.e., XPT_CONT_IO actions). This way we also don't get confused when the upstream listener stalls try and look at a CCB which has already been freed (by CAM).
2006-04-18 21:52:00 +00:00
CAMLOCK_2_MPTLOCK(mpt);
Some fairly major changes to this driver. A) Fibre Channel Target Mode support mostly works (SAS/SPI won't be too far behind). I'd say that this probably works just about as well as isp(4) does right now. Still, it and isp(4) and the whole target mode stack need a bit of tightening. B) The startup sequence has been changed so that after all attaches are done, a set of enable functions are called. The idea here is that the attaches do whatever needs to be done *prior* to a port being enabled and the enables do what need to be done for enabling stuff for a port after it's been enabled. This means that we also have events handled by their proper handlers as we start up. C) Conditional code that means that this driver goes back all the way to RELENG_4 in terms of support. D) Quite a lot of little nitty bug fixes- some discovered by doing RELENG_4 support. We've been living under Giant *waaaayyyyy* too long and it's made some of us (me) sloppy. E) Some shutdown hook stuff that makes sure we don't blow up during a reboot (like by the arrival of a new command from an initiator). There's been some testing and LINT checking, but not as complete as would be liked. Regression testing with Fusion RAID instances has not been possible. Caveat Emptor. Sponsored by: LSI-Logic.
2006-03-25 07:08:27 +00:00
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
static void
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_tgt_dump_tgt_state(struct mpt_softc *mpt, request_t *req)
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
{
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, req);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_prt(mpt, "req %p:%u tgt:rdesc 0x%x resid %u xfrd %u ccb %p treq %p "
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"nx %d tag 0x%08x state=%d\n", req, req->serno, tgt->reply_desc,
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->resid, tgt->bytes_xfered, tgt->ccb, tgt->req, tgt->nxfers,
tgt->tag_id, tgt->state);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
static void
mpt_tgt_dump_req_state(struct mpt_softc *mpt, request_t *req)
{
mpt_prt(mpt, "req %p:%u index %u (%x) state %x\n", req, req->serno,
req->index, req->index, req->state);
mpt_tgt_dump_tgt_state(mpt, req);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
static int
mpt_scsi_tgt_reply_handler(struct mpt_softc *mpt, request_t *req,
uint32_t reply_desc, MSG_DEFAULT_REPLY *reply_frame)
{
int dbg;
union ccb *ccb;
U16 status;
if (reply_frame == NULL) {
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
* Figure out what the state of the command is.
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_tgt_state_t *tgt = MPT_TGT_STATE(mpt, req);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_spcl(mpt, req, "turbo scsi_tgt_reply", __LINE__);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (tgt->req) {
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_req_not_spcl(mpt, tgt->req,
"turbo scsi_tgt_reply associated req", __LINE__);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#endif
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
switch(tgt->state) {
case TGT_STATE_LOADED:
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* This is a new command starting.
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_scsi_tgt_atio(mpt, req, reply_desc);
break;
case TGT_STATE_MOVING_DATA:
{
uint8_t *sp = NULL, sense[MPT_SENSE_SIZE];
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
ccb = tgt->ccb;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (tgt->req == NULL) {
panic("mpt: turbo target reply with null "
"associated request moving data");
/* NOTREACHED */
}
if (ccb == NULL) {
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
if (tgt->is_local == 0) {
panic("mpt: turbo target reply with "
"null associated ccb moving data");
/* NOTREACHED */
}
mpt_lprt(mpt, MPT_PRT_DEBUG,
"TARGET_ASSIST local done\n");
TAILQ_REMOVE(&mpt->request_pending_list,
tgt->req, links);
mpt_free_request(mpt, tgt->req);
tgt->req = NULL;
mpt_scsi_tgt_status(mpt, NULL, req,
0, NULL);
return (TRUE);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->ccb = NULL;
tgt->nxfers++;
untimeout(mpt_timeout, ccb, ccb->ccb_h.timeout_ch);
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"TARGET_ASSIST %p (req %p:%u) done tag 0x%x\n",
ccb, tgt->req, tgt->req->serno, ccb->csio.tag_id);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
* Free the Target Assist Request
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
KASSERT(tgt->req->ccb == ccb,
("tgt->req %p:%u tgt->req->ccb %p", tgt->req,
tgt->req->serno, tgt->req->ccb));
TAILQ_REMOVE(&mpt->request_pending_list,
tgt->req, links);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_free_request(mpt, tgt->req);
tgt->req = NULL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Do we need to send status now? That is, are
* we done with all our data transfers?
*/
if ((ccb->ccb_h.flags & CAM_SEND_STATUS) == 0) {
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
KASSERT(ccb->ccb_h.status,
("zero ccb sts at %d\n", __LINE__));
tgt->state = TGT_STATE_IN_CAM;
if (mpt->outofbeer) {
ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
mpt->outofbeer = 0;
mpt_lprt(mpt, MPT_PRT_DEBUG, "THAWQ\n");
}
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
break;
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
/*
* Otherwise, send status (and sense)
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (ccb->ccb_h.flags & CAM_SEND_SENSE) {
sp = sense;
memcpy(sp, &ccb->csio.sense_data,
min(ccb->csio.sense_len, MPT_SENSE_SIZE));
}
mpt_scsi_tgt_status(mpt, ccb, req,
ccb->csio.scsi_status, sp);
break;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
case TGT_STATE_SENDING_STATUS:
case TGT_STATE_MOVING_DATA_AND_STATUS:
{
int ioindex;
ccb = tgt->ccb;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (tgt->req == NULL) {
panic("mpt: turbo target reply with null "
"associated request sending status");
/* NOTREACHED */
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (ccb) {
tgt->ccb = NULL;
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
if (tgt->state ==
TGT_STATE_MOVING_DATA_AND_STATUS) {
tgt->nxfers++;
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
untimeout(mpt_timeout, ccb,
ccb->ccb_h.timeout_ch);
if (ccb->ccb_h.flags & CAM_SEND_SENSE) {
ccb->ccb_h.status |= CAM_SENT_SENSE;
}
mpt_lprt(mpt, MPT_PRT_DEBUG,
"TARGET_STATUS tag %x sts %x flgs %x req "
"%p\n", ccb->csio.tag_id, ccb->ccb_h.status,
ccb->ccb_h.flags, tgt->req);
/*
* Free the Target Send Status Request
*/
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
KASSERT(tgt->req->ccb == ccb,
("tgt->req %p:%u tgt->req->ccb %p",
tgt->req, tgt->req->serno, tgt->req->ccb));
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* Notify CAM that we're done
*/
mpt_set_ccb_status(ccb, CAM_REQ_CMP);
ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
KASSERT(ccb->ccb_h.status,
("ZERO ccb sts at %d\n", __LINE__));
tgt->ccb = NULL;
} else {
mpt_lprt(mpt, MPT_PRT_DEBUG,
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
"TARGET_STATUS non-CAM for req %p:%u\n",
tgt->req, tgt->req->serno);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
TAILQ_REMOVE(&mpt->request_pending_list,
tgt->req, links);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_free_request(mpt, tgt->req);
tgt->req = NULL;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
/*
* And re-post the Command Buffer.
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
* This will reset the state.
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
*/
ioindex = GET_IO_INDEX(reply_desc);
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
If the card has target mode enabled, and we hang out ELS buffers but *don't* hang out commands, we hang folks on the SAN because the LSI-Logic f/w apparently sends back BUSY or QFULL or some darn thing. If we add command buffers, we have to respond to them sensibly even if we don't have any upstream listeners (scsi_targ or scsi_targ_bh), so put in some local command reponse stuff. MFC after: 2 weeks
2006-07-15 22:58:09 +00:00
tgt->is_local = 0;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_post_target_command(mpt, req, ioindex);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
/*
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
* And post a done for anyone who cares
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
*/
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
if (ccb) {
if (mpt->outofbeer) {
ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
mpt->outofbeer = 0;
mpt_lprt(mpt, MPT_PRT_DEBUG, "THAWQ\n");
}
MPTLOCK_2_CAMLOCK(mpt);
xpt_done(ccb);
CAMLOCK_2_MPTLOCK(mpt);
}
break;
}
case TGT_STATE_NIL: /* XXX This Never Happens XXX */
tgt->state = TGT_STATE_LOADED;
break;
default:
mpt_prt(mpt, "Unknown Target State 0x%x in Context "
"Reply Function\n", tgt->state);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return (TRUE);
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
status = le16toh(reply_frame->IOCStatus);
if (status != MPI_IOCSTATUS_SUCCESS) {
dbg = MPT_PRT_ERROR;
} else {
dbg = MPT_PRT_DEBUG1;
}
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_lprt(mpt, dbg,
"SCSI_TGT REPLY: req=%p:%u reply=%p func=%x IOCstatus 0x%x\n",
req, req->serno, reply_frame, reply_frame->Function, status);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
switch (reply_frame->Function) {
case MPI_FUNCTION_TARGET_CMD_BUFFER_POST:
{
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
mpt_tgt_state_t *tgt;
#ifdef INVARIANTS
mpt_req_spcl(mpt, req, "tgt reply BUFFER POST", __LINE__);
#endif
if (status != MPI_IOCSTATUS_SUCCESS) {
/*
* XXX What to do?
*/
break;
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
}
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
tgt = MPT_TGT_STATE(mpt, req);
KASSERT(tgt->state == TGT_STATE_LOADING,
("bad state 0x%x on reply to buffer post\n", tgt->state));
mpt_assign_serno(mpt, req);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
tgt->state = TGT_STATE_LOADED;
break;
}
case MPI_FUNCTION_TARGET_ASSIST:
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_not_spcl(mpt, req, "tgt reply TARGET ASSIST", __LINE__);
#endif
mpt_prt(mpt, "target assist completion\n");
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_free_request(mpt, req);
break;
case MPI_FUNCTION_TARGET_STATUS_SEND:
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_not_spcl(mpt, req, "tgt reply STATUS SEND", __LINE__);
#endif
mpt_prt(mpt, "status send completion\n");
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_free_request(mpt, req);
break;
case MPI_FUNCTION_TARGET_MODE_ABORT:
{
PTR_MSG_TARGET_MODE_ABORT_REPLY abtrp =
(PTR_MSG_TARGET_MODE_ABORT_REPLY) reply_frame;
PTR_MSG_TARGET_MODE_ABORT abtp =
(PTR_MSG_TARGET_MODE_ABORT) req->req_vbuf;
uint32_t cc = GET_IO_INDEX(le32toh(abtp->ReplyWord));
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
#ifdef INVARIANTS
mpt_req_not_spcl(mpt, req, "tgt reply TMODE ABORT", __LINE__);
#endif
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_prt(mpt, "ABORT RX_ID 0x%x Complete; status 0x%x cnt %u\n",
cc, le16toh(abtrp->IOCStatus), le32toh(abtrp->AbortCount));
A large set of changes: + Add boatloads of KASSERTs and *really* check out more locking issues (to catch recursions when we actually go to real locking in CAM soon). The KASSERTs also caught lots of other issues like using commands that were put back on free lists, etc. + Target mode: role setting is derived directly from port capabilities. There is no need to set a role any more. Some target mode resources are allocated early on (ELS), but target command buffer allocation is deferred until the first lun enable. + Fix some breakages I introduced with target mode in that some commands are *repeating* commands. That is, the reply shows up but the command isn't really done (we don't free it). We still need to take it off the pending list because when we resubmit it, bad things then happen. + Fix more of the way that timed out commands and bus reset is done. The actual TMF response code was being ignored. + For SPI, honor BIOS settings. This doesn't quite fix the problems we've seen where we can't seem to (re)negotiate U320 on all drives but avoids it instead by letting us honor the BIOS settings. I'm sure this is not quite right and will have to change again soon.
2006-04-11 16:47:30 +00:00
TAILQ_REMOVE(&mpt->request_pending_list, req, links);
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
mpt_free_request(mpt, req);
break;
}
default:
mpt_prt(mpt, "Unknown Target Address Reply Function code: "
"0x%x\n", reply_frame->Function);
break;
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Fix some of the previus changes 'better'. There's something strange going on with async events. They seem to be be treated differently for different Fusion implementations. Some will really tell you when it's okay to free the request that started them. Some won't. Very disconcerting. This is particularily bad when the chip (FC in this case) tells you in the reply that it's not a continuation reply, which means you can free the request that its associated with. However, if you do that, I've found that additional async event replies come back for that message context after you freed it. Very Bad Things Happen. Put in a reply register debounce. Warn about out of range context indices. Use more MPILIB defines where possible. Replace bzero with memset. Add tons more KASSERTS. Do a *lot* more request free list auditting and serial number usages. Get rid of the warning about the short IOC Facts Reply. Go back to 16 bits of context index. Do a lot more target state auditting as well. Make a tag out of not only the ioindex but the request index as well and worry less about keeping a full serial number.
2006-04-01 07:12:18 +00:00
return (TRUE);
Massive overhaul of MPT Fusion driver: o Add timeout error recovery (from a thread context to avoid the deferral of other critical interrupts). o Properly recover commands across controller reset events. o Update the driver to handle events and status codes that have been added to the MPI spec since the driver was originally written. o Make the driver more modular to improve maintainability and support dynamic "personality" registration (e.g. SCSI Initiator, RAID, SAS, FC, etc). o Shorten and simplify the common I/O path to improve driver performance. o Add RAID volume and RAID member state/settings reporting. o Add periodic volume resynchronization status reporting. o Add support for sysctl tunable resync rate, member write cache enable, and volume transaction queue depth. Sponsored by ---------------- Avid Technologies Inc: SCSI error recovery, driver re-organization, update of MPI library headers, portions of dynamic personality registration, and misc bug fixes. Wheel Open Technologies: RAID event notification, RAID member pass-thru support, firmware upload/download support, enhanced RAID resync speed, portions of dynamic personality registration, and misc bug fixes. Detailed Changes ================ mpt.c mpt_cam.c mpt_raid.c mpt_pci.c: o Add support for personality modules. Each module exports load, and unload module scope methods as well as probe, attach, event, reset, shutdown, and detach per-device instance methods mpt.c mpt.h mpt_pci.c: o The driver now associates a callback function (via an index) with every transaction submitted to the controller. This allows the main interrupt handler to absolve itself of any knowledge of individual transaction/response types by simply calling the callback function "registered" for the transaction. We use a callback index instead of a callback function pointer in each requests so we can properly handle responses (e.g. event notifications) that are not associated with a transaction. Personality modules dynamically register their callbacks with the driver core to receive the callback index to use for their handlers. o Move the interrupt handler into mpt.c. The ISR algorithm is bus transport and OS independent and thus had no reason to be in mpt_pci.c. o Simplify configuration message reply handling by copying reply frame data for the requester and storing completion status in the original request structure. o Add the mpt_complete_request_chain() helper method and use it to implement reset handlers that must abort transactions. o Keep track of all pending requests on the new requests_pending_list in the softc. o Add default handlers to mpt.c to handle generic event notifications and controller reset activities. The event handler code is largely the same as in the original driver. The reset handler is new and terminates any pending transactions with a status code indicating the controller needs to be re-initialized. o Add some endian support to the driver. A complete audit is still required for this driver to have any hope of operating in a big-endian environment. o Use inttypes.h and __inline. Come closer to being style(9) compliant. o Remove extraneous use of typedefs. o Convert request state from a strict enumeration to a series of flags. This allows us to, for example, tag transactions that have timed-out while retaining the state that the transaction is still in-flight on the controller. o Add mpt_wait_req() which allows a caller to poll or sleep for the completion of a request. Use this to simplify and factor code out from many initialization routines. We also use this to sleep for task management request completions in our CAM timeout handler. mpt.c: o Correct a bug in the event handler where request structures were freed even if the request reply was marked as a continuation reply. Continuation replies indicate that the controller still owns the request and freeing these replies prematurely corrupted controller state. o Implement firmware upload and download. On controllers that do not have dedicated NVRAM (as in the Sun v20/v40z), the firmware image is downloaded to the controller by the system BIOS. This image occupies precious controller RAM space until the host driver fetches the image, reducing the number of concurrent I/Os the controller can processes. The uploaded image is used to re-program the controller during hard reset events since the controller cannot fetch the firmware on its own. Implementing this feature allows much higher queue depths when RAID volumes are configured. o Changed configuration page accessors to allow threads to sleep rather than busy wait for completion. o Removed hard coded data transfer sizes from configuration page routines so that RAID configuration page processing is possible. mpt_reg.h: o Move controller register definitions into a separate file. mpt.h: o Re-arrange includes to allow inlined functions to be defined in mpt.h. o Add reply, event, and reset handler definitions. o Add softc fields for handling timeout and controller reset recovery. mpt_cam.c: o Move mpt_freebsd.c to mpt_cam.c. Move all core functionality, such as event handling, into mpt.c leaving only CAM SCSI support here. o Revamp completion handler to provide correct CAM status for all currently defined SCSI MPI message result codes. o Register event and reset handlers with the MPT core. Modify the event handler to notify CAM of bus reset events. The controller reset handler will abort any transactions that have timed out. All other pending CAM transactions are correctly aborted by the core driver's reset handler. o Allocate a single request up front to perform task management operations. This guarantees that we can always perform a TMF operation even when the controller is saturated with other operations. The single request also serves as a perfect mechanism of guaranteeing that only a single TMF is in flight at a time - something that is required according to the MPT Fusion documentation. o Add a helper function for issuing task management requests to the controller. This is used to abort individual requests or perform a bus reset. o Modify the CAM XPT_BUS_RESET ccb handler to wait for and properly handle the status of the bus reset task management frame used to reset the bus. The previous code assumed that the reset request would always succeed. o Add timeout recovery support. When a timeout occurs, the timed-out request is added to a queue to be processed by our recovery thread and the thread is woken up. The recovery thread processes timed-out command serially, attempting first to abort them and then falling back to a bus reset if an abort fails. o Add calls to mpt_reset() to reset the controller if any handshake command, bus reset attempt or abort attempt fails due to a timeout. o Export a secondary "bus" to CAM that exposes all volume drive members as pass-thru devices, allowing CAM to perform proper speed negotiation to hidden devices. o Add a CAM async event handler tracking the AC_FOUND_DEVICE event. Use this to trigger calls to set the per-volume queue depth once the volume is fully registered with CAM. This is required to avoid hitting firmware limits on volume queue depth. Exceeding the limit causes the firmware to hang. mpt_cam.h: o Add several helper functions for interfacing to CAM and performing timeout recovery. mpt_pci.c: o Disable interrupts on the controller before registering and enabling interrupt delivery to the OS. Otherwise we risk receiving interrupts before the driver is ready to receive them. o Make use of compatibility macros that allow the driver to be compiled under 4.x and 5.x. mpt_raid.c: o Add a per-controller instance RAID thread to perform settings changes and query status (minimizes CPU busy wait loops). o Use a shutdown handler to disable "Member Write Cache Enable" (MWCE) setting for RAID arrays set to enable MWCE During Rebuild. o Change reply handler function signature to allow handlers to defer the deletion of reply frames. Use this to allow the event reply handler to queue up events that need to be acked if no resources are available to immediately ack an event. Queued events are processed in mpt_free_request() where resources are freed. This avoids a panic on resource shortage. o Parse and print out RAID controller capabilities during driver probe. o Define, allocate, and maintain RAID data structures for volumes, hidden member physical disks and spare disks. o Add dynamic sysctls for per-instance setting of the log level, array resync rate, array member cache enable, and volume queue depth. mpt_debug.c: o Add mpt_lprt and mpt_lprtc for printing diagnostics conditioned on a particular log level to aid in tracking down driver issues. o Add mpt_decode_value() which parses the bits in an integer value based on a parsing table (mask, value, name string, tuples). mpilib/*: o Update mpi library header files to latest distribution from LSI. Submitted by: gibbs Approved by: re
2005-07-10 15:05:39 +00:00
}
Reference in New Issue Copy Permalink