d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
bd247e9ddd
freebsd-nq/sys/dev/mfi/mfi.c

3801 lines
102 KiB
C
Raw Normal View History

Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*-
* Copyright (c) 2006 IronPort Systems
* 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 the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* 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.
*/
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
/*-
* Copyright (c) 2007 LSI Corp.
* Copyright (c) 2007 Rajesh Prabhakaran.
* 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 the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* 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.
*/
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#include "opt_compat.h"
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
#include "opt_mfi.h"
#include <sys/param.h>
#include <sys/systm.h>
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
#include <sys/sysctl.h>
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
#include <sys/malloc.h>
#include <sys/kernel.h>
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
#include <sys/poll.h>
#include <sys/selinfo.h>
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
#include <sys/bus.h>
#include <sys/conf.h>
#include <sys/eventhandler.h>
#include <sys/rman.h>
#include <sys/bus_dma.h>
#include <sys/bio.h>
#include <sys/ioccom.h>
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
#include <sys/uio.h>
#include <sys/proc.h>
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
#include <sys/signalvar.h>
Fix panics triggered by older mfiutil binaries run on the new mfi(4) driver. The new driver changed the size of the mfi_dcmd_frame structure in such a way that a MFI_IOC_PASSTHRU ioctl from an old amd64 binary is treated as an MFI_IOC_PASSTHRU32 ioctl in the new driver. As a result, the user pointer is treated as the buffer length. mfi_user_command() doesn't have a bounds check on the buffer length, so it passes a really big value to malloc() which panics when it tries to exhaust the kmem_map. Fix this two ways: - Only honor MFI_IOC_PASSTHRU32 if the binary has the SV_ILP32 flag set, otherwise treat it as an unknown ioctl. - Add a bounds check on the buffer length passed by the user. For now it fails any user attempts to use a buffer larger than 1MB. While here, fix a few other nits: - Remove an unnecessary check for a NULL return from malloc(M_WAITOK). - Use the ENOTTY errno for invalid ioctl commands instead of ENOENT. MFC after: 3 days
2012-07-03 16:12:57 +00:00
#include <sys/sysent.h>
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#include <sys/taskqueue.h>
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
#include <machine/bus.h>
#include <machine/resource.h>
#include <dev/mfi/mfireg.h>
#include <dev/mfi/mfi_ioctl.h>
#include <dev/mfi/mfivar.h>
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
#include <sys/interrupt.h>
#include <sys/priority.h>
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int mfi_alloc_commands(struct mfi_softc *);
static int mfi_comms_init(struct mfi_softc *);
static int mfi_get_controller_info(struct mfi_softc *);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int mfi_get_log_state(struct mfi_softc *,
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
struct mfi_evt_log_state **);
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
static int mfi_parse_entries(struct mfi_softc *, int, int);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static void mfi_data_cb(void *, bus_dma_segment_t *, int, int);
static void mfi_startup(void *arg);
static void mfi_intr(void *arg);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
static void mfi_ldprobe(struct mfi_softc *sc);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static void mfi_syspdprobe(struct mfi_softc *sc);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static void mfi_handle_evt(void *context, int pending);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int mfi_aen_register(struct mfi_softc *sc, int seq, int locale);
static void mfi_aen_complete(struct mfi_command *);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
static int mfi_add_ld(struct mfi_softc *sc, int);
static void mfi_add_ld_complete(struct mfi_command *);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static int mfi_add_sys_pd(struct mfi_softc *sc, int);
static void mfi_add_sys_pd_complete(struct mfi_command *);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static struct mfi_command * mfi_bio_command(struct mfi_softc *);
static void mfi_bio_complete(struct mfi_command *);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static struct mfi_command *mfi_build_ldio(struct mfi_softc *,struct bio*);
static struct mfi_command *mfi_build_syspdio(struct mfi_softc *,struct bio*);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int mfi_send_frame(struct mfi_softc *, struct mfi_command *);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
static int mfi_std_send_frame(struct mfi_softc *, struct mfi_command *);
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
static int mfi_abort(struct mfi_softc *, struct mfi_command **);
We no longer need to use d_thread_t, migrate to struct thread *.
2009-05-20 17:29:21 +00:00
static int mfi_linux_ioctl_int(struct cdev *, u_long, caddr_t, int, struct thread *);
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
static void mfi_timeout(void *);
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
static int mfi_user_command(struct mfi_softc *,
struct mfi_ioc_passthru *);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static void mfi_enable_intr_xscale(struct mfi_softc *sc);
static void mfi_enable_intr_ppc(struct mfi_softc *sc);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static int32_t mfi_read_fw_status_xscale(struct mfi_softc *sc);
static int32_t mfi_read_fw_status_ppc(struct mfi_softc *sc);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static int mfi_check_clear_intr_xscale(struct mfi_softc *sc);
static int mfi_check_clear_intr_ppc(struct mfi_softc *sc);
static void mfi_issue_cmd_xscale(struct mfi_softc *sc, bus_addr_t bus_add,
uint32_t frame_cnt);
static void mfi_issue_cmd_ppc(struct mfi_softc *sc, bus_addr_t bus_add,
uint32_t frame_cnt);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static int mfi_config_lock(struct mfi_softc *sc, uint32_t opcode);
static void mfi_config_unlock(struct mfi_softc *sc, int locked);
static int mfi_check_command_pre(struct mfi_softc *sc, struct mfi_command *cm);
static void mfi_check_command_post(struct mfi_softc *sc, struct mfi_command *cm);
static int mfi_check_for_sscd(struct mfi_softc *sc, struct mfi_command *cm);
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
Add single-message MSI support to mfi(4). It is disabled by default but can be enabled via the hw.mfi.msi tunable. Many mfi(4) controllers also support MSI-X, but in testing it seems that many adapters do not work with MSI-X but do work with MSI. MFC after: 2 weeks
2011-11-16 15:39:27 +00:00
SYSCTL_NODE(_hw, OID_AUTO, mfi, CTLFLAG_RD, 0, "MFI driver parameters");
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
static int mfi_event_locale = MFI_EVT_LOCALE_ALL;
Allow hw.mfi.event_locale/hw.mfi.event_class to be set via loader. If an event doesn't match the criteria then don't print it. Some events are not saved in the log (<0 class events).
2006-09-20 18:49:35 +00:00
TUNABLE_INT("hw.mfi.event_locale", &mfi_event_locale);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
SYSCTL_INT(_hw_mfi, OID_AUTO, event_locale, CTLFLAG_RWTUN, &mfi_event_locale,
0, "event message locale");
Allow hw.mfi.event_locale/hw.mfi.event_class to be set via loader. If an event doesn't match the criteria then don't print it. Some events are not saved in the log (<0 class events).
2006-09-20 18:49:35 +00:00
Up the event class reporting from 10 (basically, nothing) to EVT_CLASS_INFO. Submitted by: Doug Ambrisko
2007-01-07 06:43:25 +00:00
static int mfi_event_class = MFI_EVT_CLASS_INFO;
Allow hw.mfi.event_locale/hw.mfi.event_class to be set via loader. If an event doesn't match the criteria then don't print it. Some events are not saved in the log (<0 class events).
2006-09-20 18:49:35 +00:00
TUNABLE_INT("hw.mfi.event_class", &mfi_event_class);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
SYSCTL_INT(_hw_mfi, OID_AUTO, event_class, CTLFLAG_RWTUN, &mfi_event_class,
0, "event message class");
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
static int mfi_max_cmds = 128;
TUNABLE_INT("hw.mfi.max_cmds", &mfi_max_cmds);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
SYSCTL_INT(_hw_mfi, OID_AUTO, max_cmds, CTLFLAG_RDTUN, &mfi_max_cmds,
0, "Max commands limit (-1 = controller limit)");
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static int mfi_detect_jbod_change = 1;
TUNABLE_INT("hw.mfi.detect_jbod_change", &mfi_detect_jbod_change);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
SYSCTL_INT(_hw_mfi, OID_AUTO, detect_jbod_change, CTLFLAG_RWTUN,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
&mfi_detect_jbod_change, 0, "Detect a change to a JBOD");
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int mfi_polled_cmd_timeout = MFI_POLL_TIMEOUT_SECS;
TUNABLE_INT("hw.mfi.polled_cmd_timeout", &mfi_polled_cmd_timeout);
SYSCTL_INT(_hw_mfi, OID_AUTO, polled_cmd_timeout, CTLFLAG_RWTUN,
&mfi_polled_cmd_timeout, 0,
"Polled command timeout - used for firmware flash etc (in seconds)");
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
static int mfi_cmd_timeout = MFI_CMD_TIMEOUT;
TUNABLE_INT("hw.mfi.cmd_timeout", &mfi_cmd_timeout);
SYSCTL_INT(_hw_mfi, OID_AUTO, cmd_timeout, CTLFLAG_RWTUN, &mfi_cmd_timeout,
0, "Command timeout (in seconds)");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/* Management interface */
static d_open_t mfi_open;
static d_close_t mfi_close;
static d_ioctl_t mfi_ioctl;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static d_poll_t mfi_poll;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static struct cdevsw mfi_cdevsw = {
.d_version = D_VERSION,
.d_flags = 0,
.d_open = mfi_open,
.d_close = mfi_close,
.d_ioctl = mfi_ioctl,
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
.d_poll = mfi_poll,
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
.d_name = "mfi",
};
MALLOC_DEFINE(M_MFIBUF, "mfibuf", "Buffers for the MFI driver");
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
#define MFI_INQ_LENGTH SHORT_INQUIRY_LENGTH
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_skinny_dma_info mfi_skinny;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
static void
mfi_enable_intr_xscale(struct mfi_softc *sc)
{
MFI_WRITE4(sc, MFI_OMSK, 0x01);
}
static void
mfi_enable_intr_ppc(struct mfi_softc *sc)
{
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
if (sc->mfi_flags & MFI_FLAGS_1078) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
MFI_WRITE4(sc, MFI_ODCR0, 0xFFFFFFFF);
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
MFI_WRITE4(sc, MFI_OMSK, ~MFI_1078_EIM);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
else if (sc->mfi_flags & MFI_FLAGS_GEN2) {
MFI_WRITE4(sc, MFI_ODCR0, 0xFFFFFFFF);
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
MFI_WRITE4(sc, MFI_OMSK, ~MFI_GEN2_EIM);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
else if (sc->mfi_flags & MFI_FLAGS_SKINNY) {
MFI_WRITE4(sc, MFI_OMSK, ~0x00000001);
}
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
}
static int32_t
mfi_read_fw_status_xscale(struct mfi_softc *sc)
{
return MFI_READ4(sc, MFI_OMSG0);
}
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
static int32_t
mfi_read_fw_status_ppc(struct mfi_softc *sc)
{
return MFI_READ4(sc, MFI_OSP0);
}
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
static int
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
mfi_check_clear_intr_xscale(struct mfi_softc *sc)
{
int32_t status;
status = MFI_READ4(sc, MFI_OSTS);
if ((status & MFI_OSTS_INTR_VALID) == 0)
return 1;
MFI_WRITE4(sc, MFI_OSTS, status);
return 0;
Fix style nit: s/^ }/}/ in two places.
2008-08-23 23:30:54 +00:00
}
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
static int
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
mfi_check_clear_intr_ppc(struct mfi_softc *sc)
{
int32_t status;
status = MFI_READ4(sc, MFI_OSTS);
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
if (sc->mfi_flags & MFI_FLAGS_1078) {
if (!(status & MFI_1078_RM)) {
return 1;
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
else if (sc->mfi_flags & MFI_FLAGS_GEN2) {
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
if (!(status & MFI_GEN2_RM)) {
return 1;
}
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
else if (sc->mfi_flags & MFI_FLAGS_SKINNY) {
if (!(status & MFI_SKINNY_RM)) {
return 1;
}
}
if (sc->mfi_flags & MFI_FLAGS_SKINNY)
MFI_WRITE4(sc, MFI_OSTS, status);
else
MFI_WRITE4(sc, MFI_ODCR0, status);
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
return 0;
Fix style nit: s/^ }/}/ in two places.
2008-08-23 23:30:54 +00:00
}
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
static void
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_issue_cmd_xscale(struct mfi_softc *sc, bus_addr_t bus_add, uint32_t frame_cnt)
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
{
MFI_WRITE4(sc, MFI_IQP,(bus_add >>3)|frame_cnt);
}
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
static void
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_issue_cmd_ppc(struct mfi_softc *sc, bus_addr_t bus_add, uint32_t frame_cnt)
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
{
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY) {
MFI_WRITE4(sc, MFI_IQPL, (bus_add | frame_cnt <<1)|1 );
MFI_WRITE4(sc, MFI_IQPH, 0x00000000);
} else {
MFI_WRITE4(sc, MFI_IQP, (bus_add | frame_cnt <<1)|1 );
}
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_transition_firmware(struct mfi_softc *sc)
{
fw_state ad cur_state are holding unsigned bitfields, so declare then as unsigned as well. Submitted by: rdivacky
2009-06-24 16:11:29 +00:00
uint32_t fw_state, cur_state;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
int max_wait, i;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t cur_abs_reg_val = 0;
uint32_t prev_abs_reg_val = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cur_abs_reg_val = sc->mfi_read_fw_status(sc);
fw_state = cur_abs_reg_val & MFI_FWSTATE_MASK;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
while (fw_state != MFI_FWSTATE_READY) {
if (bootverbose)
device_printf(sc->mfi_dev, "Waiting for firmware to "
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
"become ready\n");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cur_state = fw_state;
switch (fw_state) {
case MFI_FWSTATE_FAULT:
device_printf(sc->mfi_dev, "Firmware fault\n");
return (ENXIO);
case MFI_FWSTATE_WAIT_HANDSHAKE:
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY || sc->mfi_flags & MFI_FLAGS_TBOLT)
MFI_WRITE4(sc, MFI_SKINNY_IDB, MFI_FWINIT_CLEAR_HANDSHAKE);
else
MFI_WRITE4(sc, MFI_IDB, MFI_FWINIT_CLEAR_HANDSHAKE);
max_wait = MFI_RESET_WAIT_TIME;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
case MFI_FWSTATE_OPERATIONAL:
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY || sc->mfi_flags & MFI_FLAGS_TBOLT)
MFI_WRITE4(sc, MFI_SKINNY_IDB, 7);
else
MFI_WRITE4(sc, MFI_IDB, MFI_FWINIT_READY);
max_wait = MFI_RESET_WAIT_TIME;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
case MFI_FWSTATE_UNDEFINED:
case MFI_FWSTATE_BB_INIT:
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
max_wait = MFI_RESET_WAIT_TIME;
break;
case MFI_FWSTATE_FW_INIT_2:
max_wait = MFI_RESET_WAIT_TIME;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
case MFI_FWSTATE_FW_INIT:
case MFI_FWSTATE_FLUSH_CACHE:
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
max_wait = MFI_RESET_WAIT_TIME;
break;
case MFI_FWSTATE_DEVICE_SCAN:
max_wait = MFI_RESET_WAIT_TIME; /* wait for 180 seconds */
prev_abs_reg_val = cur_abs_reg_val;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
Support controllers whose option ROM is disabled in the BIOS by kicking the firmware so that it boots. PR: kern/154978 Submitted by: Andrew Boyer aboyer of averesystems com Obtained from: LSI MFC after: 1 week
2011-07-14 20:20:33 +00:00
case MFI_FWSTATE_BOOT_MESSAGE_PENDING:
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY || sc->mfi_flags & MFI_FLAGS_TBOLT)
MFI_WRITE4(sc, MFI_SKINNY_IDB, MFI_FWINIT_HOTPLUG);
else
MFI_WRITE4(sc, MFI_IDB, MFI_FWINIT_HOTPLUG);
max_wait = MFI_RESET_WAIT_TIME;
Support controllers whose option ROM is disabled in the BIOS by kicking the firmware so that it boots. PR: kern/154978 Submitted by: Andrew Boyer aboyer of averesystems com Obtained from: LSI MFC after: 1 week
2011-07-14 20:20:33 +00:00
break;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
default:
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "Unknown firmware state %#x\n",
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
fw_state);
return (ENXIO);
}
for (i = 0; i < (max_wait * 10); i++) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cur_abs_reg_val = sc->mfi_read_fw_status(sc);
fw_state = cur_abs_reg_val & MFI_FWSTATE_MASK;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (fw_state == cur_state)
DELAY(100000);
else
break;
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (fw_state == MFI_FWSTATE_DEVICE_SCAN) {
/* Check the device scanning progress */
if (prev_abs_reg_val != cur_abs_reg_val) {
continue;
}
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (fw_state == cur_state) {
Support controllers whose option ROM is disabled in the BIOS by kicking the firmware so that it boots. PR: kern/154978 Submitted by: Andrew Boyer aboyer of averesystems com Obtained from: LSI MFC after: 1 week
2011-07-14 20:20:33 +00:00
device_printf(sc->mfi_dev, "Firmware stuck in state "
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
"%#x\n", fw_state);
return (ENXIO);
}
}
return (0);
}
static void
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_addr_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
bus_addr_t *addr;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
addr = arg;
*addr = segs[0].ds_addr;
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
int
mfi_attach(struct mfi_softc *sc)
{
uint32_t status;
int error, commsz, framessz, sensesz;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int frames, unit, max_fw_sge, max_fw_cmds;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t tb_mem_size = 0;
Change Dell's gen2 catch-all from Dell PERC H700/H800 to Dell PERC Gen2 and bump the driver version from 2 to 3 and fix the related style problem. Suggested by: LSI
2008-12-15 17:11:40 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc == NULL)
return EINVAL;
Change Dell's gen2 catch-all from Dell PERC H700/H800 to Dell PERC Gen2 and bump the driver version from 2 to 3 and fix the related style problem. Suggested by: LSI
2008-12-15 17:11:40 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "Megaraid SAS driver Ver %s \n",
MEGASAS_VERSION);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mtx_init(&sc->mfi_io_lock, "MFI I/O lock", NULL, MTX_DEF);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_init(&sc->mfi_config_lock, "MFI config");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
TAILQ_INIT(&sc->mfi_ld_tqh);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
TAILQ_INIT(&sc->mfi_syspd_tqh);
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
TAILQ_INIT(&sc->mfi_ld_pend_tqh);
TAILQ_INIT(&sc->mfi_syspd_pend_tqh);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
TAILQ_INIT(&sc->mfi_evt_queue);
TASK_INIT(&sc->mfi_evt_task, 0, mfi_handle_evt, sc);
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
TASK_INIT(&sc->mfi_map_sync_task, 0, mfi_handle_map_sync, sc);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
TAILQ_INIT(&sc->mfi_aen_pids);
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
TAILQ_INIT(&sc->mfi_cam_ccbq);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_initq_free(sc);
mfi_initq_ready(sc);
mfi_initq_busy(sc);
mfi_initq_bio(sc);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->adpreset = 0;
sc->last_seq_num = 0;
sc->disableOnlineCtrlReset = 1;
sc->issuepend_done = 1;
sc->hw_crit_error = 0;
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
if (sc->mfi_flags & MFI_FLAGS_1064R) {
sc->mfi_enable_intr = mfi_enable_intr_xscale;
sc->mfi_read_fw_status = mfi_read_fw_status_xscale;
sc->mfi_check_clear_intr = mfi_check_clear_intr_xscale;
sc->mfi_issue_cmd = mfi_issue_cmd_xscale;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
} else if (sc->mfi_flags & MFI_FLAGS_TBOLT) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_enable_intr = mfi_tbolt_enable_intr_ppc;
sc->mfi_disable_intr = mfi_tbolt_disable_intr_ppc;
sc->mfi_read_fw_status = mfi_tbolt_read_fw_status_ppc;
sc->mfi_check_clear_intr = mfi_tbolt_check_clear_intr_ppc;
sc->mfi_issue_cmd = mfi_tbolt_issue_cmd_ppc;
sc->mfi_adp_reset = mfi_tbolt_adp_reset;
sc->mfi_tbolt = 1;
TAILQ_INIT(&sc->mfi_cmd_tbolt_tqh);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
} else {
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
sc->mfi_enable_intr = mfi_enable_intr_ppc;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_read_fw_status = mfi_read_fw_status_ppc;
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
sc->mfi_check_clear_intr = mfi_check_clear_intr_ppc;
sc->mfi_issue_cmd = mfi_issue_cmd_ppc;
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/* Before we get too far, see if the firmware is working */
if ((error = mfi_transition_firmware(sc)) != 0) {
device_printf(sc->mfi_dev, "Firmware not in READY state, "
"error %d\n", error);
return (ENXIO);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/* Start: LSIP200113393 */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MEGASAS_MAX_NAME*sizeof(bus_addr_t), /* maxsize */
1, /* msegments */
MEGASAS_MAX_NAME*sizeof(bus_addr_t), /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->verbuf_h_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate verbuf_h_dmat DMA tag\n");
return (ENOMEM);
}
if (bus_dmamem_alloc(sc->verbuf_h_dmat, (void **)&sc->verbuf,
BUS_DMA_NOWAIT, &sc->verbuf_h_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate verbuf_h_dmamap memory\n");
return (ENOMEM);
}
bzero(sc->verbuf, MEGASAS_MAX_NAME*sizeof(bus_addr_t));
bus_dmamap_load(sc->verbuf_h_dmat, sc->verbuf_h_dmamap,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->verbuf, MEGASAS_MAX_NAME*sizeof(bus_addr_t),
mfi_addr_cb, &sc->verbuf_h_busaddr, 0);
/* End: LSIP200113393 */
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Get information needed for sizing the contiguous memory for the
* frame pool. Size down the sgl parameter since we know that
* we will never need more than what's required for MAXPHYS.
* It would be nice if these constants were available at runtime
* instead of compile time.
*/
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
status = sc->mfi_read_fw_status(sc);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
max_fw_cmds = status & MFI_FWSTATE_MAXCMD_MASK;
if (mfi_max_cmds > 0 && mfi_max_cmds < max_fw_cmds) {
device_printf(sc->mfi_dev, "FW MaxCmds = %d, limiting to %d\n",
max_fw_cmds, mfi_max_cmds);
sc->mfi_max_fw_cmds = mfi_max_cmds;
} else {
sc->mfi_max_fw_cmds = max_fw_cmds;
}
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
max_fw_sge = (status & MFI_FWSTATE_MAXSGL_MASK) >> 16;
Separate the parallel scsi knowledge out of the core of the XPT, and modularize it so that new transports can be created. Add a transport for SATA Add a periph+protocol layer for ATA Add a driver for AHCI-compliant hardware. Add a maxio field to CAM so that drivers can advertise their max I/O capability. Modify various drivers so that they are insulated from the value of MAXPHYS. The new ATA/SATA code supports AHCI-compliant hardware, and will override the classic ATA driver if it is loaded as a module at boot time or compiled into the kernel. The stack now support NCQ (tagged queueing) for increased performance on modern SATA drives. It also supports port multipliers. ATA drives are accessed via 'ada' device nodes. ATAPI drives are accessed via 'cd' device nodes. They can all be enumerated and manipulated via camcontrol, just like SCSI drives. SCSI commands are not translated to their ATA equivalents; ATA native commands are used throughout the entire stack, including camcontrol. See the camcontrol manpage for further details. Testing this code may require that you update your fstab, and possibly modify your BIOS to enable AHCI functionality, if available. This code is very experimental at the moment. The userland ABI/API has changed, so applications will need to be recompiled. It may change further in the near future. The 'ada' device name may also change as more infrastructure is completed in this project. The goal is to eventually put all CAM busses and devices until newbus, allowing for interesting topology and management options. Few functional changes will be seen with existing SCSI/SAS/FC drivers, though the userland ABI has still changed. In the future, transports specific modules for SAS and FC may appear in order to better support the topologies and capabilities of these technologies. The modularization of CAM and the addition of the ATA/SATA modules is meant to break CAM out of the mold of being specific to SCSI, letting it grow to be a framework for arbitrary transports and protocols. It also allows drivers to be written to support discrete hardware without jeopardizing the stability of non-related hardware. While only an AHCI driver is provided now, a Silicon Image driver is also in the works. Drivers for ICH1-4, ICH5-6, PIIX, classic IDE, and any other hardware is possible and encouraged. Help with new transports is also encouraged. Submitted by: scottl, mav Approved by: re
2009-07-10 08:18:08 +00:00
sc->mfi_max_sge = min(max_fw_sge, ((MFI_MAXPHYS / PAGE_SIZE) + 1));
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* ThunderBolt Support get the contiguous memory */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->mfi_flags & MFI_FLAGS_TBOLT) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_tbolt_init_globals(sc);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
device_printf(sc->mfi_dev, "MaxCmd = %d, Drv MaxCmd = %d, "
"MaxSgl = %d, state = %#x\n", max_fw_cmds,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_max_fw_cmds, sc->mfi_max_sge, status);
tb_mem_size = mfi_tbolt_get_memory_requirement(sc);
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
tb_mem_size, /* maxsize */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
1, /* msegments */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
tb_mem_size, /* maxsegsize */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_tb_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate comms DMA tag\n");
return (ENOMEM);
}
if (bus_dmamem_alloc(sc->mfi_tb_dmat, (void **)&sc->request_message_pool,
BUS_DMA_NOWAIT, &sc->mfi_tb_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate comms memory\n");
return (ENOMEM);
}
bzero(sc->request_message_pool, tb_mem_size);
bus_dmamap_load(sc->mfi_tb_dmat, sc->mfi_tb_dmamap,
sc->request_message_pool, tb_mem_size, mfi_addr_cb, &sc->mfi_tb_busaddr, 0);
/* For ThunderBolt memory init */
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
0x100, 0, /* alignmnt, boundary */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
NULL, NULL, /* filter, filterarg */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
MFI_FRAME_SIZE, /* maxsize */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
1, /* msegments */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
MFI_FRAME_SIZE, /* maxsegsize */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
&sc->mfi_tb_init_dmat)) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
device_printf(sc->mfi_dev, "Cannot allocate init DMA tag\n");
return (ENOMEM);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
if (bus_dmamem_alloc(sc->mfi_tb_init_dmat, (void **)&sc->mfi_tb_init,
BUS_DMA_NOWAIT, &sc->mfi_tb_init_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate init memory\n");
return (ENOMEM);
}
bzero(sc->mfi_tb_init, MFI_FRAME_SIZE);
bus_dmamap_load(sc->mfi_tb_init_dmat, sc->mfi_tb_init_dmamap,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->mfi_tb_init, MFI_FRAME_SIZE, mfi_addr_cb,
&sc->mfi_tb_init_busaddr, 0);
if (mfi_tbolt_init_desc_pool(sc, sc->request_message_pool,
tb_mem_size)) {
device_printf(sc->mfi_dev,
"Thunderbolt pool preparation error\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return 0;
}
/*
Allocate DMA memory mapping for MPI2 IOC Init descriptor,
we are taking it diffrent from what we have allocated for Request
and reply descriptors to avoid confusion later
*/
tb_mem_size = sizeof(struct MPI2_IOC_INIT_REQUEST);
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
tb_mem_size, /* maxsize */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
1, /* msegments */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
tb_mem_size, /* maxsegsize */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_tb_ioc_init_dmat)) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Cannot allocate comms DMA tag\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return (ENOMEM);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (bus_dmamem_alloc(sc->mfi_tb_ioc_init_dmat,
(void **)&sc->mfi_tb_ioc_init_desc,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
BUS_DMA_NOWAIT, &sc->mfi_tb_ioc_init_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate comms memory\n");
return (ENOMEM);
}
bzero(sc->mfi_tb_ioc_init_desc, tb_mem_size);
bus_dmamap_load(sc->mfi_tb_ioc_init_dmat, sc->mfi_tb_ioc_init_dmamap,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->mfi_tb_ioc_init_desc, tb_mem_size, mfi_addr_cb,
&sc->mfi_tb_ioc_init_busaddr, 0);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Create the dma tag for data buffers. Used both for block I/O
* and for various internal data queries.
*/
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_max_sge, /* nsegments */
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
BUS_DMA_ALLOCNOW, /* flags */
busdma_lock_mutex, /* lockfunc */
&sc->mfi_io_lock, /* lockfuncarg */
&sc->mfi_buffer_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate buffer DMA tag\n");
return (ENOMEM);
}
/*
* Allocate DMA memory for the comms queues. Keep it under 4GB for
* efficiency. The mfi_hwcomms struct includes space for 1 reply queue
* entry, so the calculated size here will be will be 1 more than
* mfi_max_fw_cmds. This is apparently a requirement of the hardware.
*/
commsz = (sizeof(uint32_t) * sc->mfi_max_fw_cmds) +
sizeof(struct mfi_hwcomms);
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
commsz, /* maxsize */
1, /* msegments */
commsz, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_comms_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate comms DMA tag\n");
return (ENOMEM);
}
if (bus_dmamem_alloc(sc->mfi_comms_dmat, (void **)&sc->mfi_comms,
BUS_DMA_NOWAIT, &sc->mfi_comms_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate comms memory\n");
return (ENOMEM);
}
bzero(sc->mfi_comms, commsz);
bus_dmamap_load(sc->mfi_comms_dmat, sc->mfi_comms_dmamap,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_comms, commsz, mfi_addr_cb, &sc->mfi_comms_busaddr, 0);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Allocate DMA memory for the command frames. Keep them in the
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
* lower 4GB for efficiency. Calculate the size of the commands at
* the same time; each command is one 64 byte frame plus a set of
* additional frames for holding sg lists or other data.
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
* The assumption here is that the SG list will start at the second
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
* frame and not use the unused bytes in the first frame. While this
* isn't technically correct, it simplifies the calculation and allows
* for command frames that might be larger than an mfi_io_frame.
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
*/
if (sizeof(bus_addr_t) == 8) {
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_sge_size = sizeof(struct mfi_sg64);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc->mfi_flags |= MFI_FLAGS_SG64;
} else {
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_sge_size = sizeof(struct mfi_sg32);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY)
sc->mfi_sge_size = sizeof(struct mfi_sg_skinny);
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
frames = (sc->mfi_sge_size * sc->mfi_max_sge - 1) / MFI_FRAME_SIZE + 2;
sc->mfi_cmd_size = frames * MFI_FRAME_SIZE;
framessz = sc->mfi_cmd_size * sc->mfi_max_fw_cmds;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
64, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
framessz, /* maxsize */
1, /* nsegments */
framessz, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_frames_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate frame DMA tag\n");
return (ENOMEM);
}
if (bus_dmamem_alloc(sc->mfi_frames_dmat, (void **)&sc->mfi_frames,
BUS_DMA_NOWAIT, &sc->mfi_frames_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate frames memory\n");
return (ENOMEM);
}
bzero(sc->mfi_frames, framessz);
bus_dmamap_load(sc->mfi_frames_dmat, sc->mfi_frames_dmamap,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_frames, framessz, mfi_addr_cb, &sc->mfi_frames_busaddr,0);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Allocate DMA memory for the frame sense data. Keep them in the
* lower 4GB for efficiency
*/
sensesz = sc->mfi_max_fw_cmds * MFI_SENSE_LEN;
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
4, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
sensesz, /* maxsize */
1, /* nsegments */
sensesz, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_sense_dmat)) {
device_printf(sc->mfi_dev, "Cannot allocate sense DMA tag\n");
return (ENOMEM);
}
if (bus_dmamem_alloc(sc->mfi_sense_dmat, (void **)&sc->mfi_sense,
BUS_DMA_NOWAIT, &sc->mfi_sense_dmamap)) {
device_printf(sc->mfi_dev, "Cannot allocate sense memory\n");
return (ENOMEM);
}
bus_dmamap_load(sc->mfi_sense_dmat, sc->mfi_sense_dmamap,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_sense, sensesz, mfi_addr_cb, &sc->mfi_sense_busaddr, 0);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if ((error = mfi_alloc_commands(sc)) != 0)
return (error);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Before moving the FW to operational state, check whether
* hostmemory is required by the FW or not
*/
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* ThunderBolt MFI_IOC2 INIT */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->mfi_flags & MFI_FLAGS_TBOLT) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_disable_intr(sc);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_lock(&sc->mfi_io_lock);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if ((error = mfi_tbolt_init_MFI_queue(sc)) != 0) {
device_printf(sc->mfi_dev,
"TB Init has failed with error %d\n",error);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_unlock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return error;
}
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if ((error = mfi_tbolt_alloc_cmd(sc)) != 0)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return error;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (bus_setup_intr(sc->mfi_dev, sc->mfi_irq,
INTR_MPSAFE|INTR_TYPE_BIO, NULL, mfi_intr_tbolt, sc,
&sc->mfi_intr)) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
device_printf(sc->mfi_dev, "Cannot set up interrupt\n");
return (EINVAL);
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
sc->mfi_intr_ptr = mfi_intr_tbolt;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_enable_intr(sc);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
} else {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((error = mfi_comms_init(sc)) != 0)
return (error);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (bus_setup_intr(sc->mfi_dev, sc->mfi_irq,
INTR_MPSAFE|INTR_TYPE_BIO, NULL, mfi_intr, sc, &sc->mfi_intr)) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
device_printf(sc->mfi_dev, "Cannot set up interrupt\n");
return (EINVAL);
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
sc->mfi_intr_ptr = mfi_intr;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_enable_intr(sc);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((error = mfi_get_controller_info(sc)) != 0)
return (error);
sc->disableOnlineCtrlReset = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/* Register a config hook to probe the bus for arrays */
sc->mfi_ich.ich_func = mfi_startup;
sc->mfi_ich.ich_arg = sc;
if (config_intrhook_establish(&sc->mfi_ich) != 0) {
device_printf(sc->mfi_dev, "Cannot establish configuration "
"hook\n");
return (EINVAL);
}
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_lock(&sc->mfi_io_lock);
if ((error = mfi_aen_setup(sc, 0), 0) != 0) {
mtx_unlock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return (error);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
}
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Register a shutdown handler.
*/
if ((sc->mfi_eh = EVENTHANDLER_REGISTER(shutdown_final, mfi_shutdown,
sc, SHUTDOWN_PRI_DEFAULT)) == NULL) {
device_printf(sc->mfi_dev, "Warning: shutdown event "
"registration failed\n");
}
/*
* Create the control device for doing management
*/
unit = device_get_unit(sc->mfi_dev);
sc->mfi_cdev = make_dev(&mfi_cdevsw, unit, UID_ROOT, GID_OPERATOR,
0640, "mfi%d", unit);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (unit == 0)
make_dev_alias(sc->mfi_cdev, "megaraid_sas_ioctl_node");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (sc->mfi_cdev != NULL)
sc->mfi_cdev->si_drv1 = sc;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
SYSCTL_ADD_INT(device_get_sysctl_ctx(sc->mfi_dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(sc->mfi_dev)),
OID_AUTO, "delete_busy_volumes", CTLFLAG_RW,
&sc->mfi_delete_busy_volumes, 0, "Allow removal of busy volumes");
SYSCTL_ADD_INT(device_get_sysctl_ctx(sc->mfi_dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(sc->mfi_dev)),
OID_AUTO, "keep_deleted_volumes", CTLFLAG_RW,
&sc->mfi_keep_deleted_volumes, 0,
"Don't detach the mfid device for a busy volume that is deleted");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
device_add_child(sc->mfi_dev, "mfip", -1);
bus_generic_attach(sc->mfi_dev);
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
/* Start the timeout watchdog */
Replace callout_init(..., 1) with callout_init(..., CALLOUT_MPSAFE) for better grep-compliance and to standardize with the rest of the kernel. Reviewed by: jhb MFC after: 1 week
2008-04-16 16:47:14 +00:00
callout_init(&sc->mfi_watchdog_callout, CALLOUT_MPSAFE);
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
callout_reset(&sc->mfi_watchdog_callout, mfi_cmd_timeout * hz,
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
mfi_timeout, sc);
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
if (sc->mfi_flags & MFI_FLAGS_TBOLT) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_lock(&sc->mfi_io_lock);
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
mfi_tbolt_sync_map_info(sc);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_unlock(&sc->mfi_io_lock);
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (0);
}
static int
mfi_alloc_commands(struct mfi_softc *sc)
{
struct mfi_command *cm;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int i, j;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* XXX Should we allocate all the commands up front, or allocate on
* demand later like 'aac' does?
*/
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
sc->mfi_commands = malloc(sizeof(sc->mfi_commands[0]) *
sc->mfi_max_fw_cmds, M_MFIBUF, M_WAITOK | M_ZERO);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
for (i = 0; i < sc->mfi_max_fw_cmds; i++) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm = &sc->mfi_commands[i];
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
cm->cm_frame = (union mfi_frame *)((uintptr_t)sc->mfi_frames +
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_cmd_size * i);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_frame_busaddr = sc->mfi_frames_busaddr +
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_cmd_size * i;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_frame->header.context = i;
cm->cm_sense = &sc->mfi_sense[i];
cm->cm_sense_busaddr= sc->mfi_sense_busaddr + MFI_SENSE_LEN * i;
cm->cm_sc = sc;
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
cm->cm_index = i;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (bus_dmamap_create(sc->mfi_buffer_dmat, 0,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
&cm->cm_dmamap) == 0) {
mtx_lock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(cm);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mtx_unlock(&sc->mfi_io_lock);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
} else {
device_printf(sc->mfi_dev, "Failed to allocate %d "
"command blocks, only allocated %d\n",
sc->mfi_max_fw_cmds, i - 1);
for (j = 0; j < i; j++) {
cm = &sc->mfi_commands[i];
bus_dmamap_destroy(sc->mfi_buffer_dmat,
cm->cm_dmamap);
}
free(sc->mfi_commands, M_MFIBUF);
sc->mfi_commands = NULL;
return (ENOMEM);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
return (0);
}
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
void
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(struct mfi_command *cm)
{
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
struct mfi_frame_header *hdr;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
uint32_t *hdr_data;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mtx_assert(&cm->cm_sc->mfi_io_lock, MA_OWNED);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Zero out the important fields of the frame, but make sure the
Make sure that all of the fields in the header are clean. It was possible for unsafe flags to leak from one command to another.
2007-01-02 04:12:34 +00:00
* context field is preserved. For efficiency, handle the fields
* as 32 bit words. Clear out the first S/G entry too for safety.
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
*/
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
hdr = &cm->cm_frame->header;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
if (cm->cm_data != NULL && hdr->sg_count) {
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
cm->cm_sg->sg32[0].len = 0;
cm->cm_sg->sg32[0].addr = 0;
}
Make sure that all of the fields in the header are clean. It was possible for unsafe flags to leak from one command to another.
2007-01-02 04:12:34 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
/*
* Command may be on other queues e.g. busy queue depending on the
* flow of a previous call to mfi_mapcmd, so ensure its dequeued
* properly
*/
if ((cm->cm_flags & MFI_ON_MFIQ_BUSY) != 0)
mfi_remove_busy(cm);
if ((cm->cm_flags & MFI_ON_MFIQ_READY) != 0)
mfi_remove_ready(cm);
/* We're not expecting it to be on any other queue but check */
if ((cm->cm_flags & MFI_ON_MFIQ_MASK) != 0) {
panic("Command %p is still on another queue, flags = %#x",
cm, cm->cm_flags);
}
/* tbolt cleanup */
if ((cm->cm_flags & MFI_CMD_TBOLT) != 0) {
mfi_tbolt_return_cmd(cm->cm_sc,
cm->cm_sc->mfi_cmd_pool_tbolt[cm->cm_extra_frames - 1],
cm);
}
Make sure that all of the fields in the header are clean. It was possible for unsafe flags to leak from one command to another.
2007-01-02 04:12:34 +00:00
hdr_data = (uint32_t *)cm->cm_frame;
hdr_data[0] = 0; /* cmd, sense_len, cmd_status, scsi_status */
hdr_data[1] = 0; /* target_id, lun_id, cdb_len, sg_count */
hdr_data[4] = 0; /* flags, timeout */
hdr_data[5] = 0; /* data_len */
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_extra_frames = 0;
cm->cm_flags = 0;
cm->cm_complete = NULL;
cm->cm_private = NULL;
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
cm->cm_data = NULL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_sg = 0;
cm->cm_total_frame_size = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->retry_for_fw_reset = 0;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_enqueue_free(cm);
}
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
int
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_dcmd_command(struct mfi_softc *sc, struct mfi_command **cmp,
uint32_t opcode, void **bufp, size_t bufsize)
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
{
struct mfi_command *cm;
struct mfi_dcmd_frame *dcmd;
void *buf = NULL;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
cm = mfi_dequeue_free(sc);
if (cm == NULL)
return (EBUSY);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Zero out the MFI frame */
context = cm->cm_frame->header.context;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(cm->cm_frame, sizeof(union mfi_frame));
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.context = context;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
if ((bufsize > 0) && (bufp != NULL)) {
if (*bufp == NULL) {
buf = malloc(bufsize, M_MFIBUF, M_NOWAIT|M_ZERO);
if (buf == NULL) {
mfi_release_command(cm);
return (ENOMEM);
}
*bufp = buf;
} else {
buf = *bufp;
}
}
dcmd = &cm->cm_frame->dcmd;
bzero(dcmd->mbox, MFI_MBOX_SIZE);
dcmd->header.cmd = MFI_CMD_DCMD;
dcmd->header.timeout = 0;
dcmd->header.flags = 0;
dcmd->header.data_len = bufsize;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
dcmd->header.scsi_status = 0;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
dcmd->opcode = opcode;
cm->cm_sg = &dcmd->sgl;
cm->cm_total_frame_size = MFI_DCMD_FRAME_SIZE;
cm->cm_flags = 0;
cm->cm_data = buf;
cm->cm_private = buf;
cm->cm_len = bufsize;
*cmp = cm;
if ((bufp != NULL) && (*bufp == NULL) && (buf != NULL))
*bufp = buf;
return (0);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int
mfi_comms_init(struct mfi_softc *sc)
{
struct mfi_command *cm;
struct mfi_init_frame *init;
struct mfi_init_qinfo *qinfo;
int error;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL) {
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (EBUSY);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Zero out the MFI frame */
context = cm->cm_frame->header.context;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(cm->cm_frame, sizeof(union mfi_frame));
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.context = context;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* Abuse the SG list area of the frame to hold the init_qinfo
* object;
*/
init = &cm->cm_frame->init;
qinfo = (struct mfi_init_qinfo *)((uintptr_t)init + MFI_FRAME_SIZE);
bzero(qinfo, sizeof(struct mfi_init_qinfo));
qinfo->rq_entries = sc->mfi_max_fw_cmds + 1;
qinfo->rq_addr_lo = sc->mfi_comms_busaddr +
offsetof(struct mfi_hwcomms, hw_reply_q);
qinfo->pi_addr_lo = sc->mfi_comms_busaddr +
offsetof(struct mfi_hwcomms, hw_pi);
qinfo->ci_addr_lo = sc->mfi_comms_busaddr +
offsetof(struct mfi_hwcomms, hw_ci);
init->header.cmd = MFI_CMD_INIT;
init->header.data_len = sizeof(struct mfi_init_qinfo);
init->qinfo_new_addr_lo = cm->cm_frame_busaddr + MFI_FRAME_SIZE;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
cm->cm_data = NULL;
cm->cm_flags = MFI_CMD_POLLED;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
device_printf(sc->mfi_dev, "failed to send init command\n");
mfi_release_command(cm);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
return (error);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
static int
mfi_get_controller_info(struct mfi_softc *sc)
{
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
struct mfi_command *cm = NULL;
struct mfi_ctrl_info *ci = NULL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
uint32_t max_sectors_1, max_sectors_2;
int error;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
mtx_lock(&sc->mfi_io_lock);
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_CTRL_GETINFO,
(void **)&ci, sizeof(*ci));
if (error)
goto out;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_POLLED;
if ((error = mfi_mapcmd(sc, cm)) != 0) {
device_printf(sc->mfi_dev, "Failed to get controller info\n");
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
sc->mfi_max_io = (sc->mfi_max_sge - 1) * PAGE_SIZE /
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
MFI_SECTOR_LEN;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
error = 0;
goto out;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
max_sectors_1 = (1 << ci->stripe_sz_ops.max) * ci->max_strips_per_io;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
max_sectors_2 = ci->max_request_size;
sc->mfi_max_io = min(max_sectors_1, max_sectors_2);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->disableOnlineCtrlReset =
ci->properties.OnOffProperties.disableOnlineCtrlReset;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
out:
if (ci)
free(ci, M_MFIBUF);
if (cm)
mfi_release_command(cm);
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (error);
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
mfi_get_log_state(struct mfi_softc *sc, struct mfi_evt_log_state **log_state)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
Fix a potential problem when mfi_get_log_state and only release a command if one was allocated. Also release the command in mfi_shutdown.
2006-06-20 23:08:35 +00:00
struct mfi_command *cm = NULL;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
int error;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_CTRL_EVENT_GETINFO,
(void **)log_state, sizeof(**log_state));
if (error)
goto out;
Fix a typo when getting the log_state and set the flags on the command to signify it is being polled and expecting data to be dma'd from the card.
2006-06-20 22:17:54 +00:00
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_POLLED;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0) {
Clean up some comments and device printf failures in mfi_get_log_state and mfi_get_controller_info.
2006-06-20 20:41:54 +00:00
device_printf(sc->mfi_dev, "Failed to get log state\n");
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
goto out;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
out:
Fix a potential problem when mfi_get_log_state and only release a command if one was allocated. Also release the command in mfi_shutdown.
2006-06-20 23:08:35 +00:00
if (cm)
mfi_release_command(cm);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
return (error);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
mfi_aen_setup(struct mfi_softc *sc, uint32_t seq_start)
{
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
struct mfi_evt_log_state *log_state = NULL;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
union mfi_evt class_locale;
int error = 0;
uint32_t seq;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
class_locale.members.reserved = 0;
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
class_locale.members.locale = mfi_event_locale;
There are a couple of structs in mfireg.h with members named 'class'. These cause problems when trying to include the header in a C++ project. Rename them to 'evt_class', and track the change in mfi and mfiutil. Submitted by: Mark Johnston Sponsored by: Sandvine Incorporated Reviewed by: jhb@ MFC after: 1 week
2011-06-02 00:43:16 +00:00
class_locale.members.evt_class = mfi_event_class;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (seq_start == 0) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_get_log_state(sc, &log_state)) != 0)
goto out;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->mfi_boot_seq_num = log_state->boot_seq_num;
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
/*
* Walk through any events that fired since the last
* shutdown.
*/
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_parse_entries(sc, log_state->shutdown_seq_num,
log_state->newest_seq_num)) != 0)
goto out;
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
seq = log_state->newest_seq_num;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
} else
seq = seq_start;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
error = mfi_aen_register(sc, seq, class_locale.word);
out:
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
free(log_state, M_MFIBUF);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
return (error);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_wait_command(struct mfi_softc *sc, struct mfi_command *cm)
{
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
cm->cm_complete = NULL;
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
/*
* MegaCli can issue a DCMD of 0. In this case do nothing
* and return 0 to it as status
*/
if (cm->cm_frame->dcmd.opcode == 0) {
cm->cm_frame->header.cmd_status = MFI_STAT_OK;
cm->cm_error = 0;
return (cm->cm_error);
}
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_enqueue_ready(cm);
mfi_startio(sc);
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
if ((cm->cm_flags & MFI_CMD_COMPLETED) == 0)
msleep(cm, &sc->mfi_io_lock, PRIBIO, "mfiwait", 0);
return (cm->cm_error);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
void
mfi_free(struct mfi_softc *sc)
{
struct mfi_command *cm;
int i;
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
callout_drain(&sc->mfi_watchdog_callout);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (sc->mfi_cdev != NULL)
destroy_dev(sc->mfi_cdev);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (sc->mfi_commands != NULL) {
for (i = 0; i < sc->mfi_max_fw_cmds; i++) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm = &sc->mfi_commands[i];
bus_dmamap_destroy(sc->mfi_buffer_dmat, cm->cm_dmamap);
}
free(sc->mfi_commands, M_MFIBUF);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
sc->mfi_commands = NULL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
if (sc->mfi_intr)
bus_teardown_intr(sc->mfi_dev, sc->mfi_irq, sc->mfi_intr);
if (sc->mfi_irq != NULL)
bus_release_resource(sc->mfi_dev, SYS_RES_IRQ, sc->mfi_irq_rid,
sc->mfi_irq);
if (sc->mfi_sense_busaddr != 0)
bus_dmamap_unload(sc->mfi_sense_dmat, sc->mfi_sense_dmamap);
if (sc->mfi_sense != NULL)
bus_dmamem_free(sc->mfi_sense_dmat, sc->mfi_sense,
sc->mfi_sense_dmamap);
if (sc->mfi_sense_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_sense_dmat);
if (sc->mfi_frames_busaddr != 0)
bus_dmamap_unload(sc->mfi_frames_dmat, sc->mfi_frames_dmamap);
if (sc->mfi_frames != NULL)
bus_dmamem_free(sc->mfi_frames_dmat, sc->mfi_frames,
sc->mfi_frames_dmamap);
if (sc->mfi_frames_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_frames_dmat);
if (sc->mfi_comms_busaddr != 0)
bus_dmamap_unload(sc->mfi_comms_dmat, sc->mfi_comms_dmamap);
if (sc->mfi_comms != NULL)
bus_dmamem_free(sc->mfi_comms_dmat, sc->mfi_comms,
sc->mfi_comms_dmamap);
if (sc->mfi_comms_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_comms_dmat);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* ThunderBolt contiguous memory free here */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->mfi_flags & MFI_FLAGS_TBOLT) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_tb_busaddr != 0)
bus_dmamap_unload(sc->mfi_tb_dmat, sc->mfi_tb_dmamap);
if (sc->request_message_pool != NULL)
bus_dmamem_free(sc->mfi_tb_dmat, sc->request_message_pool,
sc->mfi_tb_dmamap);
if (sc->mfi_tb_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_tb_dmat);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Version buffer memory free */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/* Start LSIP200113393 */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->verbuf_h_busaddr != 0)
bus_dmamap_unload(sc->verbuf_h_dmat, sc->verbuf_h_dmamap);
if (sc->verbuf != NULL)
bus_dmamem_free(sc->verbuf_h_dmat, sc->verbuf,
sc->verbuf_h_dmamap);
if (sc->verbuf_h_dmat != NULL)
bus_dma_tag_destroy(sc->verbuf_h_dmat);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/* End LSIP200113393 */
/* ThunderBolt INIT packet memory Free */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_tb_init_busaddr != 0)
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
bus_dmamap_unload(sc->mfi_tb_init_dmat,
sc->mfi_tb_init_dmamap);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_tb_init != NULL)
bus_dmamem_free(sc->mfi_tb_init_dmat, sc->mfi_tb_init,
sc->mfi_tb_init_dmamap);
if (sc->mfi_tb_init_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_tb_init_dmat);
/* ThunderBolt IOC Init Desc memory free here */
if (sc->mfi_tb_ioc_init_busaddr != 0)
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bus_dmamap_unload(sc->mfi_tb_ioc_init_dmat,
sc->mfi_tb_ioc_init_dmamap);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_tb_ioc_init_desc != NULL)
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bus_dmamem_free(sc->mfi_tb_ioc_init_dmat,
sc->mfi_tb_ioc_init_desc,
sc->mfi_tb_ioc_init_dmamap);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_tb_ioc_init_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_tb_ioc_init_dmat);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (sc->mfi_cmd_pool_tbolt != NULL) {
for (int i = 0; i < sc->mfi_max_fw_cmds; i++) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->mfi_cmd_pool_tbolt[i] != NULL) {
free(sc->mfi_cmd_pool_tbolt[i],
M_MFIBUF);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->mfi_cmd_pool_tbolt[i] = NULL;
}
}
free(sc->mfi_cmd_pool_tbolt, M_MFIBUF);
sc->mfi_cmd_pool_tbolt = NULL;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->request_desc_pool != NULL) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
free(sc->request_desc_pool, M_MFIBUF);
sc->request_desc_pool = NULL;
}
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (sc->mfi_buffer_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_buffer_dmat);
if (sc->mfi_parent_dmat != NULL)
bus_dma_tag_destroy(sc->mfi_parent_dmat);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
if (mtx_initialized(&sc->mfi_io_lock)) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mtx_destroy(&sc->mfi_io_lock);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_destroy(&sc->mfi_config_lock);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return;
}
static void
mfi_startup(void *arg)
{
struct mfi_softc *sc;
sc = (struct mfi_softc *)arg;
config_intrhook_disestablish(&sc->mfi_ich);
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
sc->mfi_enable_intr(sc);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_xlock(&sc->mfi_config_lock);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_ldprobe(sc);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->mfi_flags & MFI_FLAGS_SKINNY)
mfi_syspdprobe(sc);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_xunlock(&sc->mfi_config_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
static void
mfi_intr(void *arg)
{
struct mfi_softc *sc;
struct mfi_command *cm;
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
uint32_t pi, ci, context;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc = (struct mfi_softc *)arg;
Update the MFI driver to support new "1078" series of hardware. This includes the upcoming Dell PERC6 series. Many thanks to LSI for contributing this code. Submitted by: LSI Approved by: re
2007-08-25 23:58:45 +00:00
if (sc->mfi_check_clear_intr(sc))
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
restart:
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
pi = sc->mfi_comms->hw_pi;
ci = sc->mfi_comms->hw_ci;
mtx_lock(&sc->mfi_io_lock);
while (ci != pi) {
context = sc->mfi_comms->hw_reply_q[ci];
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
if (context < sc->mfi_max_fw_cmds) {
cm = &sc->mfi_commands[context];
mfi_remove_busy(cm);
cm->cm_error = 0;
mfi_complete(sc, cm);
}
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (++ci == (sc->mfi_max_fw_cmds + 1))
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
ci = 0;
}
sc->mfi_comms->hw_ci = ci;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
/* Give defered I/O a chance to run */
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
sc->mfi_flags &= ~MFI_FLAGS_QFRZN;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mfi_startio(sc);
mtx_unlock(&sc->mfi_io_lock);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/*
* Dummy read to flush the bus; this ensures that the indexes are up
* to date. Restart processing if more commands have come it.
*/
(void)sc->mfi_read_fw_status(sc);
if (pi != sc->mfi_comms->hw_pi)
goto restart;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return;
}
int
mfi_shutdown(struct mfi_softc *sc)
{
struct mfi_dcmd_frame *dcmd;
struct mfi_command *cm;
int error;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (sc->mfi_aen_cm != NULL) {
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
sc->cm_aen_abort = 1;
mfi_abort(sc, &sc->mfi_aen_cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (sc->mfi_map_sync_cm != NULL) {
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
sc->cm_map_abort = 1;
mfi_abort(sc, &sc->mfi_map_sync_cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
mtx_lock(&sc->mfi_io_lock);
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_CTRL_SHUTDOWN, NULL, 0);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
if (error) {
mtx_unlock(&sc->mfi_io_lock);
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
return (error);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
dcmd = &cm->cm_frame->dcmd;
dcmd->header.flags = MFI_FRAME_DIR_NONE;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
cm->cm_flags = MFI_CMD_POLLED;
cm->cm_data = NULL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
device_printf(sc->mfi_dev, "Failed to shutdown controller\n");
Fix a potential problem when mfi_get_log_state and only release a command if one was allocated. Also release the command in mfi_shutdown.
2006-06-20 23:08:35 +00:00
mfi_release_command(cm);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (error);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static void
mfi_syspdprobe(struct mfi_softc *sc)
{
struct mfi_frame_header *hdr;
struct mfi_command *cm = NULL;
struct mfi_pd_list *pdlist = NULL;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
struct mfi_system_pd *syspd, *tmp;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_system_pending *syspd_pend;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
int error, i, found;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sx_assert(&sc->mfi_config_lock, SA_XLOCKED);
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Add SYSTEM PD's */
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_PD_LIST_QUERY,
(void **)&pdlist, sizeof(*pdlist));
Some style improvements.
2012-05-04 16:22:13 +00:00
if (error) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Error while forming SYSTEM PD list\n");
goto out;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_POLLED;
cm->cm_frame->dcmd.mbox[0] = MR_PD_QUERY_TYPE_EXPOSED_TO_HOST;
cm->cm_frame->dcmd.mbox[1] = 0;
if (mfi_mapcmd(sc, cm) != 0) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Failed to get syspd device listing\n");
goto out;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
bus_dmamap_sync(sc->mfi_buffer_dmat,cm->cm_dmamap,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
BUS_DMASYNC_POSTREAD);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
hdr = &cm->cm_frame->header;
if (hdr->cmd_status != MFI_STAT_OK) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"MFI_DCMD_PD_LIST_QUERY failed %x\n", hdr->cmd_status);
goto out;
}
/* Get each PD and add it to the system */
for (i = 0; i < pdlist->count; i++) {
if (pdlist->addr[i].device_id ==
pdlist->addr[i].encl_device_id)
continue;
found = 0;
TAILQ_FOREACH(syspd, &sc->mfi_syspd_tqh, pd_link) {
if (syspd->pd_id == pdlist->addr[i].device_id)
found = 1;
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
TAILQ_FOREACH(syspd_pend, &sc->mfi_syspd_pend_tqh, pd_link) {
if (syspd_pend->pd_id == pdlist->addr[i].device_id)
found = 1;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (found == 0)
mfi_add_sys_pd(sc, pdlist->addr[i].device_id);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
/* Delete SYSPD's whose state has been changed */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
TAILQ_FOREACH_SAFE(syspd, &sc->mfi_syspd_tqh, pd_link, tmp) {
found = 0;
for (i = 0; i < pdlist->count; i++) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (syspd->pd_id == pdlist->addr[i].device_id) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
found = 1;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
break;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
}
if (found == 0) {
printf("DELETE\n");
mtx_unlock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mtx_lock(&Giant);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_delete_child(sc->mfi_dev, syspd->pd_dev);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mtx_unlock(&Giant);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mtx_lock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
}
out:
if (pdlist)
free(pdlist, M_MFIBUF);
if (cm)
mfi_release_command(cm);
return;
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static void
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_ldprobe(struct mfi_softc *sc)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_frame_header *hdr;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
struct mfi_command *cm = NULL;
struct mfi_ld_list *list = NULL;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
struct mfi_disk *ld;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_disk_pending *ld_pend;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
int error, i;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_assert(&sc->mfi_config_lock, SA_XLOCKED);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_LD_GET_LIST,
(void **)&list, sizeof(*list));
if (error)
goto out;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
cm->cm_flags = MFI_CMD_DATAIN;
if (mfi_wait_command(sc, cm) != 0) {
device_printf(sc->mfi_dev, "Failed to get device listing\n");
goto out;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
hdr = &cm->cm_frame->header;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
if (hdr->cmd_status != MFI_STAT_OK) {
device_printf(sc->mfi_dev, "MFI_DCMD_LD_GET_LIST failed %x\n",
hdr->cmd_status);
goto out;
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
for (i = 0; i < list->ld_count; i++) {
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
if (ld->ld_id == list->ld_list[i].ld.v.target_id)
goto skip_add;
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
TAILQ_FOREACH(ld_pend, &sc->mfi_ld_pend_tqh, ld_link) {
if (ld_pend->ld_id == list->ld_list[i].ld.v.target_id)
goto skip_add;
}
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mfi_add_ld(sc, list->ld_list[i].ld.v.target_id);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
skip_add:;
}
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
out:
if (list)
free(list, M_MFIBUF);
if (cm)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(cm);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
return;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
/*
* The timestamp is the number of seconds since 00:00 Jan 1, 2000. If
* the bits in 24-31 are all set, then it is the number of seconds since
* boot.
*/
static const char *
format_timestamp(uint32_t timestamp)
{
static char buffer[32];
if ((timestamp & 0xff000000) == 0xff000000)
snprintf(buffer, sizeof(buffer), "boot + %us", timestamp &
0x00ffffff);
else
snprintf(buffer, sizeof(buffer), "%us", timestamp);
return (buffer);
}
static const char *
format_class(int8_t class)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
static char buffer[6];
switch (class) {
case MFI_EVT_CLASS_DEBUG:
return ("debug");
case MFI_EVT_CLASS_PROGRESS:
return ("progress");
case MFI_EVT_CLASS_INFO:
return ("info");
case MFI_EVT_CLASS_WARNING:
return ("WARN");
case MFI_EVT_CLASS_CRITICAL:
return ("CRIT");
case MFI_EVT_CLASS_FATAL:
return ("FATAL");
case MFI_EVT_CLASS_DEAD:
return ("DEAD");
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
default:
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
snprintf(buffer, sizeof(buffer), "%d", class);
return (buffer);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
}
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
static void
mfi_decode_evt(struct mfi_softc *sc, struct mfi_evt_detail *detail)
{
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_system_pd *syspd = NULL;
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
Revert r200231. It was already taken cared by jhb long ago. Pointed out by: jhb Pointy hat: jkim
2009-12-07 21:24:07 +00:00
device_printf(sc->mfi_dev, "%d (%s/0x%04x/%s) - %s\n", detail->seq,
There are a couple of structs in mfireg.h with members named 'class'. These cause problems when trying to include the header in a C++ project. Rename them to 'evt_class', and track the change in mfi and mfiutil. Submitted by: Mark Johnston Sponsored by: Sandvine Incorporated Reviewed by: jhb@ MFC after: 1 week
2011-06-02 00:43:16 +00:00
format_timestamp(detail->time), detail->evt_class.members.locale,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
format_class(detail->evt_class.members.evt_class),
detail->description);
/* Don't act on old AEN's or while shutting down */
if (detail->seq < sc->mfi_boot_seq_num || sc->mfi_detaching)
return;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
switch (detail->arg_type) {
case MR_EVT_ARGS_NONE:
if (detail->code == MR_EVT_CTRL_HOST_BUS_SCAN_REQUESTED) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "HostBus scan raised\n");
if (mfi_detect_jbod_change) {
/*
* Probe for new SYSPD's and Delete
* invalid SYSPD's
*/
sx_xlock(&sc->mfi_config_lock);
mtx_lock(&sc->mfi_io_lock);
mfi_syspdprobe(sc);
mtx_unlock(&sc->mfi_io_lock);
sx_xunlock(&sc->mfi_config_lock);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
break;
case MR_EVT_ARGS_LD_STATE:
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/* During load time driver reads all the events starting
* from the one that has been logged after shutdown. Avoid
* these old events.
*/
if (detail->args.ld_state.new_state == MFI_LD_STATE_OFFLINE ) {
/* Remove the LD */
struct mfi_disk *ld;
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
if (ld->ld_id ==
detail->args.ld_state.ld.target_id)
break;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/*
Fix: for kernel panics when SSCD is removed
KASSERT(ld != NULL, ("volume dissappeared"));
*/
if (ld != NULL) {
mtx_lock(&Giant);
device_delete_child(sc->mfi_dev, ld->ld_dev);
mtx_unlock(&Giant);
}
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
case MR_EVT_ARGS_PD:
if (detail->code == MR_EVT_PD_REMOVED) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (mfi_detect_jbod_change) {
/*
* If the removed device is a SYSPD then
* delete it
*/
TAILQ_FOREACH(syspd, &sc->mfi_syspd_tqh,
pd_link) {
if (syspd->pd_id ==
detail->args.pd.device_id) {
mtx_lock(&Giant);
device_delete_child(
sc->mfi_dev,
syspd->pd_dev);
mtx_unlock(&Giant);
break;
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
if (detail->code == MR_EVT_PD_INSERTED) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (mfi_detect_jbod_change) {
/* Probe for new SYSPD's */
sx_xlock(&sc->mfi_config_lock);
mtx_lock(&sc->mfi_io_lock);
mfi_syspdprobe(sc);
mtx_unlock(&sc->mfi_io_lock);
sx_xunlock(&sc->mfi_config_lock);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
Add support for SCSI pass through devices to be attached and detached. PR: 172864 Submitted by: rstone@
2012-11-08 00:32:36 +00:00
if (sc->mfi_cam_rescan_cb != NULL &&
(detail->code == MR_EVT_PD_INSERTED ||
detail->code == MR_EVT_PD_REMOVED)) {
sc->mfi_cam_rescan_cb(sc, detail->args.pd.device_id);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
}
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static void
mfi_queue_evt(struct mfi_softc *sc, struct mfi_evt_detail *detail)
{
struct mfi_evt_queue_elm *elm;
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
elm = malloc(sizeof(*elm), M_MFIBUF, M_NOWAIT|M_ZERO);
if (elm == NULL)
return;
memcpy(&elm->detail, detail, sizeof(*detail));
TAILQ_INSERT_TAIL(&sc->mfi_evt_queue, elm, link);
taskqueue_enqueue(taskqueue_swi, &sc->mfi_evt_task);
}
static void
mfi_handle_evt(void *context, int pending)
{
TAILQ_HEAD(,mfi_evt_queue_elm) queue;
struct mfi_softc *sc;
struct mfi_evt_queue_elm *elm;
sc = context;
TAILQ_INIT(&queue);
mtx_lock(&sc->mfi_io_lock);
TAILQ_CONCAT(&queue, &sc->mfi_evt_queue, link);
mtx_unlock(&sc->mfi_io_lock);
while ((elm = TAILQ_FIRST(&queue)) != NULL) {
TAILQ_REMOVE(&queue, elm, link);
mfi_decode_evt(sc, &elm->detail);
free(elm, M_MFIBUF);
}
Tweak the output of event log messages from the controller: - Each log entry contains a text description in the "description" field of the entry. The existing decode logic always ended up duplicating information that was already in the description string. This made the logs overly verbose. Now we just print out the description string. - Add some simple parsing of the timestamp and event classes. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:36:38 +00:00
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int
mfi_aen_register(struct mfi_softc *sc, int seq, int locale)
{
struct mfi_command *cm;
struct mfi_dcmd_frame *dcmd;
union mfi_evt current_aen, prior_aen;
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
struct mfi_evt_detail *ed = NULL;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
int error = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
current_aen.word = locale;
if (sc->mfi_aen_cm != NULL) {
prior_aen.word =
((uint32_t *)&sc->mfi_aen_cm->cm_frame->dcmd.mbox)[1];
There are a couple of structs in mfireg.h with members named 'class'. These cause problems when trying to include the header in a C++ project. Rename them to 'evt_class', and track the change in mfi and mfiutil. Submitted by: Mark Johnston Sponsored by: Sandvine Incorporated Reviewed by: jhb@ MFC after: 1 week
2011-06-02 00:43:16 +00:00
if (prior_aen.members.evt_class <= current_aen.members.evt_class &&
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
!((prior_aen.members.locale & current_aen.members.locale)
^current_aen.members.locale)) {
return (0);
} else {
prior_aen.members.locale |= current_aen.members.locale;
There are a couple of structs in mfireg.h with members named 'class'. These cause problems when trying to include the header in a C++ project. Rename them to 'evt_class', and track the change in mfi and mfiutil. Submitted by: Mark Johnston Sponsored by: Sandvine Incorporated Reviewed by: jhb@ MFC after: 1 week
2011-06-02 00:43:16 +00:00
if (prior_aen.members.evt_class
< current_aen.members.evt_class)
current_aen.members.evt_class =
prior_aen.members.evt_class;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
mfi_abort(sc, &sc->mfi_aen_cm);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
}
Cleanup dcmd firmware processing into a single function, mfi_dcmd_command to avoid duplication and mistakes when setting up firmware commands for submission. Reviewed by: scottl
2006-06-20 21:06:05 +00:00
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_CTRL_EVENT_WAIT,
(void **)&ed, sizeof(*ed));
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (error)
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
goto out;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
dcmd = &cm->cm_frame->dcmd;
((uint32_t *)&dcmd->mbox)[0] = seq;
((uint32_t *)&dcmd->mbox)[1] = locale;
cm->cm_flags = MFI_CMD_DATAIN;
cm->cm_complete = mfi_aen_complete;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sc->last_seq_num = seq;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
sc->mfi_aen_cm = cm;
mfi_enqueue_ready(cm);
mfi_startio(sc);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
out:
return (error);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
static void
mfi_aen_complete(struct mfi_command *cm)
{
struct mfi_frame_header *hdr;
struct mfi_softc *sc;
struct mfi_evt_detail *detail;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
struct mfi_aen *mfi_aen_entry, *tmp;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
int seq = 0, aborted = 0;
sc = cm->cm_sc;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (sc->mfi_aen_cm == NULL)
return;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
hdr = &cm->cm_frame->header;
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
if (sc->cm_aen_abort ||
Use MFI_STAT_* constants for cmd_status field values rather than magic numbers in a few places. PR: bin/145960 Submitted by: gcooper MFC after: 1 week
2011-07-14 20:11:27 +00:00
hdr->cmd_status == MFI_STAT_INVALID_STATUS) {
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
sc->cm_aen_abort = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
aborted = 1;
} else {
sc->mfi_aen_triggered = 1;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
if (sc->mfi_poll_waiting) {
sc->mfi_poll_waiting = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
selwakeup(&sc->mfi_select);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
detail = cm->cm_data;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_queue_evt(sc, detail);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
seq = detail->seq + 1;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
TAILQ_FOREACH_SAFE(mfi_aen_entry, &sc->mfi_aen_pids, aen_link,
tmp) {
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
TAILQ_REMOVE(&sc->mfi_aen_pids, mfi_aen_entry,
aen_link);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
PROC_LOCK(mfi_aen_entry->p);
In order to maximize the re-usability of kernel code in user space this patch modifies makesyscalls.sh to prefix all of the non-compatibility calls (e.g. not linux_, freebsd32_) with sys_ and updates the kernel entry points and all places in the code that use them. It also fixes an additional name space collision between the kernel function psignal and the libc function of the same name by renaming the kernel psignal kern_psignal(). By introducing this change now we will ease future MFCs that change syscalls. Reviewed by: rwatson Approved by: re (bz)
2011-09-16 13:58:51 +00:00
kern_psignal(mfi_aen_entry->p, SIGIO);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
PROC_UNLOCK(mfi_aen_entry->p);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
free(mfi_aen_entry, M_MFIBUF);
}
}
free(cm->cm_data, M_MFIBUF);
wakeup(&sc->mfi_aen_cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
sc->mfi_aen_cm = NULL;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
mfi_release_command(cm);
/* set it up again so the driver can catch more events */
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (!aborted)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
mfi_aen_setup(sc, seq);
}
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
#define MAX_EVENTS 15
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
mfi_parse_entries(struct mfi_softc *sc, int start_seq, int stop_seq)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
struct mfi_command *cm;
struct mfi_dcmd_frame *dcmd;
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
struct mfi_evt_list *el;
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
union mfi_evt class_locale;
int error, i, seq, size;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
class_locale.members.reserved = 0;
class_locale.members.locale = mfi_event_locale;
There are a couple of structs in mfireg.h with members named 'class'. These cause problems when trying to include the header in a C++ project. Rename them to 'evt_class', and track the change in mfi and mfiutil. Submitted by: Mark Johnston Sponsored by: Sandvine Incorporated Reviewed by: jhb@ MFC after: 1 week
2011-06-02 00:43:16 +00:00
class_locale.members.evt_class = mfi_event_class;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
size = sizeof(struct mfi_evt_list) + sizeof(struct mfi_evt_detail)
* (MAX_EVENTS - 1);
el = malloc(size, M_MFIBUF, M_NOWAIT | M_ZERO);
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
if (el == NULL)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
return (ENOMEM);
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
for (seq = start_seq;;) {
if ((cm = mfi_dequeue_free(sc)) == NULL) {
free(el, M_MFIBUF);
return (EBUSY);
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
dcmd = &cm->cm_frame->dcmd;
bzero(dcmd->mbox, MFI_MBOX_SIZE);
dcmd->header.cmd = MFI_CMD_DCMD;
dcmd->header.timeout = 0;
dcmd->header.data_len = size;
dcmd->opcode = MFI_DCMD_CTRL_EVENT_GET;
((uint32_t *)&dcmd->mbox)[0] = seq;
((uint32_t *)&dcmd->mbox)[1] = class_locale.word;
cm->cm_sg = &dcmd->sgl;
cm->cm_total_frame_size = MFI_DCMD_FRAME_SIZE;
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_POLLED;
cm->cm_data = el;
cm->cm_len = size;
if ((error = mfi_mapcmd(sc, cm)) != 0) {
device_printf(sc->mfi_dev,
"Failed to get controller entries\n");
mfi_release_command(cm);
break;
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
if (dcmd->header.cmd_status == MFI_STAT_NOT_FOUND) {
mfi_release_command(cm);
break;
}
if (dcmd->header.cmd_status != MFI_STAT_OK) {
device_printf(sc->mfi_dev,
"Error %d fetching controller entries\n",
dcmd->header.cmd_status);
mfi_release_command(cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
error = EIO;
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
break;
}
mfi_release_command(cm);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Allow hw.mfi.event_locale/hw.mfi.event_class to be set via loader. If an event doesn't match the criteria then don't print it. Some events are not saved in the log (<0 class events).
2006-09-20 18:49:35 +00:00
for (i = 0; i < el->count; i++) {
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
/*
* If this event is newer than 'stop_seq' then
* break out of the loop. Note that the log
* is a circular buffer so we have to handle
* the case that our stop point is earlier in
* the buffer than our start point.
*/
if (el->event[i].seq >= stop_seq) {
if (start_seq <= stop_seq)
break;
else if (el->event[i].seq < start_seq)
break;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_queue_evt(sc, &el->event[i]);
Allow hw.mfi.event_locale/hw.mfi.event_class to be set via loader. If an event doesn't match the criteria then don't print it. Some events are not saved in the log (<0 class events).
2006-09-20 18:49:35 +00:00
}
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
seq = el->event[el->count - 1].seq + 1;
Change the event log dump on initial boot to use get_event versus AEN. This makes the boot messages cleaner. I now know how this structure works so I can implement it versus guessing. Remove the not ready type code since it is ready now. I added the time stamp/locale/class so people can parse messages better. Create a sysctl so that we can set the locale/class level.
2006-09-07 18:40:49 +00:00
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Adjust the handling of pending log events during boot: - Fetch events from the controller in batches of 15 rather than a single event at a time. - When fetching events from the controller, honor the event class and locale settings (via hw.mfi tunables). This also allows the firmware to skip over unwanted log entries resulting in fewer requests to the controller if there many unwanted log entries since the last clean shutdown. - Don't drop the driver mutex while decoding an event. - If we get an error other than MFI_STAT_NOT_FOUND (basically EOF for hitting the end of the event log) then emit a warning and bail on processing further log entries. Reviewed by: ambrisko, scottl MFC after: 2 weeks
2008-06-26 22:33:24 +00:00
free(el, M_MFIBUF);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
return (error);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_add_ld(struct mfi_softc *sc, int id)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_command *cm;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
struct mfi_dcmd_frame *dcmd = NULL;
struct mfi_ld_info *ld_info = NULL;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_disk_pending *ld_pend;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
int error;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
ld_pend = malloc(sizeof(*ld_pend), M_MFIBUF, M_NOWAIT | M_ZERO);
if (ld_pend != NULL) {
ld_pend->ld_id = id;
TAILQ_INSERT_TAIL(&sc->mfi_ld_pend_tqh, ld_pend, ld_link);
}
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_LD_GET_INFO,
(void **)&ld_info, sizeof(*ld_info));
if (error) {
device_printf(sc->mfi_dev,
"Failed to allocate for MFI_DCMD_LD_GET_INFO %d\n", error);
if (ld_info)
free(ld_info, M_MFIBUF);
return (error);
Fix some small bugs. Submitted by: pjd Found by: Coverity Prevent (tm)
2006-04-10 06:44:30 +00:00
}
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
cm->cm_flags = MFI_CMD_DATAIN;
dcmd = &cm->cm_frame->dcmd;
dcmd->mbox[0] = id;
Change mfi_add_ld to "immediate command" mode since we need to enumerate potential boot disks during the probe so they are read for mount root. Reviewed by: ps, scottl
2006-06-30 18:59:08 +00:00
if (mfi_wait_command(sc, cm) != 0) {
device_printf(sc->mfi_dev,
"Failed to get logical drive: %d\n", id);
free(ld_info, M_MFIBUF);
return (0);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (ld_info->ld_config.params.isSSCD != 1)
mfi_add_ld_complete(cm);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
else {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_release_command(cm);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (ld_info) /* SSCD drives ld_info free here */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
free(ld_info, M_MFIBUF);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (0);
}
static void
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
mfi_add_ld_complete(struct mfi_command *cm)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_frame_header *hdr;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
struct mfi_ld_info *ld_info;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
struct mfi_softc *sc;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
device_t child;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc = cm->cm_sc;
hdr = &cm->cm_frame->header;
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
ld_info = cm->cm_private;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
if (sc->cm_map_abort || hdr->cmd_status != MFI_STAT_OK) {
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
free(ld_info, M_MFIBUF);
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
wakeup(&sc->mfi_map_sync_cm);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(cm);
return;
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
wakeup(&sc->mfi_map_sync_cm);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(cm);
Handle invalid capacity parameters from the firmware.
2006-03-28 01:59:11 +00:00
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
mtx_unlock(&sc->mfi_io_lock);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_lock(&Giant);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if ((child = device_add_child(sc->mfi_dev, "mfid", -1)) == NULL) {
device_printf(sc->mfi_dev, "Failed to add logical disk\n");
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
free(ld_info, M_MFIBUF);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_unlock(&Giant);
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
mtx_lock(&sc->mfi_io_lock);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
return;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
Collapse the mfi_ld object. Add an ioctl to help management apps map array Id's to FreeBSD device names.
2007-05-10 15:33:41 +00:00
device_set_ivars(child, ld_info);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
device_set_desc(child, "MFI Logical Disk");
bus_generic_attach(sc->mfi_dev);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_unlock(&Giant);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mtx_lock(&sc->mfi_io_lock);
}
Remove some debugging code that accidentally crept in.
2006-10-16 04:30:09 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
static int mfi_add_sys_pd(struct mfi_softc *sc, int id)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_command *cm;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_dcmd_frame *dcmd = NULL;
struct mfi_pd_info *pd_info = NULL;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_system_pending *syspd_pend;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int error;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
syspd_pend = malloc(sizeof(*syspd_pend), M_MFIBUF, M_NOWAIT | M_ZERO);
if (syspd_pend != NULL) {
syspd_pend->pd_id = id;
TAILQ_INSERT_TAIL(&sc->mfi_syspd_pend_tqh, syspd_pend, pd_link);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
error = mfi_dcmd_command(sc, &cm, MFI_DCMD_PD_GET_INFO,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
(void **)&pd_info, sizeof(*pd_info));
if (error) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Failed to allocated for MFI_DCMD_PD_GET_INFO %d\n",
error);
if (pd_info)
free(pd_info, M_MFIBUF);
return (error);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_POLLED;
dcmd = &cm->cm_frame->dcmd;
dcmd->mbox[0]=id;
dcmd->header.scsi_status = 0;
dcmd->header.pad0 = 0;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Failed to get physical drive info %d\n", id);
free(pd_info, M_MFIBUF);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mfi_release_command(cm);
return (error);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
}
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_add_sys_pd_complete(cm);
return (0);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static void
mfi_add_sys_pd_complete(struct mfi_command *cm)
{
struct mfi_frame_header *hdr;
struct mfi_pd_info *pd_info;
struct mfi_softc *sc;
device_t child;
sc = cm->cm_sc;
hdr = &cm->cm_frame->header;
pd_info = cm->cm_private;
if (hdr->cmd_status != MFI_STAT_OK) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
free(pd_info, M_MFIBUF);
mfi_release_command(cm);
return;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
if (pd_info->fw_state != MFI_PD_STATE_SYSTEM) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "PD=%x is not SYSTEM PD\n",
pd_info->ref.v.device_id);
free(pd_info, M_MFIBUF);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_release_command(cm);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
return;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
mfi_release_command(cm);
mtx_unlock(&sc->mfi_io_lock);
mtx_lock(&Giant);
if ((child = device_add_child(sc->mfi_dev, "mfisyspd", -1)) == NULL) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "Failed to add system pd\n");
free(pd_info, M_MFIBUF);
mtx_unlock(&Giant);
mtx_lock(&sc->mfi_io_lock);
return;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
device_set_ivars(child, pd_info);
device_set_desc(child, "MFI System PD");
bus_generic_attach(sc->mfi_dev);
mtx_unlock(&Giant);
mtx_lock(&sc->mfi_io_lock);
}
Some style improvements.
2012-05-04 16:22:13 +00:00
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static struct mfi_command *
mfi_bio_command(struct mfi_softc *sc)
{
struct bio *bio;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_command *cm = NULL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/*reserving two commands to avoid starvation for IOCTL*/
Some style improvements.
2012-05-04 16:22:13 +00:00
if (sc->mfi_qstat[MFIQ_FREE].q_length < 2) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (NULL);
}
if ((bio = mfi_dequeue_bio(sc)) == NULL) {
return (NULL);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((uintptr_t)bio->bio_driver2 == MFI_LD_IO) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
cm = mfi_build_ldio(sc, bio);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
} else if ((uintptr_t) bio->bio_driver2 == MFI_SYS_PD_IO) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
cm = mfi_build_syspdio(sc, bio);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
if (!cm)
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_enqueue_bio(sc, bio);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return cm;
}
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
/*
* mostly copied from cam/scsi/scsi_all.c:scsi_read_write
*/
int
mfi_build_cdb(int readop, uint8_t byte2, u_int64_t lba, u_int32_t block_count, uint8_t *cdb)
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
{
int cdb_len;
if (((lba & 0x1fffff) == lba)
&& ((block_count & 0xff) == block_count)
&& (byte2 == 0)) {
/* We can fit in a 6 byte cdb */
struct scsi_rw_6 *scsi_cmd;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
scsi_cmd = (struct scsi_rw_6 *)cdb;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
scsi_cmd->opcode = readop ? READ_6 : WRITE_6;
scsi_ulto3b(lba, scsi_cmd->addr);
scsi_cmd->length = block_count & 0xff;
scsi_cmd->control = 0;
cdb_len = sizeof(*scsi_cmd);
} else if (((block_count & 0xffff) == block_count) && ((lba & 0xffffffff) == lba)) {
/* Need a 10 byte CDB */
struct scsi_rw_10 *scsi_cmd;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
scsi_cmd = (struct scsi_rw_10 *)cdb;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
scsi_cmd->opcode = readop ? READ_10 : WRITE_10;
scsi_cmd->byte2 = byte2;
scsi_ulto4b(lba, scsi_cmd->addr);
scsi_cmd->reserved = 0;
scsi_ulto2b(block_count, scsi_cmd->length);
scsi_cmd->control = 0;
cdb_len = sizeof(*scsi_cmd);
} else if (((block_count & 0xffffffff) == block_count) &&
((lba & 0xffffffff) == lba)) {
/* Block count is too big for 10 byte CDB use a 12 byte CDB */
struct scsi_rw_12 *scsi_cmd;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
scsi_cmd = (struct scsi_rw_12 *)cdb;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
scsi_cmd->opcode = readop ? READ_12 : WRITE_12;
scsi_cmd->byte2 = byte2;
scsi_ulto4b(lba, scsi_cmd->addr);
scsi_cmd->reserved = 0;
scsi_ulto4b(block_count, scsi_cmd->length);
scsi_cmd->control = 0;
cdb_len = sizeof(*scsi_cmd);
} else {
/*
* 16 byte CDB. We'll only get here if the LBA is larger
* than 2^32
*/
struct scsi_rw_16 *scsi_cmd;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
scsi_cmd = (struct scsi_rw_16 *)cdb;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
scsi_cmd->opcode = readop ? READ_16 : WRITE_16;
scsi_cmd->byte2 = byte2;
scsi_u64to8b(lba, scsi_cmd->addr);
scsi_cmd->reserved = 0;
scsi_ulto4b(block_count, scsi_cmd->length);
scsi_cmd->control = 0;
cdb_len = sizeof(*scsi_cmd);
}
return cdb_len;
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static struct mfi_command *
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_build_syspdio(struct mfi_softc *sc, struct bio *bio)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
{
struct mfi_command *cm;
struct mfi_pass_frame *pass;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
uint32_t context = 0;
int flags = 0, blkcount = 0, readop;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
uint8_t cdb_len;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL)
return (NULL);
/* Zero out the MFI frame */
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
context = cm->cm_frame->header.context;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(cm->cm_frame, sizeof(union mfi_frame));
cm->cm_frame->header.context = context;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass = &cm->cm_frame->pass;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(pass->cdb, 16);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass->header.cmd = MFI_CMD_PD_SCSI_IO;
switch (bio->bio_cmd & 0x03) {
case BIO_READ:
flags = MFI_CMD_DATAIN;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
readop = 1;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
case BIO_WRITE:
flags = MFI_CMD_DATAOUT;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
readop = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
default:
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
/* TODO: what about BIO_DELETE??? */
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
panic("Unsupported bio command %x\n", bio->bio_cmd);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
/* Cheat with the sector length to avoid a non-constant division */
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
blkcount = (bio->bio_bcount + MFI_SECTOR_LEN - 1) / MFI_SECTOR_LEN;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Fill the LBA and Transfer length in CDB */
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
cdb_len = mfi_build_cdb(readop, 0, bio->bio_pblkno, blkcount,
pass->cdb);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass->header.target_id = (uintptr_t)bio->bio_driver1;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
pass->header.lun_id = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass->header.timeout = 0;
pass->header.flags = 0;
pass->header.scsi_status = 0;
pass->header.sense_len = MFI_SENSE_LEN;
pass->header.data_len = bio->bio_bcount;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
pass->header.cdb_len = cdb_len;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
pass->sense_addr_lo = (uint32_t)cm->cm_sense_busaddr;
pass->sense_addr_hi = (uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_complete = mfi_bio_complete;
cm->cm_private = bio;
cm->cm_data = bio->bio_data;
cm->cm_len = bio->bio_bcount;
cm->cm_sg = &pass->sgl;
cm->cm_total_frame_size = MFI_PASS_FRAME_SIZE;
cm->cm_flags = flags;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return (cm);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static struct mfi_command *
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_build_ldio(struct mfi_softc *sc, struct bio *bio)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
{
struct mfi_io_frame *io;
struct mfi_command *cm;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
int flags;
uint32_t blkcount;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL)
return (NULL);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Zero out the MFI frame */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
context = cm->cm_frame->header.context;
bzero(cm->cm_frame, sizeof(union mfi_frame));
cm->cm_frame->header.context = context;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io = &cm->cm_frame->io;
switch (bio->bio_cmd & 0x03) {
case BIO_READ:
io->header.cmd = MFI_CMD_LD_READ;
flags = MFI_CMD_DATAIN;
break;
case BIO_WRITE:
io->header.cmd = MFI_CMD_LD_WRITE;
flags = MFI_CMD_DATAOUT;
break;
default:
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
/* TODO: what about BIO_DELETE??? */
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
panic("Unsupported bio command %x\n", bio->bio_cmd);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
/* Cheat with the sector length to avoid a non-constant division */
blkcount = (bio->bio_bcount + MFI_SECTOR_LEN - 1) / MFI_SECTOR_LEN;
io->header.target_id = (uintptr_t)bio->bio_driver1;
io->header.timeout = 0;
io->header.flags = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
io->header.scsi_status = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io->header.sense_len = MFI_SENSE_LEN;
io->header.data_len = blkcount;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
io->sense_addr_lo = (uint32_t)cm->cm_sense_busaddr;
io->sense_addr_hi = (uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io->lba_hi = (bio->bio_pblkno & 0xffffffff00000000) >> 32;
io->lba_lo = bio->bio_pblkno & 0xffffffff;
cm->cm_complete = mfi_bio_complete;
cm->cm_private = bio;
cm->cm_data = bio->bio_data;
cm->cm_len = bio->bio_bcount;
cm->cm_sg = &io->sgl;
cm->cm_total_frame_size = MFI_IO_FRAME_SIZE;
cm->cm_flags = flags;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (cm);
}
static void
mfi_bio_complete(struct mfi_command *cm)
{
struct bio *bio;
struct mfi_frame_header *hdr;
struct mfi_softc *sc;
bio = cm->cm_private;
hdr = &cm->cm_frame->header;
sc = cm->cm_sc;
Use MFI_STAT_* constants for cmd_status field values rather than magic numbers in a few places. PR: bin/145960 Submitted by: gcooper MFC after: 1 week
2011-07-14 20:11:27 +00:00
if ((hdr->cmd_status != MFI_STAT_OK) || (hdr->scsi_status != 0)) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
bio->bio_flags |= BIO_ERROR;
bio->bio_error = EIO;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
device_printf(sc->mfi_dev, "I/O error, cmd=%p, status=%#x, "
"scsi_status=%#x\n", cm, hdr->cmd_status, hdr->scsi_status);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_print_sense(cm->cm_sc, cm->cm_sense);
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
} else if (cm->cm_error != 0) {
bio->bio_flags |= BIO_ERROR;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
bio->bio_error = cm->cm_error;
device_printf(sc->mfi_dev, "I/O error, cmd=%p, error=%#x\n",
cm, cm->cm_error);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
mfi_release_command(cm);
mfi_disk_complete(bio);
}
void
mfi_startio(struct mfi_softc *sc)
{
struct mfi_command *cm;
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
struct ccb_hdr *ccbh;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
for (;;) {
/* Don't bother if we're short on resources */
if (sc->mfi_flags & MFI_FLAGS_QFRZN)
break;
/* Try a command that has already been prepared */
cm = mfi_dequeue_ready(sc);
Catch up to driver changes required for the mfip sub-driver.
2007-05-16 17:23:36 +00:00
if (cm == NULL) {
if ((ccbh = TAILQ_FIRST(&sc->mfi_cam_ccbq)) != NULL)
cm = sc->mfi_cam_start(ccbh);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/* Nope, so look for work on the bioq */
if (cm == NULL)
cm = mfi_bio_command(sc);
/* No work available, so exit */
if (cm == NULL)
break;
/* Send the command to the controller */
if (mfi_mapcmd(sc, cm) != 0) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
device_printf(sc->mfi_dev, "Failed to startio\n");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_requeue_ready(cm);
break;
}
}
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_mapcmd(struct mfi_softc *sc, struct mfi_command *cm)
{
int error, polled;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((cm->cm_data != NULL) && (cm->cm_frame->header.cmd != MFI_CMD_STP )) {
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
polled = (cm->cm_flags & MFI_CMD_POLLED) ? BUS_DMA_NOWAIT : 0;
Reform the busdma API so that new types may be added without modifying every architecture's busdma_machdep.c. It is done by unifying the bus_dmamap_load_buffer() routines so that they may be called from MI code. The MD busdma is then given a chance to do any final processing in the complete() callback. The cam changes unify the bus_dmamap_load* handling in cam drivers. The arm and mips implementations are updated to track virtual addresses for sync(). Previously this was done in a type specific way. Now it is done in a generic way by recording the list of virtuals in the map. Submitted by: jeff (sponsored by EMC/Isilon) Reviewed by: kan (previous version), scottl, mjacob (isp(4), no objections for target mode changes) Discussed with: ian (arm changes) Tested by: marius (sparc64), mips (jmallet), isci(4) on x86 (jharris), amd64 (Fabian Keil <freebsd-listen@fabiankeil.de>)
2013-02-12 16:57:20 +00:00
if (cm->cm_flags & MFI_CMD_CCB)
error = bus_dmamap_load_ccb(sc->mfi_buffer_dmat,
cm->cm_dmamap, cm->cm_data, mfi_data_cb, cm,
polled);
else
error = bus_dmamap_load(sc->mfi_buffer_dmat,
cm->cm_dmamap, cm->cm_data, cm->cm_len,
mfi_data_cb, cm, polled);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (error == EINPROGRESS) {
sc->mfi_flags |= MFI_FLAGS_QFRZN;
return (0);
}
} else {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
error = mfi_send_frame(sc, cm);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
return (error);
}
static void
mfi_data_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
{
struct mfi_frame_header *hdr;
struct mfi_command *cm;
Fix 64-bit DMA. The problem was an incorrect flag check. Thanks to Paul Saab for helping to track this down. Fix a error with 32bit DMA size calculation that seemed to be harmless. Add a few micro-optimizations while I'm here.
2006-03-28 23:59:07 +00:00
union mfi_sgl *sgl;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
struct mfi_softc *sc;
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
int i, j, first, dir;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int sge_size, locked;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm = (struct mfi_command *)arg;
sc = cm->cm_sc;
Fix 64-bit DMA. The problem was an incorrect flag check. Thanks to Paul Saab for helping to track this down. Fix a error with 32bit DMA size calculation that seemed to be harmless. Add a few micro-optimizations while I'm here.
2006-03-28 23:59:07 +00:00
hdr = &cm->cm_frame->header;
sgl = cm->cm_sg;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
/*
* We need to check if we have the lock as this is async
* callback so even though our caller mfi_mapcmd asserts
* it has the lock, there is no garantee that hasn't been
* dropped if bus_dmamap_load returned prior to our
* completion.
*/
if ((locked = mtx_owned(&sc->mfi_io_lock)) == 0)
mtx_lock(&sc->mfi_io_lock);
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
if (error) {
printf("error %d in callback\n", error);
cm->cm_error = error;
mfi_complete(sc, cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
goto out;
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Use IEEE sgl only for IO's on a SKINNY controller
* For other commands on a SKINNY controller use either
* sg32 or sg64 based on the sizeof(bus_addr_t).
* Also calculate the total frame size based on the type
* of SGL used.
*/
if (((cm->cm_frame->header.cmd == MFI_CMD_PD_SCSI_IO) ||
(cm->cm_frame->header.cmd == MFI_CMD_LD_READ) ||
(cm->cm_frame->header.cmd == MFI_CMD_LD_WRITE)) &&
(sc->mfi_flags & MFI_FLAGS_SKINNY)) {
Fix 64-bit DMA. The problem was an incorrect flag check. Thanks to Paul Saab for helping to track this down. Fix a error with 32bit DMA size calculation that seemed to be harmless. Add a few micro-optimizations while I'm here.
2006-03-28 23:59:07 +00:00
for (i = 0; i < nsegs; i++) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sgl->sg_skinny[i].addr = segs[i].ds_addr;
sgl->sg_skinny[i].len = segs[i].ds_len;
sgl->sg_skinny[i].flag = 0;
Fix 64-bit DMA. The problem was an incorrect flag check. Thanks to Paul Saab for helping to track this down. Fix a error with 32bit DMA size calculation that seemed to be harmless. Add a few micro-optimizations while I'm here.
2006-03-28 23:59:07 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
hdr->flags |= MFI_FRAME_IEEE_SGL | MFI_FRAME_SGL64;
sge_size = sizeof(struct mfi_sg_skinny);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
hdr->sg_count = nsegs;
Fix 64-bit DMA. The problem was an incorrect flag check. Thanks to Paul Saab for helping to track this down. Fix a error with 32bit DMA size calculation that seemed to be harmless. Add a few micro-optimizations while I'm here.
2006-03-28 23:59:07 +00:00
} else {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
j = 0;
if (cm->cm_frame->header.cmd == MFI_CMD_STP) {
first = cm->cm_stp_len;
if ((sc->mfi_flags & MFI_FLAGS_SG64) == 0) {
sgl->sg32[j].addr = segs[0].ds_addr;
sgl->sg32[j++].len = first;
} else {
sgl->sg64[j].addr = segs[0].ds_addr;
sgl->sg64[j++].len = first;
}
} else
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
first = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((sc->mfi_flags & MFI_FLAGS_SG64) == 0) {
for (i = 0; i < nsegs; i++) {
sgl->sg32[j].addr = segs[i].ds_addr + first;
sgl->sg32[j++].len = segs[i].ds_len - first;
first = 0;
}
} else {
for (i = 0; i < nsegs; i++) {
sgl->sg64[j].addr = segs[i].ds_addr + first;
sgl->sg64[j++].len = segs[i].ds_len - first;
first = 0;
}
hdr->flags |= MFI_FRAME_SGL64;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
hdr->sg_count = j;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sge_size = sc->mfi_sge_size;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
dir = 0;
if (cm->cm_flags & MFI_CMD_DATAIN) {
dir |= BUS_DMASYNC_PREREAD;
hdr->flags |= MFI_FRAME_DIR_READ;
}
if (cm->cm_flags & MFI_CMD_DATAOUT) {
dir |= BUS_DMASYNC_PREWRITE;
hdr->flags |= MFI_FRAME_DIR_WRITE;
}
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap, dir);
cm->cm_flags |= MFI_CMD_MAPPED;
/*
* Instead of calculating the total number of frames in the
* compound frame, it's already assumed that there will be at
* least 1 frame, so don't compensate for the modulo of the
* following division.
*/
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
cm->cm_total_frame_size += (sc->mfi_sge_size * nsegs);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
cm->cm_extra_frames = (cm->cm_total_frame_size - 1) / MFI_FRAME_SIZE;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_send_frame(sc, cm)) != 0) {
printf("error %d in callback from mfi_send_frame\n", error);
cm->cm_error = error;
mfi_complete(sc, cm);
goto out;
}
out:
/* leave the lock in the state we found it */
if (locked == 0)
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return;
}
static int
mfi_send_frame(struct mfi_softc *sc, struct mfi_command *cm)
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
{
int error;
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
if (sc->MFA_enabled)
error = mfi_tbolt_send_frame(sc, cm);
else
error = mfi_std_send_frame(sc, cm);
if (error != 0 && (cm->cm_flags & MFI_ON_MFIQ_BUSY) != 0)
mfi_remove_busy(cm);
return (error);
}
static int
mfi_std_send_frame(struct mfi_softc *sc, struct mfi_command *cm)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
struct mfi_frame_header *hdr;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int tm = mfi_polled_cmd_timeout * 1000;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
hdr = &cm->cm_frame->header;
if ((cm->cm_flags & MFI_CMD_POLLED) == 0) {
cm->cm_timestamp = time_uptime;
mfi_enqueue_busy(cm);
} else {
Use MFI_STAT_* constants for cmd_status field values rather than magic numbers in a few places. PR: bin/145960 Submitted by: gcooper MFC after: 1 week
2011-07-14 20:11:27 +00:00
hdr->cmd_status = MFI_STAT_INVALID_STATUS;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
hdr->flags |= MFI_FRAME_DONT_POST_IN_REPLY_QUEUE;
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
/*
* The bus address of the command is aligned on a 64 byte boundary,
* leaving the least 6 bits as zero. For whatever reason, the
* hardware wants the address shifted right by three, leaving just
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
* 3 zero bits. These three bits are then used as a prefetching
* hint for the hardware to predict how many frames need to be
* fetched across the bus. If a command has more than 8 frames
* then the 3 bits are set to 0x7 and the firmware uses other
* information in the command to determine the total amount to fetch.
* However, FreeBSD doesn't support I/O larger than 128K, so 8 frames
* is enough for both 32bit and 64bit systems.
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
*/
Change some variable names and update some comments to help clarify some confusing issues.
2006-09-20 06:58:02 +00:00
if (cm->cm_extra_frames > 7)
cm->cm_extra_frames = 7;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
sc->mfi_issue_cmd(sc, cm->cm_frame_busaddr, cm->cm_extra_frames);
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
if ((cm->cm_flags & MFI_CMD_POLLED) == 0)
return (0);
/* This is a polled command, so busy-wait for it to complete. */
Use MFI_STAT_* constants for cmd_status field values rather than magic numbers in a few places. PR: bin/145960 Submitted by: gcooper MFC after: 1 week
2011-07-14 20:11:27 +00:00
while (hdr->cmd_status == MFI_STAT_INVALID_STATUS) {
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
DELAY(1000);
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
tm -= 1;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
if (tm <= 0)
break;
}
Use MFI_STAT_* constants for cmd_status field values rather than magic numbers in a few places. PR: bin/145960 Submitted by: gcooper MFC after: 1 week
2011-07-14 20:11:27 +00:00
if (hdr->cmd_status == MFI_STAT_INVALID_STATUS) {
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
device_printf(sc->mfi_dev, "Frame %p timed out "
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
"command 0x%X\n", hdr, cm->cm_frame->dcmd.opcode);
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
return (ETIMEDOUT);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (0);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
void
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
mfi_complete(struct mfi_softc *sc, struct mfi_command *cm)
{
int dir;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if ((cm->cm_flags & MFI_CMD_MAPPED) != 0) {
dir = 0;
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
if ((cm->cm_flags & MFI_CMD_DATAIN) ||
(cm->cm_frame->header.cmd == MFI_CMD_STP))
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
dir |= BUS_DMASYNC_POSTREAD;
if (cm->cm_flags & MFI_CMD_DATAOUT)
dir |= BUS_DMASYNC_POSTWRITE;
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap, dir);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
cm->cm_flags &= ~MFI_CMD_MAPPED;
}
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
cm->cm_flags |= MFI_CMD_COMPLETED;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if (cm->cm_complete != NULL)
cm->cm_complete(cm);
Instead of using scsi probes to do device discovery, use the firmware commands to grab the device listing. This resolves issues using multiple volumes, where each volume was actually internally pointing to target 0.
2006-06-20 22:41:44 +00:00
else
wakeup(cm);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
mfi_abort(struct mfi_softc *sc, struct mfi_command **cm_abort)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
struct mfi_command *cm;
struct mfi_abort_frame *abort;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
int i = 0, error;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
mtx_lock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
return (EBUSY);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Zero out the MFI frame */
context = cm->cm_frame->header.context;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(cm->cm_frame, sizeof(union mfi_frame));
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.context = context;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
abort = &cm->cm_frame->abort;
abort->header.cmd = MFI_CMD_ABORT;
abort->header.flags = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
abort->header.scsi_status = 0;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
abort->abort_context = (*cm_abort)->cm_frame->header.context;
abort->abort_mfi_addr_lo = (uint32_t)(*cm_abort)->cm_frame_busaddr;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
abort->abort_mfi_addr_hi =
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
(uint32_t)((uint64_t)(*cm_abort)->cm_frame_busaddr >> 32);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
cm->cm_data = NULL;
Change the internal API for polled commands. Calling mfi_polled_command after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED flag is set. This change eliminates the possibility of a polled command getting posted twice to the driver. This is turn fixes panics on shutdown when INVARIANTS is set. Sponsored by: Ironport
2006-11-18 07:33:53 +00:00
cm->cm_flags = MFI_CMD_POLLED;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0)
device_printf(sc->mfi_dev, "failed to abort command\n");
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
mfi_release_command(cm);
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
mtx_unlock(&sc->mfi_io_lock);
while (i < 5 && *cm_abort != NULL) {
tsleep(cm_abort, 0, "mfiabort",
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
5 * hz);
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
i++;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
if (*cm_abort != NULL) {
/* Force a complete if command didn't abort */
mtx_lock(&sc->mfi_io_lock);
(*cm_abort)->cm_complete(*cm_abort);
mtx_unlock(&sc->mfi_io_lock);
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
return (error);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
int
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_dump_blocks(struct mfi_softc *sc, int id, uint64_t lba, void *virt,
int len)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_command *cm;
struct mfi_io_frame *io;
int error;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL)
return (EBUSY);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Zero out the MFI frame */
context = cm->cm_frame->header.context;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(cm->cm_frame, sizeof(union mfi_frame));
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.context = context;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io = &cm->cm_frame->io;
io->header.cmd = MFI_CMD_LD_WRITE;
io->header.target_id = id;
io->header.timeout = 0;
io->header.flags = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
io->header.scsi_status = 0;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io->header.sense_len = MFI_SENSE_LEN;
io->header.data_len = (len + MFI_SECTOR_LEN - 1) / MFI_SECTOR_LEN;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
io->sense_addr_lo = (uint32_t)cm->cm_sense_busaddr;
io->sense_addr_hi = (uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
io->lba_hi = (lba & 0xffffffff00000000) >> 32;
io->lba_lo = lba & 0xffffffff;
cm->cm_data = virt;
cm->cm_len = len;
cm->cm_sg = &io->sgl;
cm->cm_total_frame_size = MFI_IO_FRAME_SIZE;
cm->cm_flags = MFI_CMD_POLLED | MFI_CMD_DATAOUT;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0)
device_printf(sc->mfi_dev, "failed dump blocks\n");
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
mfi_release_command(cm);
return (error);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
mfi_dump_syspd_blocks(struct mfi_softc *sc, int id, uint64_t lba, void *virt,
int len)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
{
struct mfi_command *cm;
struct mfi_pass_frame *pass;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
int error, readop, cdb_len;
Copy code from scsi_read_write() as mfi_build_syspd_cdb() to build SCSI command properly. Without this change, mfi(4) always sends 10 byte READ and WRITE commands, which will cause data corruption when device is larger than 2^32 sectors. PR: kern/173291 Submitted by: Steven Hartland <steven.hartland multiplay.co.uk> Reviewed by: mav MFC after: 2 weeks
2012-11-02 22:07:45 +00:00
uint32_t blkcount;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if ((cm = mfi_dequeue_free(sc)) == NULL)
return (EBUSY);
pass = &cm->cm_frame->pass;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bzero(pass->cdb, 16);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass->header.cmd = MFI_CMD_PD_SCSI_IO;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
readop = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
blkcount = (len + MFI_SECTOR_LEN - 1) / MFI_SECTOR_LEN;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
cdb_len = mfi_build_cdb(readop, 0, lba, blkcount, pass->cdb);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
pass->header.target_id = id;
pass->header.timeout = 0;
pass->header.flags = 0;
pass->header.scsi_status = 0;
pass->header.sense_len = MFI_SENSE_LEN;
pass->header.data_len = len;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
pass->header.cdb_len = cdb_len;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
pass->sense_addr_lo = (uint32_t)cm->cm_sense_busaddr;
pass->sense_addr_hi = (uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_data = virt;
cm->cm_len = len;
cm->cm_sg = &pass->sgl;
cm->cm_total_frame_size = MFI_PASS_FRAME_SIZE;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
cm->cm_flags = MFI_CMD_POLLED | MFI_CMD_DATAOUT | MFI_CMD_SCSI;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if ((error = mfi_mapcmd(sc, cm)) != 0)
device_printf(sc->mfi_dev, "failed dump blocks\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
bus_dmamap_sync(sc->mfi_buffer_dmat, cm->cm_dmamap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->mfi_buffer_dmat, cm->cm_dmamap);
mfi_release_command(cm);
return (error);
}
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int
We no longer need to use d_thread_t, migrate to struct thread *.
2009-05-20 17:29:21 +00:00
mfi_open(struct cdev *dev, int flags, int fmt, struct thread *td)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_softc *sc;
Fix a few nits relative to the previous changes: - Don't leak the config lock if detach() fails due to the controller char dev being open. - Close a race between detach() and a process opening the controller char dev. MFC after: 1 week Approved by: re (bmah)
2007-08-13 21:14:16 +00:00
int error;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc = dev->si_drv1;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Fix a few nits relative to the previous changes: - Don't leak the config lock if detach() fails due to the controller char dev being open. - Close a race between detach() and a process opening the controller char dev. MFC after: 1 week Approved by: re (bmah)
2007-08-13 21:14:16 +00:00
if (sc->mfi_detaching)
error = ENXIO;
else {
sc->mfi_flags |= MFI_FLAGS_OPEN;
error = 0;
}
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
Fix a few nits relative to the previous changes: - Don't leak the config lock if detach() fails due to the controller char dev being open. - Close a race between detach() and a process opening the controller char dev. MFC after: 1 week Approved by: re (bmah)
2007-08-13 21:14:16 +00:00
return (error);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
}
static int
We no longer need to use d_thread_t, migrate to struct thread *.
2009-05-20 17:29:21 +00:00
mfi_close(struct cdev *dev, int flags, int fmt, struct thread *td)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_softc *sc;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
struct mfi_aen *mfi_aen_entry, *tmp;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc = dev->si_drv1;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc->mfi_flags &= ~MFI_FLAGS_OPEN;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
TAILQ_FOREACH_SAFE(mfi_aen_entry, &sc->mfi_aen_pids, aen_link, tmp) {
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (mfi_aen_entry->p == curproc) {
TAILQ_REMOVE(&sc->mfi_aen_pids, mfi_aen_entry,
aen_link);
free(mfi_aen_entry, M_MFIBUF);
}
}
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
return (0);
}
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
static int
mfi_config_lock(struct mfi_softc *sc, uint32_t opcode)
{
switch (opcode) {
case MFI_DCMD_LD_DELETE:
case MFI_DCMD_CFG_ADD:
case MFI_DCMD_CFG_CLEAR:
Grab the mfi_config_lock while performing a MFI_DCMD_CFG_FOREIGN_IMPORT request on behalf of a user utility. Submitted by: Steven Hartland killing multiplay co uk MFC after: 1 week
2012-09-26 14:14:06 +00:00
case MFI_DCMD_CFG_FOREIGN_IMPORT:
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
sx_xlock(&sc->mfi_config_lock);
return (1);
default:
return (0);
}
}
static void
mfi_config_unlock(struct mfi_softc *sc, int locked)
{
if (locked)
sx_xunlock(&sc->mfi_config_lock);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
/*
* Perform pre-issue checks on commands from userland and possibly veto
* them.
*/
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
static int
mfi_check_command_pre(struct mfi_softc *sc, struct mfi_command *cm)
{
struct mfi_disk *ld, *ld2;
int error;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_system_pd *syspd = NULL;
uint16_t syspd_id;
uint16_t *mbox;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mtx_assert(&sc->mfi_io_lock, MA_OWNED);
error = 0;
switch (cm->cm_frame->dcmd.opcode) {
case MFI_DCMD_LD_DELETE:
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
if (ld->ld_id == cm->cm_frame->dcmd.mbox[0])
break;
}
if (ld == NULL)
error = ENOENT;
else
error = mfi_disk_disable(ld);
break;
case MFI_DCMD_CFG_CLEAR:
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
error = mfi_disk_disable(ld);
if (error)
break;
}
if (error) {
TAILQ_FOREACH(ld2, &sc->mfi_ld_tqh, ld_link) {
if (ld2 == ld)
break;
mfi_disk_enable(ld2);
}
}
break;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
case MFI_DCMD_PD_STATE_SET:
mbox = (uint16_t *) cm->cm_frame->dcmd.mbox;
syspd_id = mbox[0];
if (mbox[2] == MFI_PD_STATE_UNCONFIGURED_GOOD) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
TAILQ_FOREACH(syspd, &sc->mfi_syspd_tqh, pd_link) {
if (syspd->pd_id == syspd_id)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
}
}
else
break;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (syspd)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
error = mfi_syspd_disable(syspd);
break;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
default:
break;
}
return (error);
}
/* Perform post-issue checks on commands from userland. */
static void
mfi_check_command_post(struct mfi_softc *sc, struct mfi_command *cm)
{
struct mfi_disk *ld, *ldn;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_system_pd *syspd = NULL;
uint16_t syspd_id;
uint16_t *mbox;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
switch (cm->cm_frame->dcmd.opcode) {
case MFI_DCMD_LD_DELETE:
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
if (ld->ld_id == cm->cm_frame->dcmd.mbox[0])
break;
}
KASSERT(ld != NULL, ("volume dissappeared"));
if (cm->cm_frame->header.cmd_status == MFI_STAT_OK) {
mtx_unlock(&sc->mfi_io_lock);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_lock(&Giant);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
device_delete_child(sc->mfi_dev, ld->ld_dev);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_unlock(&Giant);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mtx_lock(&sc->mfi_io_lock);
} else
mfi_disk_enable(ld);
break;
case MFI_DCMD_CFG_CLEAR:
if (cm->cm_frame->header.cmd_status == MFI_STAT_OK) {
mtx_unlock(&sc->mfi_io_lock);
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_lock(&Giant);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
TAILQ_FOREACH_SAFE(ld, &sc->mfi_ld_tqh, ld_link, ldn) {
device_delete_child(sc->mfi_dev, ld->ld_dev);
}
Temporarily revert the new-bus locking for 8.0 release. It will be reintroduced after HEAD is reopened for commits by re@. Approved by: re (kib), attilio
2009-08-20 19:17:53 +00:00
mtx_unlock(&Giant);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mtx_lock(&sc->mfi_io_lock);
} else {
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link)
mfi_disk_enable(ld);
}
break;
case MFI_DCMD_CFG_ADD:
mfi_ldprobe(sc);
break;
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
case MFI_DCMD_CFG_FOREIGN_IMPORT:
mfi_ldprobe(sc);
break;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
case MFI_DCMD_PD_STATE_SET:
mbox = (uint16_t *) cm->cm_frame->dcmd.mbox;
syspd_id = mbox[0];
if (mbox[2] == MFI_PD_STATE_UNCONFIGURED_GOOD) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
TAILQ_FOREACH(syspd, &sc->mfi_syspd_tqh,pd_link) {
if (syspd->pd_id == syspd_id)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
break;
}
}
else
break;
/* If the transition fails then enable the syspd again */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (syspd && cm->cm_frame->header.cmd_status != MFI_STAT_OK)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_syspd_enable(syspd);
break;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
}
}
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
static int
mfi_check_for_sscd(struct mfi_softc *sc, struct mfi_command *cm)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
{
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_config_data *conf_data;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
struct mfi_command *ld_cm = NULL;
struct mfi_ld_info *ld_info = NULL;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
struct mfi_ld_config *ld;
char *p;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int error = 0;
- Extend the prior commit to use the generic SCSI command building function use that for JBOD and Thunderbolt disk write command. Now we only have one implementation in mfi. - Fix dumping on Thunderbolt cards. Polled IO commands do not seem to be normally acknowledged by changing cmd_status to MFI_STAT_OK. In order to get acknowledgement of the IO is complete, the Thunderbolt command queue needs to be run through. I added a flag MFI_CMD_SCSI to indicate this command is being polled and to complete the Thunderbolt wrapper and indicate the result. This flag needs to be set in the JBOD case in case if that us using Thunderbolt card. When in the polling loop check for completed commands. - Remove mfi_tbolt_is_ldio and just do the check when needed. - Fix an issue when attaching of disk device happens when a device is already scheduled to be attached but hasn't attached. - add a tunable to allow raw disk attachment to CAM via: hw.mfi.allow_cam_disk_passthrough=1 - fixup aborting of commands (AEN and LD state change). Use a generic abort function and only wait the command being aborted not both. Thunderbolt cards don't seem to abort commands so the abort times out.
2012-11-06 23:25:06 +00:00
conf_data = (struct mfi_config_data *)cm->cm_data;
if (cm->cm_frame->dcmd.opcode == MFI_DCMD_CFG_ADD) {
p = (char *)conf_data->array;
p += conf_data->array_size * conf_data->array_count;
ld = (struct mfi_ld_config *)p;
if (ld->params.isSSCD == 1)
error = 1;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
} else if (cm->cm_frame->dcmd.opcode == MFI_DCMD_LD_DELETE) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
error = mfi_dcmd_command (sc, &ld_cm, MFI_DCMD_LD_GET_INFO,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
(void **)&ld_info, sizeof(*ld_info));
Some style improvements.
2012-05-04 16:22:13 +00:00
if (error) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "Failed to allocate"
"MFI_DCMD_LD_GET_INFO %d", error);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (ld_info)
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
free(ld_info, M_MFIBUF);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return 0;
}
ld_cm->cm_flags = MFI_CMD_DATAIN;
ld_cm->cm_frame->dcmd.mbox[0]= cm->cm_frame->dcmd.mbox[0];
ld_cm->cm_frame->header.target_id = cm->cm_frame->dcmd.mbox[0];
Some style improvements.
2012-05-04 16:22:13 +00:00
if (mfi_wait_command(sc, ld_cm) != 0) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev, "failed to get log drv\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_release_command(ld_cm);
free(ld_info, M_MFIBUF);
return 0;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (ld_cm->cm_frame->header.cmd_status != MFI_STAT_OK) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
free(ld_info, M_MFIBUF);
mfi_release_command(ld_cm);
return 0;
}
else
ld_info = (struct mfi_ld_info *)ld_cm->cm_private;
if (ld_info->ld_config.params.isSSCD == 1)
error = 1;
mfi_release_command(ld_cm);
free(ld_info, M_MFIBUF);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return error;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
static int
mfi_stp_cmd(struct mfi_softc *sc, struct mfi_command *cm,caddr_t arg)
{
uint8_t i;
struct mfi_ioc_packet *ioc;
ioc = (struct mfi_ioc_packet *)arg;
int sge_size, error;
struct megasas_sge *kern_sge;
memset(sc->kbuff_arr, 0, sizeof(sc->kbuff_arr));
kern_sge =(struct megasas_sge *) ((uintptr_t)cm->cm_frame + ioc->mfi_sgl_off);
cm->cm_frame->header.sg_count = ioc->mfi_sge_count;
if (sizeof(bus_addr_t) == 8) {
cm->cm_frame->header.flags |= MFI_FRAME_SGL64;
cm->cm_extra_frames = 2;
sge_size = sizeof(struct mfi_sg64);
} else {
cm->cm_extra_frames = (cm->cm_total_frame_size - 1) / MFI_FRAME_SIZE;
sge_size = sizeof(struct mfi_sg32);
}
cm->cm_total_frame_size += (sge_size * ioc->mfi_sge_count);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
for (i = 0; i < ioc->mfi_sge_count; i++) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (bus_dma_tag_create( sc->mfi_parent_dmat, /* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
ioc->mfi_sgl[i].iov_len,/* maxsize */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
2, /* nsegments */
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
ioc->mfi_sgl[i].iov_len,/* maxsegsize */
BUS_DMA_ALLOCNOW, /* flags */
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
NULL, NULL, /* lockfunc, lockarg */
&sc->mfi_kbuff_arr_dmat[i])) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Cannot allocate mfi_kbuff_arr_dmat tag\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return (ENOMEM);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (bus_dmamem_alloc(sc->mfi_kbuff_arr_dmat[i],
(void **)&sc->kbuff_arr[i], BUS_DMA_NOWAIT,
&sc->mfi_kbuff_arr_dmamap[i])) {
device_printf(sc->mfi_dev,
"Cannot allocate mfi_kbuff_arr_dmamap memory\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return (ENOMEM);
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
bus_dmamap_load(sc->mfi_kbuff_arr_dmat[i],
sc->mfi_kbuff_arr_dmamap[i], sc->kbuff_arr[i],
ioc->mfi_sgl[i].iov_len, mfi_addr_cb,
&sc->mfi_kbuff_arr_busaddr[i], 0);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (!sc->kbuff_arr[i]) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
device_printf(sc->mfi_dev,
"Could not allocate memory for kbuff_arr info\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return -1;
}
kern_sge[i].phys_addr = sc->mfi_kbuff_arr_busaddr[i];
kern_sge[i].length = ioc->mfi_sgl[i].iov_len;
if (sizeof(bus_addr_t) == 8) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
cm->cm_frame->stp.sgl.sg64[i].addr =
kern_sge[i].phys_addr;
cm->cm_frame->stp.sgl.sg64[i].len =
ioc->mfi_sgl[i].iov_len;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
} else {
Fix a typo in mfi_stp_cmd() that would give wrong assignment. Submitted by: Sascha Wildner <saw online de> Obtained from: DragonFly rev 0dc98fff2206d7bb78ce5e07ac34d6954e4bd96a MFC after: 3 days
2013-03-01 23:18:20 +00:00
cm->cm_frame->stp.sgl.sg32[i].addr =
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
kern_sge[i].phys_addr;
cm->cm_frame->stp.sgl.sg32[i].len =
ioc->mfi_sgl[i].iov_len;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
error = copyin(ioc->mfi_sgl[i].iov_base,
sc->kbuff_arr[i],
ioc->mfi_sgl[i].iov_len);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (error != 0) {
device_printf(sc->mfi_dev, "Copy in failed\n");
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return error;
}
}
cm->cm_flags |=MFI_CMD_MAPPED;
return 0;
}
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
static int
mfi_user_command(struct mfi_softc *sc, struct mfi_ioc_passthru *ioc)
{
struct mfi_command *cm;
struct mfi_dcmd_frame *dcmd;
void *ioc_buf = NULL;
uint32_t context;
int error = 0, locked;
if (ioc->buf_size > 0) {
Fix panics triggered by older mfiutil binaries run on the new mfi(4) driver. The new driver changed the size of the mfi_dcmd_frame structure in such a way that a MFI_IOC_PASSTHRU ioctl from an old amd64 binary is treated as an MFI_IOC_PASSTHRU32 ioctl in the new driver. As a result, the user pointer is treated as the buffer length. mfi_user_command() doesn't have a bounds check on the buffer length, so it passes a really big value to malloc() which panics when it tries to exhaust the kmem_map. Fix this two ways: - Only honor MFI_IOC_PASSTHRU32 if the binary has the SV_ILP32 flag set, otherwise treat it as an unknown ioctl. - Add a bounds check on the buffer length passed by the user. For now it fails any user attempts to use a buffer larger than 1MB. While here, fix a few other nits: - Remove an unnecessary check for a NULL return from malloc(M_WAITOK). - Use the ENOTTY errno for invalid ioctl commands instead of ENOENT. MFC after: 3 days
2012-07-03 16:12:57 +00:00
if (ioc->buf_size > 1024 * 1024)
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
return (ENOMEM);
Fix panics triggered by older mfiutil binaries run on the new mfi(4) driver. The new driver changed the size of the mfi_dcmd_frame structure in such a way that a MFI_IOC_PASSTHRU ioctl from an old amd64 binary is treated as an MFI_IOC_PASSTHRU32 ioctl in the new driver. As a result, the user pointer is treated as the buffer length. mfi_user_command() doesn't have a bounds check on the buffer length, so it passes a really big value to malloc() which panics when it tries to exhaust the kmem_map. Fix this two ways: - Only honor MFI_IOC_PASSTHRU32 if the binary has the SV_ILP32 flag set, otherwise treat it as an unknown ioctl. - Add a bounds check on the buffer length passed by the user. For now it fails any user attempts to use a buffer larger than 1MB. While here, fix a few other nits: - Remove an unnecessary check for a NULL return from malloc(M_WAITOK). - Use the ENOTTY errno for invalid ioctl commands instead of ENOENT. MFC after: 3 days
2012-07-03 16:12:57 +00:00
ioc_buf = malloc(ioc->buf_size, M_MFIBUF, M_WAITOK);
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
error = copyin(ioc->buf, ioc_buf, ioc->buf_size);
if (error) {
device_printf(sc->mfi_dev, "failed to copyin\n");
free(ioc_buf, M_MFIBUF);
return (error);
}
}
locked = mfi_config_lock(sc, ioc->ioc_frame.opcode);
mtx_lock(&sc->mfi_io_lock);
while ((cm = mfi_dequeue_free(sc)) == NULL)
msleep(mfi_user_command, &sc->mfi_io_lock, 0, "mfiioc", hz);
/* Save context for later */
context = cm->cm_frame->header.context;
dcmd = &cm->cm_frame->dcmd;
bcopy(&ioc->ioc_frame, dcmd, sizeof(struct mfi_dcmd_frame));
cm->cm_sg = &dcmd->sgl;
cm->cm_total_frame_size = MFI_DCMD_FRAME_SIZE;
cm->cm_data = ioc_buf;
cm->cm_len = ioc->buf_size;
/* restore context */
cm->cm_frame->header.context = context;
/* Cheat since we don't know if we're writing or reading */
cm->cm_flags = MFI_CMD_DATAIN | MFI_CMD_DATAOUT;
error = mfi_check_command_pre(sc, cm);
if (error)
goto out;
error = mfi_wait_command(sc, cm);
if (error) {
device_printf(sc->mfi_dev, "ioctl failed %d\n", error);
goto out;
}
bcopy(dcmd, &ioc->ioc_frame, sizeof(struct mfi_dcmd_frame));
mfi_check_command_post(sc, cm);
out:
mfi_release_command(cm);
mtx_unlock(&sc->mfi_io_lock);
mfi_config_unlock(sc, locked);
if (ioc->buf_size > 0)
error = copyout(ioc_buf, ioc->buf, ioc->buf_size);
if (ioc_buf)
free(ioc_buf, M_MFIBUF);
return (error);
}
#define PTRIN(p) ((void *)(uintptr_t)(p))
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
static int
We no longer need to use d_thread_t, migrate to struct thread *.
2009-05-20 17:29:21 +00:00
mfi_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int flag, struct thread *td)
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
{
struct mfi_softc *sc;
union mfi_statrequest *ms;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
struct mfi_ioc_packet *ioc;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
Add support to talk to the LSI ioctl path on with FreeBSD 32 bit app's on amd64. Note the only difference is the iovec32 part so I use the native structure for everything else. Also I plan to MFC all the changes in -current to 7-stable and 6-stable shortly since I've been running them. This does not include the cam changes. MFC after: 3 days
2008-05-28 23:19:27 +00:00
struct mfi_ioc_packet32 *ioc32;
#endif
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
struct mfi_ioc_aen *aen;
struct mfi_command *cm = NULL;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
union mfi_sense_ptr sense_ptr;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint8_t *data = NULL, *temp, *addr, skip_pre_post = 0;
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
size_t len;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
int i, res;
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
struct mfi_ioc_passthru *iop = (struct mfi_ioc_passthru *)arg;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
struct mfi_ioc_passthru32 *iop32 = (struct mfi_ioc_passthru32 *)arg;
struct mfi_ioc_passthru iop_swab;
#endif
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
int error, locked;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
union mfi_sgl *sgl;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
sc = dev->si_drv1;
error = 0;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (sc->adpreset)
return EBUSY;
if (sc->hw_crit_error)
return EBUSY;
if (sc->issuepend_done == 0)
return EBUSY;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
switch (cmd) {
case MFIIO_STATS:
ms = (union mfi_statrequest *)arg;
switch (ms->ms_item) {
case MFIQ_FREE:
case MFIQ_BIO:
case MFIQ_READY:
case MFIQ_BUSY:
bcopy(&sc->mfi_qstat[ms->ms_item], &ms->ms_qstat,
sizeof(struct mfi_qstat));
break;
default:
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
error = ENOIOCTL;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
}
break;
Collapse the mfi_ld object. Add an ioctl to help management apps map array Id's to FreeBSD device names.
2007-05-10 15:33:41 +00:00
case MFIIO_QUERY_DISK:
{
struct mfi_query_disk *qd;
struct mfi_disk *ld;
qd = (struct mfi_query_disk *)arg;
mtx_lock(&sc->mfi_io_lock);
TAILQ_FOREACH(ld, &sc->mfi_ld_tqh, ld_link) {
if (ld->ld_id == qd->array_id)
break;
}
if (ld == NULL) {
qd->present = 0;
mtx_unlock(&sc->mfi_io_lock);
return (0);
}
qd->present = 1;
if (ld->ld_flags & MFI_DISK_FLAGS_OPEN)
qd->open = 1;
bzero(qd->devname, SPECNAMELEN + 1);
snprintf(qd->devname, SPECNAMELEN, "mfid%d", ld->ld_unit);
mtx_unlock(&sc->mfi_io_lock);
break;
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
case MFI_CMD:
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
Add support to talk to the LSI ioctl path on with FreeBSD 32 bit app's on amd64. Note the only difference is the iovec32 part so I use the native structure for everything else. Also I plan to MFC all the changes in -current to 7-stable and 6-stable shortly since I've been running them. This does not include the cam changes. MFC after: 3 days
2008-05-28 23:19:27 +00:00
case MFI_CMD32:
#endif
Add in a compat. mode so you can either open the card's device node or directly open mfi0 and specify the card you want to talk to in the ioctl.
2008-03-22 02:57:49 +00:00
{
devclass_t devclass;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
ioc = (struct mfi_ioc_packet *)arg;
Add in a compat. mode so you can either open the card's device node or directly open mfi0 and specify the card you want to talk to in the ioctl.
2008-03-22 02:57:49 +00:00
int adapter;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
Add in a compat. mode so you can either open the card's device node or directly open mfi0 and specify the card you want to talk to in the ioctl.
2008-03-22 02:57:49 +00:00
adapter = ioc->mfi_adapter_no;
if (device_get_unit(sc->mfi_dev) == 0 && adapter != 0) {
devclass = devclass_find("mfi");
sc = devclass_get_softc(devclass, adapter);
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
mtx_lock(&sc->mfi_io_lock);
if ((cm = mfi_dequeue_free(sc)) == NULL) {
mtx_unlock(&sc->mfi_io_lock);
return (EBUSY);
}
mtx_unlock(&sc->mfi_io_lock);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
locked = 0;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
/*
* save off original context since copying from user
* will clobber some data
*/
context = cm->cm_frame->header.context;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.context = cm->cm_index;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
bcopy(ioc->mfi_frame.raw, cm->cm_frame,
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
2 * MEGAMFI_FRAME_SIZE);
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
cm->cm_total_frame_size = (sizeof(union mfi_sgl)
* ioc->mfi_sge_count) + ioc->mfi_sgl_off;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.scsi_status = 0;
cm->cm_frame->header.pad0 = 0;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
if (ioc->mfi_sge_count) {
cm->cm_sg =
(union mfi_sgl *)&cm->cm_frame->bytes[ioc->mfi_sgl_off];
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
sgl = cm->cm_sg;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
cm->cm_flags = 0;
if (cm->cm_frame->header.flags & MFI_FRAME_DATAIN)
cm->cm_flags |= MFI_CMD_DATAIN;
if (cm->cm_frame->header.flags & MFI_FRAME_DATAOUT)
cm->cm_flags |= MFI_CMD_DATAOUT;
/* Legacy app shim */
if (cm->cm_flags == 0)
cm->cm_flags |= MFI_CMD_DATAIN | MFI_CMD_DATAOUT;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
cm->cm_len = cm->cm_frame->header.data_len;
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_STP) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
if (cmd == MFI_CMD) {
#endif
/* Native */
cm->cm_stp_len = ioc->mfi_sgl[0].iov_len;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
} else {
/* 32bit on 64bit */
ioc32 = (struct mfi_ioc_packet32 *)ioc;
cm->cm_stp_len = ioc32->mfi_sgl[0].iov_len;
}
#endif
cm->cm_len += cm->cm_stp_len;
}
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
if (cm->cm_len &&
(cm->cm_flags & (MFI_CMD_DATAIN | MFI_CMD_DATAOUT))) {
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
cm->cm_data = data = malloc(cm->cm_len, M_MFIBUF,
M_WAITOK | M_ZERO);
if (cm->cm_data == NULL) {
device_printf(sc->mfi_dev, "Malloc failed\n");
goto out;
}
} else {
cm->cm_data = 0;
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
/* restore header context */
cm->cm_frame->header.context = context;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_STP) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
res = mfi_stp_cmd(sc, cm, arg);
if (res != 0)
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
goto out;
} else {
temp = data;
if ((cm->cm_flags & MFI_CMD_DATAOUT) ||
(cm->cm_frame->header.cmd == MFI_CMD_STP)) {
for (i = 0; i < ioc->mfi_sge_count; i++) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (cmd == MFI_CMD) {
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
#endif
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Native */
addr = ioc->mfi_sgl[i].iov_base;
len = ioc->mfi_sgl[i].iov_len;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
} else {
/* 32bit on 64bit */
ioc32 = (struct mfi_ioc_packet32 *)ioc;
addr = PTRIN(ioc32->mfi_sgl[i].iov_base);
len = ioc32->mfi_sgl[i].iov_len;
}
Add support to talk to the LSI ioctl path on with FreeBSD 32 bit app's on amd64. Note the only difference is the iovec32 part so I use the native structure for everything else. Also I plan to MFC all the changes in -current to 7-stable and 6-stable shortly since I've been running them. This does not include the cam changes. MFC after: 3 days
2008-05-28 23:19:27 +00:00
#endif
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
error = copyin(addr, temp, len);
if (error != 0) {
device_printf(sc->mfi_dev,
"Copy in failed\n");
goto out;
}
temp = &temp[len];
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
}
}
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_DCMD)
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
locked = mfi_config_lock(sc,
cm->cm_frame->dcmd.opcode);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
One more piece to add to make sense data work for a user app. from LSI. Submitted by: LSI MFC after: 3 days
2008-11-13 17:13:16 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_PD_SCSI_IO) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
cm->cm_frame->pass.sense_addr_lo =
(uint32_t)cm->cm_sense_busaddr;
cm->cm_frame->pass.sense_addr_hi =
(uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
One more piece to add to make sense data work for a user app. from LSI. Submitted by: LSI MFC after: 3 days
2008-11-13 17:13:16 +00:00
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
mtx_lock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
skip_pre_post = mfi_check_for_sscd (sc, cm);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (!skip_pre_post) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
error = mfi_check_command_pre(sc, cm);
if (error) {
mtx_unlock(&sc->mfi_io_lock);
goto out;
}
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
}
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
if ((error = mfi_wait_command(sc, cm)) != 0) {
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
device_printf(sc->mfi_dev,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
"Controller polled failed\n");
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
mtx_unlock(&sc->mfi_io_lock);
goto out;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (!skip_pre_post) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
mfi_check_command_post(sc, cm);
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
mtx_unlock(&sc->mfi_io_lock);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (cm->cm_frame->header.cmd != MFI_CMD_STP) {
temp = data;
if ((cm->cm_flags & MFI_CMD_DATAIN) ||
(cm->cm_frame->header.cmd == MFI_CMD_STP)) {
for (i = 0; i < ioc->mfi_sge_count; i++) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
if (cmd == MFI_CMD) {
- Add special support for the MFI_CMD ioctl with MFI_CMD_STP command, used by present MegaCLI version. It has some special meaning for the first s/g list entry, while the main s/g list begins from the the second entry, and those lists should remain separate after loading to the busdma map. - Fix bug in 32bit ioctl compatibility shims when s/g list consists of more then on element. Sponsored by: iXsystems, inc. MFC after: 3 days
2011-09-29 08:37:53 +00:00
#endif
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
/* Native */
addr = ioc->mfi_sgl[i].iov_base;
len = ioc->mfi_sgl[i].iov_len;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
} else {
/* 32bit on 64bit */
ioc32 = (struct mfi_ioc_packet32 *)ioc;
addr = PTRIN(ioc32->mfi_sgl[i].iov_base);
len = ioc32->mfi_sgl[i].iov_len;
}
Add support to talk to the LSI ioctl path on with FreeBSD 32 bit app's on amd64. Note the only difference is the iovec32 part so I use the native structure for everything else. Also I plan to MFC all the changes in -current to 7-stable and 6-stable shortly since I've been running them. This does not include the cam changes. MFC after: 3 days
2008-05-28 23:19:27 +00:00
#endif
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
error = copyout(temp, addr, len);
if (error != 0) {
device_printf(sc->mfi_dev,
"Copy out failed\n");
goto out;
}
temp = &temp[len];
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
}
}
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
if (ioc->mfi_sense_len) {
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
/* get user-space sense ptr then copy out sense */
Shorten a variable access some using a cached value rather than casting the input argument again, making the code more readable. Reviewed by: jhb MFC after: 2 weeks Approved by: re (kib)
2011-09-07 00:43:28 +00:00
bcopy(&ioc->mfi_frame.raw[ioc->mfi_sense_off],
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
&sense_ptr.sense_ptr_data[0],
sizeof(sense_ptr.sense_ptr_data));
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
When running a 32bit app. on amd64, ensure the bits above 32bit are zero for the copyout. Confirmed by LSI.
2008-11-14 21:05:45 +00:00
if (cmd != MFI_CMD) {
/*
* not 64bit native so zero out any address
* over 32bit */
Fix compile. I was in the wrong tree when I tested it :-( Pointed out by: Andrzej
2008-11-14 23:32:31 +00:00
sense_ptr.addr.high = 0;
When running a 32bit app. on amd64, ensure the bits above 32bit are zero for the copyout. Confirmed by LSI.
2008-11-14 21:05:45 +00:00
}
#endif
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
error = copyout(cm->cm_sense, sense_ptr.user_space,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
ioc->mfi_sense_len);
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
if (error != 0) {
device_printf(sc->mfi_dev,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
"Copy out failed\n");
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
goto out;
}
}
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
ioc->mfi_frame.hdr.cmd_status = cm->cm_frame->header.cmd_status;
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
out:
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mfi_config_unlock(sc, locked);
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
if (data)
free(data, M_MFIBUF);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_STP) {
for (i = 0; i < 2; i++) {
if (sc->kbuff_arr[i]) {
if (sc->mfi_kbuff_arr_busaddr != 0)
bus_dmamap_unload(
sc->mfi_kbuff_arr_dmat[i],
sc->mfi_kbuff_arr_dmamap[i]
);
if (sc->kbuff_arr[i] != NULL)
bus_dmamem_free(
sc->mfi_kbuff_arr_dmat[i],
sc->kbuff_arr[i],
sc->mfi_kbuff_arr_dmamap[i]
);
if (sc->mfi_kbuff_arr_dmat[i] != NULL)
bus_dma_tag_destroy(
sc->mfi_kbuff_arr_dmat[i]);
}
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
}
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
if (cm) {
mtx_lock(&sc->mfi_io_lock);
mfi_release_command(cm);
mtx_unlock(&sc->mfi_io_lock);
}
break;
Add in a compat. mode so you can either open the card's device node or directly open mfi0 and specify the card you want to talk to in the ioctl.
2008-03-22 02:57:49 +00:00
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
case MFI_SET_AEN:
aen = (struct mfi_ioc_aen *)arg;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_lock(&sc->mfi_io_lock);
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
error = mfi_aen_register(sc, aen->aen_seq_num,
aen->aen_class_locale);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
mtx_unlock(&sc->mfi_io_lock);
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
break;
case MFI_LINUX_CMD_2: /* Firmware Linux ioctl shim */
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
devclass_t devclass;
struct mfi_linux_ioc_packet l_ioc;
int adapter;
devclass = devclass_find("mfi");
if (devclass == NULL)
return (ENOENT);
error = copyin(arg, &l_ioc, sizeof(l_ioc));
if (error)
return (error);
adapter = l_ioc.lioc_adapter_no;
sc = devclass_get_softc(devclass, adapter);
if (sc == NULL)
return (ENOENT);
return (mfi_linux_ioctl_int(sc->mfi_cdev,
cmd, arg, flag, td));
break;
}
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
case MFI_LINUX_SET_AEN_2: /* AEN Linux ioctl shim */
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
devclass_t devclass;
struct mfi_linux_ioc_aen l_aen;
int adapter;
devclass = devclass_find("mfi");
if (devclass == NULL)
return (ENOENT);
error = copyin(arg, &l_aen, sizeof(l_aen));
if (error)
return (error);
adapter = l_aen.laen_adapter_no;
sc = devclass_get_softc(devclass, adapter);
if (sc == NULL)
return (ENOENT);
return (mfi_linux_ioctl_int(sc->mfi_cdev,
cmd, arg, flag, td));
break;
}
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
case MFIIO_PASSTHRU32:
Fix panics triggered by older mfiutil binaries run on the new mfi(4) driver. The new driver changed the size of the mfi_dcmd_frame structure in such a way that a MFI_IOC_PASSTHRU ioctl from an old amd64 binary is treated as an MFI_IOC_PASSTHRU32 ioctl in the new driver. As a result, the user pointer is treated as the buffer length. mfi_user_command() doesn't have a bounds check on the buffer length, so it passes a really big value to malloc() which panics when it tries to exhaust the kmem_map. Fix this two ways: - Only honor MFI_IOC_PASSTHRU32 if the binary has the SV_ILP32 flag set, otherwise treat it as an unknown ioctl. - Add a bounds check on the buffer length passed by the user. For now it fails any user attempts to use a buffer larger than 1MB. While here, fix a few other nits: - Remove an unnecessary check for a NULL return from malloc(M_WAITOK). - Use the ENOTTY errno for invalid ioctl commands instead of ENOENT. MFC after: 3 days
2012-07-03 16:12:57 +00:00
if (!SV_CURPROC_FLAG(SV_ILP32)) {
error = ENOTTY;
break;
}
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
iop_swab.ioc_frame = iop32->ioc_frame;
iop_swab.buf_size = iop32->buf_size;
iop_swab.buf = PTRIN(iop32->buf);
iop = &iop_swab;
/* FALLTHROUGH */
#endif
case MFIIO_PASSTHRU:
error = mfi_user_command(sc, iop);
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
#ifdef COMPAT_FREEBSD32
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
if (cmd == MFIIO_PASSTHRU32)
iop32->ioc_frame = iop_swab.ioc_frame;
#endif
break;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
default:
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
device_printf(sc->mfi_dev, "IOCTL 0x%lx not handled\n", cmd);
Fix panics triggered by older mfiutil binaries run on the new mfi(4) driver. The new driver changed the size of the mfi_dcmd_frame structure in such a way that a MFI_IOC_PASSTHRU ioctl from an old amd64 binary is treated as an MFI_IOC_PASSTHRU32 ioctl in the new driver. As a result, the user pointer is treated as the buffer length. mfi_user_command() doesn't have a bounds check on the buffer length, so it passes a really big value to malloc() which panics when it tries to exhaust the kmem_map. Fix this two ways: - Only honor MFI_IOC_PASSTHRU32 if the binary has the SV_ILP32 flag set, otherwise treat it as an unknown ioctl. - Add a bounds check on the buffer length passed by the user. For now it fails any user attempts to use a buffer larger than 1MB. While here, fix a few other nits: - Remove an unnecessary check for a NULL return from malloc(M_WAITOK). - Use the ENOTTY errno for invalid ioctl commands instead of ENOENT. MFC after: 3 days
2012-07-03 16:12:57 +00:00
error = ENOTTY;
Add a driver for the new LSI MegaRAID SAS controller family. The 'MFI' name is derived from the phrase 'MegaRAID Firmware Interface' used by LSI. This driver provides a block interface to logical disks on the card and a minimal management device. It is MPSAFE, INTR_FAST, and 64-bit capable. Thanks to Dell for providing hardware to test with and IronPort for sponsoring the work. Sponsored by: Dell, Ironport MFC After: 3 days
2006-03-25 06:14:32 +00:00
break;
}
return (error);
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
static int
We no longer need to use d_thread_t, migrate to struct thread *.
2009-05-20 17:29:21 +00:00
mfi_linux_ioctl_int(struct cdev *dev, u_long cmd, caddr_t arg, int flag, struct thread *td)
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
{
struct mfi_softc *sc;
struct mfi_linux_ioc_packet l_ioc;
struct mfi_linux_ioc_aen l_aen;
struct mfi_command *cm = NULL;
struct mfi_aen *mfi_aen_entry;
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
union mfi_sense_ptr sense_ptr;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
uint32_t context = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
uint8_t *data = NULL, *temp;
int i;
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
int error, locked;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
sc = dev->si_drv1;
error = 0;
switch (cmd) {
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
case MFI_LINUX_CMD_2: /* Firmware Linux ioctl shim */
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
error = copyin(arg, &l_ioc, sizeof(l_ioc));
if (error != 0)
return (error);
if (l_ioc.lioc_sge_count > MAX_LINUX_IOCTL_SGE) {
return (EINVAL);
}
mtx_lock(&sc->mfi_io_lock);
if ((cm = mfi_dequeue_free(sc)) == NULL) {
mtx_unlock(&sc->mfi_io_lock);
return (EBUSY);
}
mtx_unlock(&sc->mfi_io_lock);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
locked = 0;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
/*
* save off original context since copying from user
* will clobber some data
*/
context = cm->cm_frame->header.context;
bcopy(l_ioc.lioc_frame.raw, cm->cm_frame,
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
2 * MFI_DCMD_FRAME_SIZE); /* this isn't quite right */
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
cm->cm_total_frame_size = (sizeof(union mfi_sgl)
* l_ioc.lioc_sge_count) + l_ioc.lioc_sgl_off;
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_frame->header.scsi_status = 0;
cm->cm_frame->header.pad0 = 0;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
if (l_ioc.lioc_sge_count)
cm->cm_sg =
(union mfi_sgl *)&cm->cm_frame->bytes[l_ioc.lioc_sgl_off];
cm->cm_flags = 0;
if (cm->cm_frame->header.flags & MFI_FRAME_DATAIN)
cm->cm_flags |= MFI_CMD_DATAIN;
if (cm->cm_frame->header.flags & MFI_FRAME_DATAOUT)
cm->cm_flags |= MFI_CMD_DATAOUT;
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
cm->cm_len = cm->cm_frame->header.data_len;
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
if (cm->cm_len &&
(cm->cm_flags & (MFI_CMD_DATAIN | MFI_CMD_DATAOUT))) {
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
cm->cm_data = data = malloc(cm->cm_len, M_MFIBUF,
M_WAITOK | M_ZERO);
if (cm->cm_data == NULL) {
device_printf(sc->mfi_dev, "Malloc failed\n");
goto out;
}
} else {
cm->cm_data = 0;
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
/* restore header context */
cm->cm_frame->header.context = context;
temp = data;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
if (cm->cm_flags & MFI_CMD_DATAOUT) {
for (i = 0; i < l_ioc.lioc_sge_count; i++) {
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
error = copyin(PTRIN(l_ioc.lioc_sgl[i].iov_base),
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
temp,
l_ioc.lioc_sgl[i].iov_len);
if (error != 0) {
device_printf(sc->mfi_dev,
"Copy in failed\n");
goto out;
}
temp = &temp[l_ioc.lioc_sgl[i].iov_len];
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
}
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_DCMD)
locked = mfi_config_lock(sc, cm->cm_frame->dcmd.opcode);
One more piece to add to make sense data work for a user app. from LSI. Submitted by: LSI MFC after: 3 days
2008-11-13 17:13:16 +00:00
if (cm->cm_frame->header.cmd == MFI_CMD_PD_SCSI_IO) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
cm->cm_frame->pass.sense_addr_lo =
(uint32_t)cm->cm_sense_busaddr;
cm->cm_frame->pass.sense_addr_hi =
(uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32);
One more piece to add to make sense data work for a user app. from LSI. Submitted by: LSI MFC after: 3 days
2008-11-13 17:13:16 +00:00
}
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
error = mfi_check_command_pre(sc, cm);
if (error) {
mtx_unlock(&sc->mfi_io_lock);
goto out;
}
Add in a couple of things: - In the ioctl path let command get queued up and return when complete _without_ blocking the driving waiting for the response. This way the driver doesn't "lock up" for ~30s during a flash command. Submitted by scottl. - Add a guard so that if a DCMD of 0 is sent down the ioctl path don't send it to the controller. Return with a status of OK. This is a little strange since MegaCli doesn't seem to like something and will issue some DCMD of 0. This doesn't happen under Linux. So the emulation needs to be improved but I'm not sure what. Another strange thing is that when a DCMD of 0 gets issued under i386 the controller returns OK but in amd64 the context is messed up. - Add a guard so the context has to be with-in the legal limit so we get a reasonable error assertion versus random panic. It's going to be a challenge to figure out why MegaCli is not totally happy and then sends some bogus commands. This means that flashing firmware via the Linux tool won't work since it generates a DCMD of 0 when it should be opening the firmware for a flash update. Without this problem flashing works fine. This means there is no publicly available tool to upgrade the RAID firmware under FreeBSD right now. I plan to MFC all of the mfi changes to 6.X shortly. This might not include the SCSI pass-through changes. Submitted by: scottl Reviewed by: scottl MFC after: 3 days
2007-06-04 16:39:22 +00:00
if ((error = mfi_wait_command(sc, cm)) != 0) {
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
device_printf(sc->mfi_dev,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
"Controller polled failed\n");
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
goto out;
}
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mfi_check_command_post(sc, cm);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
temp = data;
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
if (cm->cm_flags & MFI_CMD_DATAIN) {
for (i = 0; i < l_ioc.lioc_sge_count; i++) {
error = copyout(temp,
Add support for management apps. Work around an apparent firmware bug that results in hung i/o if more than 128 commands are scheduled for an array.
2008-05-12 14:09:19 +00:00
PTRIN(l_ioc.lioc_sgl[i].iov_base),
Fix some bugs in dealing with DCMD'd without data. MegaCli was sending down some DCMD's without any data. Thanks to Dell and LSI for helping to provide clues to figure out this problem. Now MegaCli can upgrade the firmware and should work identical when run on Linux. Reviewed by: scottl, LSI MFC after: 1 day
2008-02-02 17:29:28 +00:00
l_ioc.lioc_sgl[i].iov_len);
if (error != 0) {
device_printf(sc->mfi_dev,
"Copy out failed\n");
goto out;
}
temp = &temp[l_ioc.lioc_sgl[i].iov_len];
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
}
}
if (l_ioc.lioc_sense_len) {
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
/* get user-space sense ptr then copy out sense */
bcopy(&((struct mfi_linux_ioc_packet*)arg)
->lioc_frame.raw[l_ioc.lioc_sense_off],
&sense_ptr.sense_ptr_data[0],
sizeof(sense_ptr.sense_ptr_data));
When running a 32bit app. on amd64, ensure the bits above 32bit are zero for the copyout. Confirmed by LSI.
2008-11-14 21:05:45 +00:00
#ifdef __amd64__
/*
* only 32bit Linux support so zero out any
* address over 32bit
*/
Fix compile. I was in the wrong tree when I tested it :-( Pointed out by: Andrzej
2008-11-14 23:32:31 +00:00
sense_ptr.addr.high = 0;
When running a 32bit app. on amd64, ensure the bits above 32bit are zero for the copyout. Confirmed by LSI.
2008-11-14 21:05:45 +00:00
#endif
- Fix from jhb for failing I/O request when bus_dmamap_load fails. - Fix to ioctl path in which the length could be 0 which means no data in/out from LSI. - Fix to ioctl path in which the data in the sense data space of the ioctl packet is a really a pointer to some location in user-space. From LSI re-worked a bit by me. - Add HW support for next gen cards from LSI. Thanks to LSI for their support! Submitted by: jhb, LSI MFC after: 3 days
2008-11-12 22:44:50 +00:00
error = copyout(cm->cm_sense, sense_ptr.user_space,
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
l_ioc.lioc_sense_len);
if (error != 0) {
device_printf(sc->mfi_dev,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
"Copy out failed\n");
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
goto out;
}
}
error = copyout(&cm->cm_frame->header.cmd_status,
&((struct mfi_linux_ioc_packet*)arg)
->lioc_frame.hdr.cmd_status,
1);
if (error != 0) {
device_printf(sc->mfi_dev,
Some relatively minor changes and bug fixes: 1) s/mi/mfi/ in FreeBSD ioctl path 2) add in "\n" on various failure messages 3) cap the length of time to abort an AEN command 4) fix passing sense data back to user to make Dell's Linux firmware upgrade tool happy. 5) bump the MFI_POLL_TIMEOUT_SECS from 10s to 50s since the firmware flash command can take ~40s to return. This is some clean-up and enables RAID firmware to updated via Dell's tool. Note Dell's tool requires the updates to the Linux emulator that has been done in -current with TLS etc. I need to discuss with scottl how to better submit mfi commands to the firmware via the ioctl path so we don't do it in polled mode.
2006-12-14 18:29:08 +00:00
"Copy out failed\n");
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
goto out;
}
out:
Teach the mfi(4) driver to handle requests from userland management applications to add and remove volumes. MFC after: 1 week Approved by: re (bmah) Reviewed by: ambrisko, scottl
2007-08-13 19:29:17 +00:00
mfi_config_unlock(sc, locked);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (data)
free(data, M_MFIBUF);
if (cm) {
mtx_lock(&sc->mfi_io_lock);
mfi_release_command(cm);
mtx_unlock(&sc->mfi_io_lock);
}
return (error);
- Add in FreeBSD native ioctl that models the Linux version. - Add a translation so the Linux ioctl's don't conflict with the FreeBSD definition. - Assume Linux 32bit emulation on amd64. This was tested on i386 and amd64 with the 32bit Linux MegaCli. Eventually we should do a 32bit native FreeBSD translation app.
2006-11-14 16:48:00 +00:00
case MFI_LINUX_SET_AEN_2: /* AEN Linux ioctl shim */
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
error = copyin(arg, &l_aen, sizeof(l_aen));
if (error != 0)
return (error);
printf("AEN IMPLEMENTED for pid %d\n", curproc->p_pid);
mfi_aen_entry = malloc(sizeof(struct mfi_aen), M_MFIBUF,
M_WAITOK);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (mfi_aen_entry != NULL) {
mfi_aen_entry->p = curproc;
TAILQ_INSERT_TAIL(&sc->mfi_aen_pids, mfi_aen_entry,
aen_link);
}
error = mfi_aen_register(sc, l_aen.laen_seq_num,
l_aen.laen_class_locale);
if (error != 0) {
TAILQ_REMOVE(&sc->mfi_aen_pids, mfi_aen_entry,
aen_link);
free(mfi_aen_entry, M_MFIBUF);
}
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_unlock(&sc->mfi_io_lock);
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
return (error);
default:
device_printf(sc->mfi_dev, "IOCTL 0x%lx not handled\n", cmd);
error = ENOENT;
break;
}
return (error);
}
static int
mfi_poll(struct cdev *dev, int poll_events, struct thread *td)
{
struct mfi_softc *sc;
int revents = 0;
sc = dev->si_drv1;
if (poll_events & (POLLIN | POLLRDNORM)) {
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
if (sc->mfi_aen_triggered != 0) {
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
revents |= poll_events & (POLLIN | POLLRDNORM);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
sc->mfi_aen_triggered = 0;
}
Add in a bunch of things to the mfi driver: - Linux ioctl support, with the other Linux changes MegaCli will run if you mount linprocfs & linsysfs then set sysctl compat.linux.osrelease=2.6.12 or similar. This works on i386. It should work on amd64 but not well tested yet. StoreLib may or may not work. Remember to kldload mfi_linux. - Add in AEN (Async Event Notification) support so we can get messages from the firmware when something happens. Not all messages are in defined in event detail. Use event_log to try to figure out what happened. - Try to implement something like SIGIO for StoreLib. Since mrmonitor doesn't work right I can't fully test it. StoreLib works best with the rh9 base. In theory mrmonitor isn't needed due to native driver support of AEN :-) Now we can configure and monitor the RAID better. Submitted by: IronPort Systems.
2006-05-18 23:30:48 +00:00
if (sc->mfi_aen_triggered == 0 && sc->mfi_aen_cm == NULL) {
revents |= POLLERR;
}
}
if (revents == 0) {
if (poll_events & (POLLIN | POLLRDNORM)) {
sc->mfi_poll_waiting = 1;
selrecord(td, &sc->mfi_select);
}
}
return revents;
}
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
static void
mfi_dump_all(void)
{
struct mfi_softc *sc;
struct mfi_command *cm;
devclass_t dc;
time_t deadline;
int timedout;
int i;
dc = devclass_find("mfi");
if (dc == NULL) {
printf("No mfi dev class\n");
return;
}
for (i = 0; ; i++) {
sc = devclass_get_softc(dc, i);
if (sc == NULL)
break;
device_printf(sc->mfi_dev, "Dumping\n\n");
timedout = 0;
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
deadline = time_uptime - mfi_cmd_timeout;
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
mtx_lock(&sc->mfi_io_lock);
TAILQ_FOREACH(cm, &sc->mfi_busy, cm_link) {
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (cm->cm_timestamp <= deadline) {
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
device_printf(sc->mfi_dev,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
"COMMAND %p TIMEOUT AFTER %d SECONDS\n",
cm, (int)(time_uptime - cm->cm_timestamp));
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
MFI_PRINT_CMD(cm);
timedout++;
}
}
#if 0
if (timedout)
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
MFI_DUMP_CMDS(sc);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
#endif
mtx_unlock(&sc->mfi_io_lock);
}
return;
}
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
static void
mfi_timeout(void *data)
{
struct mfi_softc *sc = (struct mfi_softc *)data;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
struct mfi_command *cm, *tmp;
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
time_t deadline;
int timedout = 0;
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
deadline = time_uptime - mfi_cmd_timeout;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->adpreset == 0) {
if (!mfi_tbolt_reset(sc)) {
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
callout_reset(&sc->mfi_watchdog_callout,
mfi_cmd_timeout * hz, mfi_timeout, sc);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
return;
}
}
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
mtx_lock(&sc->mfi_io_lock);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
TAILQ_FOREACH_SAFE(cm, &sc->mfi_busy, cm_link, tmp) {
First fix pr 167226: ThunderBolt cannot read sector >= 2^32 or 2^21 with supplied patch. Second the bigger change, fix RAID operation on ThunderBolt base card such as physically removing a disk from a RAID and replacing it. The current situation is the RAID firmware effectively hangs waiting for an acknowledgement from the driver. This is due to the firmware support of the driver actually accessing the RAID from under the firmware. This is an interesting feature that the FreeBSD driver does not use. However, when the firmare detects the driver has attached it then expects the driver will synchronize LD's with the firmware. If the driver does not sync. then the management part of the firmware will hang waiting for it so a pulled driver will listed as still there. The fix for this problem isn't extremely difficult. However, figuring out why some of the code was the way it was and then redoing it was involved. Not have a spec. made it harder to try to figure out. The existing driver would send a MFI_DCMD_LD_MAP_GET_INFO command in write mode to acknowledge a LD state change. In read mode it gets the RAID map from the firmware. The FreeBSD driver doesn't do that currently. It could be added in the future with the appropriate structures. To simplify things, get the current LD state and then build the MFI_DCMD_LD_MAP_GET_INFO/write command so that it sends an acknowledgement for each LD. The map would probably state which LD's changed so then the driver could probably just acknowledge the LD's that changed versus all. This doesn't seem to be a problem. When a MFI_DCMD_LD_MAP_GET_INFO/write command is sent to the firmware, it will complete later when a change to the LD's happen. So it is very much like an AEN command returning when something happened. When the MFI_DCMD_LD_MAP_GET_INFO/write command completes, we refire the sync'ing of the LD state. This needs to be done in as an event so that MFI_DCMD_LD_GET_LIST can wait for that command to complete before issuing the MFI_DCMD_LD_MAP_GET_INFO/write. The prior code didn't use the call-back function and tried to intercept the MFI_DCMD_LD_MAP_GET_INFO/write command when processing an interrupt. This added a bunch of code complexity to the interrupt handler. Using the call-back that is done for other commands got rid of this need. So the interrupt handler is greatly simplified. It seems that even commands that shouldn't be acknowledged end up in the interrupt handler. To deal with this, code was added to check to see if a command is in the busy queue or not. This might have contributed to the interrupt storm happening without MSI enabled on these cards. Note that MFI_DCMD_LD_MAP_GET_INFO/read returns right away. It would be interesting to see what other complexity could be removed from the ThunderBolt driver that really isn't needed in our mode of operation. Letting the RAID firmware do all of the I/O to disks is a lot faster since it can use its caches. It greatly simplifies what the driver has to do and potential bugs if the driver and firmware are not in sync. Simplify the aen_abort/cm_map_abort and put it in the softc versus in the command structure. This should get merged to 9 before the driver is merged to 8. PR: 167226 Submitted by: Petr Lampa MFC after: 3 days
2012-05-04 16:00:39 +00:00
if (sc->mfi_aen_cm == cm || sc->mfi_map_sync_cm == cm)
Skip the AEN event command that is always hanging out on the card. Submitted by: Doug Ambrisko
2006-09-27 05:00:10 +00:00
continue;
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
if (cm->cm_timestamp <= deadline) {
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
if (sc->adpreset != 0 && sc->issuepend_done == 0) {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
cm->cm_timestamp = time_uptime;
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
} else {
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
device_printf(sc->mfi_dev,
MFhead_mfi r227068 First cut of new HW support from LSI and merge into FreeBSD. Supports Drake Skinny and ThunderBolt cards. MFhead_mfi r227574 Style MFhead_mfi r227579 Use bus_addr_t instead of uintXX_t. MFhead_mfi r227580 MSI support MFhead_mfi r227612 More bus_addr_t and remove "#ifdef __amd64__". MFhead_mfi r227905 Improved timeout support from Scott. MFhead_mfi r228108 Make file. MFhead_mfi r228208 Fixed botched merge of Skinny support and enhanced handling in call back routine. MFhead_mfi r228279 Remove superfluous !TAILQ_EMPTY() checks before TAILQ_FOREACH(). MFhead_mfi r228310 Move mfi_decode_evt() to taskqueue. MFhead_mfi r228320 Implement MFI_DEBUG for 64bit S/G lists. MFhead_mfi r231988 Restore structure layout by reverting the array header to use [0] instead of [1]. MFhead_mfi r232412 Put wildcard pattern later in the match table. MFhead_mfi r232413 Use lower case for hexadecimal numbers to match surrounding style. MFhead_mfi r232414 Add more Thunderbolt variants. MFhead_mfi r232888 Don't act on events prior to boot or when shutting down. Add hw.mfi.detect_jbod_change to enable or disable acting on JBOD type of disks being added on insert and removed on removing. Switch hw.mfi.msi to 1 by default since it works better on newer cards. MFhead_mfi r233016 Release driver lock before taking Giant when deleting children. Use TAILQ_FOREACH_SAFE when items can be deleted. Make code a little simplier to follow. Fix a couple more style issues. MFhead_mfi r233620 Update mfi_spare/mfi_array with the actual number of elements for array_ref and pd. Change these max. #define names to avoid name space collisions. This will require an update to mfiutil It avoids mfiutil having to do a magic calculation. Add a note and #define to state that a "SYSTEM" disk is really what the firmware calls a "JBOD" drive. Thanks to the many that helped, LSI for the initial code drop, mav, delphij, jhb, sbruno that all helped with code and testing.
2012-03-30 23:05:48 +00:00
"COMMAND %p TIMEOUT AFTER %d SECONDS\n",
cm, (int)(time_uptime - cm->cm_timestamp)
);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
MFI_PRINT_CMD(cm);
MFI_VALIDATE_CMD(sc, cm);
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
/*
* Fail the command instead of leaving it on
* the queue where it could remain stuck forever
*/
mfi_remove_busy(cm);
cm->cm_error = ETIMEDOUT;
mfi_complete(sc, cm);
First cut at updating mfi(4) to support newer LSI MegaRAID SAS cards. Specifically, add support for "Drake Skinny" and "ThunderBolt" LSI cards. Initial code was supplied by LSI under BSD license. Several improvements were done by myself. Such things like making it work in a static kernel, be able to boot of the RAID, performance improvements. I removed some fairly complicated code that seemed to directly access the disks under the firmware. It doesn't seem to be needed and significantly slowed down the performance of the driver and caused tons of sense errors to be reported. This code is being checked in this area so others can help me get it into shape to commit into the FreeBSD tree. Assistance has been volunteered by iXsystems. We might want to re-work the JBOD attachment that creates /dev/mfisyspd? node for each disk. Performance is faster then prior cards. It works okay with WITNESS and INVARIANTS on amd64 and i386. I recall seeing a use after free time bug with FreeBSD 8 and a Drake Skinny card with WITNESS and INVARIANTS on. First task is get all of the new structures to be named in FreeBSD style format. Next is probably to deal with the 64bit addressing changes that are mostly around the #ifdef __amd64__ checks. Thanks to LSI for providing the initial code. Obtained from: LSI
2011-11-04 02:34:52 +00:00
timedout++;
}
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
}
}
#if 0
if (timedout)
Fixes queuing issues where mfi_release_command blindly sets the cm_flags = 0 without first removing the command from the relavent queue. This was causing panics in the queue functions which check to ensure a command is not on another queue. Fixed some cases where the error from mfi_mapcmd was lost and where the command was never released / dequeued in error cases. Ensure that all failures to mfi_mapcmd are logged. Fixed possible null pointer exception in mfi_aen_setup if mfi_get_log_state failed. Fixed mfi_parse_entries & mfi_aen_setup not returning possible errors. Corrected MFI_DUMP_CMDS calls with invalid vars SC vs sc. Commands which have timed out now set cm_error to ETIMEDOUT and call mfi_complete which prevents them getting stuck in the busy queue forever. Fixed possible use of NULL pointer in mfi_tbolt_get_cmd. Changed output formats to be more easily recognisable when debugging. Optimised mfi_cmd_pool_tbolt cleanup. Made information about driver limiting commands always display as for modern cards this can be severe. Fixed mfi_tbolt_alloc_cmd out of memory case which previously didnt return an error. Added malloc checks for request_desc_pool including free when subsiquent errors are detected. Fixed overflow error in SIMD reply descriptor check. Fixed tbolt_cmd leak in mfi_build_and_issue_cmd if there's an error during IO build. Elimintated double checks on sc->mfi_aen_cm & sc->mfi_map_sync_cm in mfi_shutdown. Move local hdr calculation after error check in mfi_aen_complete. Fixed wakeup on NULL in mfi_aen_complete. Fixed mfi_aen_cm cleanup in mfi_process_fw_state_chg_isr not checking if it was NULL. Changed mfi_alloc_commands to error if bus_dmamap_create fails. Previously we would try to continue with the number of allocated commands but lots of places in the driver assume sc->mfi_max_fw_cmds is whats available so its unsafe to do this without lots of changes. Removed mfi_total_cmds as its no longer used due the above change. Corrected mfi_tbolt_alloc_cmd to return ENOMEM where appropriate. Fixed timeouts actually firing at double what they should. Setting hw.mfi.max_cmds=-1 now configures to use the controller max. A few style (9) fixes e.g. braced single line conditions and double blank lines Cleaned up queuing macros Removed invalid queuing tests for multiple queues Trap and deal with errors when doing sends in mfi_data_cb Refactored frame sending into one method with error checking of the return code so we can ensure commands aren't left on the queue after error. This ensures that mfi_mapcmd & mfi_data_cb leave the queue in a valid state. Refactored how commands are cleaned up, mfi_release_command now ensures that all queues and command state is maintained in a consistent state. Prevent NULL pointer use in mfi_tbolt_complete_cmd Fixed use of NULL sc->mfi_map_sync_cm in wakeup Added defines to help with output of mfi_cmd and header flags. Fixed mfi_tbolt_init_MFI_queue invalidating cm_index of the acquired mfi_cmd. Reset now reinitialises sync map as well as AEN. Fixed possible use of NULL pointer in mfi_build_and_issue_cmd Fixed mfi_tbolt_init_MFI_queue call to mfi_process_fw_state_chg_isr causing panic on failure. Ensure that tbolt cards always initialise next_host_reply_index and free_host_reply_index (based off mfi_max_fw_cmds) on both startup and reset as per the linux driver. Fixed mfi_tbolt_complete_cmd not acknowledging unknown commands so it didn't clear the controller. Prevent locks from being dropped and re-acquired in the following functions which was allowing multiple threads to enter critical methods such as mfi_tbolt_complete_cmd & mfi_process_fw_state_chg_isr:- * mfi_tbolt_init_MFI_queue * mfi_aen_complete / mfi_aen_register * mfi_tbolt_sync_map_info * mfi_get_log_state * mfi_parse_entries The locking for these functions was promoting to higher level methods. This also fixed MFI_LINUX_SET_AEN_2 which was already acquiring the lock, so would have paniced for recursive lock. This also required changing malloc of ld_sync in mfi_tbolt_sync_map_info to M_NOWAIT which can hence now fail but this was already expected as its return was being tested. Removed the assignment of cm_index in mfi_tbolt_init_MFI_queue which breaks the world if the cmd returned by mfi_dequeue_free isn't the first cmd. Fixed locking in mfi_data_cb, this is an async callback from bus_dmamap_load which could hence be called after the caller has dropped the lock. If we don't have the lock we aquire it and ensure we unlock before returning. Fixed locking mfi_comms_init when mfi_dequeue_free fails. Fixed mfi_build_and_issue_cmd not returning tbolt cmds aquired to the pool on error. Fixed mfi_abort not dropping the io lock when mfi_dequeue_free fails. Added hw.mfi.polled_cmd_timeout sysctl that enables tuning of polled timeouts. This shouldn't be reduced below 50 seconds as its used for firmware patching which can take quite some time. Added hw.mfi.fw_reset_test sysctl which is avaliable when compiled with MFI_DEBUG and allows the testing of controller reset that was provoking a large number of the issues encountered here. Reviewed by: Doug Ambrisko Approved by: pjd (mentor) MFC after: 1 month
2013-02-27 02:21:10 +00:00
MFI_DUMP_CMDS(sc);
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
#endif
mtx_unlock(&sc->mfi_io_lock);
Adds hw.mfi.cmd_timeout loader / sysctl tuneable which controls the default timeout used in the mfi driver. This is useful for long running commands such as secure erase. Reviewed by: John Baldwin Approved by: pjd (mentor)
2013-02-27 23:17:45 +00:00
callout_reset(&sc->mfi_watchdog_callout, mfi_cmd_timeout * hz,
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
mfi_timeout, sc);
- Add a command validator for use in debugging. - Fix the locking protocol to eliminate races between normal I/O and AENs. - Various small improvements and usability tweaks. Sponsored by: IronPort Portions Submitted by: Doug Ambrisko
2006-10-16 04:18:38 +00:00
if (0)
mfi_dump_all();
Add a command debugging module and a periodic watchdog timer. Sponsored by: IronPort
2006-09-25 11:35:34 +00:00
return;
}
Reference in New Issue Copy Permalink