mostly by adjustments to debugging printf() format specifiers. For high
numbered LUNs, also switch to printing them in hex as per SAM-5.
MFC after: 2 weeks
to not get scanned on boot.
The problem originated in change 253549. With the change to the mps(4)
driver to scan only targets that it knows it has (as opposed to scanning
the entire bus), scanning RAID volumes on boot was omitted.
So, for versions of FreeBSD that have the scanning changes
(__FreeBSD_version 1000039 and higher), scan RAID volumes that are added
whether or not we're booting.
PR: kern/181784
Reported by: Xiguang Wang <kurapica@gmail.com>
Tested by: Dennis Glatting <dg@pki2.com>
Sponsored by: Spectra Logic
Approved by: re (delphij)
MFC After: 3 days
manifested itself in out of chain frame conditions.
When the driver ran out of chain frames, the request in question
would get completed early, and go through mpssas_scsiio_complete().
In mpssas_scsiio_complete(), the negation of the CAM status values
(CAM_STATUS_MASK | CAM_SIM_QUEUED) was ORed in instead of being
ANDed in. This resulted in a bogus CAM CCB status value. This
didn't show up in the non-error case, because the status was reset
to something valid (e.g. CAM_REQ_CMP) later on in the function.
But in the error case, such as when the driver ran out of chain
frames, the CAM_REQUEUE_REQ status was ORed in to the bogus status
value. This led to the CAM transport layer repeatedly releasing
the SIM queue, because it though that the CAM_RELEASE_SIMQ flag had
been set. The symptom was messages like this on the console when
INVARIANTS were enabled:
xpt_release_simq: requested 1 > present 0
xpt_release_simq: requested 1 > present 0
xpt_release_simq: requested 1 > present 0
mps_sas.c: In mpssas_scsiio_complete(), use &= to take status
bits out. |= adds them in.
In the error case in mpssas_scsiio_complete(), set
the status to CAM_REQUEUE_REQ, don't OR it in.
MFC after: 3 days
Sponsored by: Spectra Logic
command register. The lazy BAR allocation code in FreeBSD sometimes
disables this bit when it detects a range conflict, and will re-enable
it on demand when a driver allocates the BAR. Thus, the bit is no longer
a reliable indication of capability, and should not be checked. This
results in the elimination of a lot of code from drivers, and also gives
the opportunity to simplify a lot of drivers to use a helper API to set
the busmaster enable bit.
This changes fixes some recent reports of disk controllers and their
associated drives/enclosures disappearing during boot.
Submitted by: jhb
Reviewed by: jfv, marius, achadd, achim
MFC after: 1 day
Add a tunable that allows such a device to be excluded from the driver.
The id parameter is the target id that the driver assigns to a given device.
dev.mps.X.exclude_ids=<id>,<id>
Obtained from: Netflix
MFC after: 3 days
---------------------------------------------------------------
System panics during a Port reset with ouststanding I/O
---------------------------------------------------------------
It is possible to call mps_mapping_free_memory after this
memory is already freed, causing a panic. Removed this extra
call to mps_mappiing_free_memory and call mps_mapping_exit
in place of the mps_mapping_free_memory call so that any
outstanding mapping items can be flushed before memory is
freed.
---------------------------------------------------------------
Correct memory leak during a Port reset with ouststanding I/O
---------------------------------------------------------------
In mps_reinit function, the mapping memory was not being
freed before being re-allocated. Added line to call the
memory free function for mapping memory.
---------------------------------------------------------------
Use CAM_SIM_QUEUED flag in Driver IO path.
---------------------------------------------------------------
This flag informs the XPT that successful abort of a CCB
requires an abort ccb to be issued to the SIM. While
processing SCSI IO's, set the CAM_SIM_QUEUED flag in the
status for the IO. When the command completes, clear this
flag.
---------------------------------------------------------------
Check for CAM_REQ_INPROG in I/O path.
---------------------------------------------------------------
Added a check in mpssas_action_scsiio for the In Progress
status for the IO. If this flag is set, the IO has already
been aborted by the upper layer (before CAM_SIM_QUEUED was
set) and there is no need to send the IO. The request will
be completed without error.
---------------------------------------------------------------
Improve "doorbell handshake method" for mps_get_iocfacts
---------------------------------------------------------------
Removed call to get Port Facts since this information is
not used currently.
Added mps_iocfacts_allocate function to allocate memory
that is based on IOC Facts data. Added mps_iocfacts_free
function to free memory that is based on IOC Facts data.
Both of the functions are used when a Diag Reset is performed
or when the driver is attached/detached. This is needed in
case IOC Facts changes after a Diag Reset, which could
happen if FW is upgraded.
Moved call of mps_bases_static_config_pages from the attach
routine to after the IOC is ready to process accesses based
on the new memory allocations (instead of polling through
the Doorbell).
---------------------------------------------------------------
Set TimeStamp in INIT message in millisecond format Set the IOC
---------------------------------------------------------------
---------------------------------------------------------------
Prefer mps_wait_command to mps_request_polled
---------------------------------------------------------------
Instead of using mps_request_polled, call mps_wait_command
whenever possible. Change the mps_wait_command function to
check the current context and either use interrupt context
or poll if required by using the pause or DELAY function.
Added a check after waiting 50mSecs to see if the command
has timed out. This is only done if polliing, the msleep
command will automatically timeout if the command has taken
too long to complete.
---------------------------------------------------------------
Integrated RAID: Volume Activation Failed error message is
displayed though the volume has been activated.
---------------------------------------------------------------
Instead of failing an IOCTL request that does not have a
large enough buffer to hold the complete reply, copy as
much data from the reply as possible into the user's buffer
and log a message saying that the user's buffer was smaller
than the returned data.
---------------------------------------------------------------
mapping_add_new_device failure due to persistent table FULL
---------------------------------------------------------------
When a new device is added, if it is determined that the
device persistent table is being used and is full, instead
of displaying a message for this condition every time, only
log a message if the MPS_INFO bit is set in the debug_flags.
Submitted by: LSI
MFC after: 1 week
Add a PIM_NOSCAN flag to the CAM path inquiry CCB. This tells CAM
not to perform a rescan on a bus when it is registered.
We now use this flag in the mps(4) driver. Since it knows what
devices it has attached, it is more efficient for it to just issue
a target rescan on the targets that are attached.
Also, remove the private rescan thread from the mps(4) driver in
favor of the rescan thread already built into CAM. Without this
change, but with the change above, the MPS scanner could run before
or during CAM's initial setup, which would cause duplicate device
reprobes and announcements.
sys/param.h:
Bump __FreeBSD_version to 1000039 for the inclusion of the
PIM_RESCAN CAM path inquiry flag.
sys/cam/cam_ccb.h:
sys/cam/cam_xpt.c:
Added a PIM_NOSCAN flag. If a SIM sets this in the path
inquiry ccb, then CAM won't rescan the bus in
xpt_bus_regsister.
sys/dev/mps/mps_sas.c
For versions of FreeBSD that have the PIM_NOSCAN path
inquiry flag, don't freeze the sim queue during scanning,
because CAM won't be scanning this bus. Instead, hold
up the boot. Don't call mpssas_rescan_target in
mpssas_startup_decrement; it's redundant and I don't
know why it was in there.
Set PIM_NOSCAN in path inquiry CCBs.
Remove methods related to the internal rescan daemon.
Always use async events to trigger a probe for EEDP support.
In older versions of FreeBSD where AC_ADVINFO_CHANGED is
not available, use AC_FOUND_DEVICE and issue the
necessary READ CAPACITY manually.
Provide a path to xpt_register_async() so that we only
receive events for our own SCSI domain.
Improve error reporting in cases where setup for EEDP
detection fails.
sys/dev/mps/mps_sas.h:
Remove softc flags and data related to the scanner thread.
sys/dev/mps/mps_sas_lsi.c:
Unconditionally rescan the target whenever a device is added.
Sponsored by: Spectra Logic
MFC after: 1 week
sys/dev/mps/mps_user.c
Fix uninitialized memory reference in mps_read_config_page. It was
referencing a field (params->hdr.Ext.ExtPageType) that would only be
set when reading an Extended config page. The symptom was that
MPSIO_READ_CFG_PAGE ioctls would randomly fail with
MPI2_IOCSTATUS_CONFIG_INVALID_PAGE errors. The solution is to
determine whether an extended or an ordinary config page is requested
by looking at the PageType field, which should be available regardless.
Similarly, mps_user_read_extcfg_header and mps_user_read_extcfg_page,
which call mps_read_config_page, had to be fixed to always set the
PageType field. They were implicitly assuming that
mps_read_config_page always operated on Extended pages.
Reviewed by: ken
Approved by: ken (mentor)
MFC after: 3 days
Stop abusing xpt_periph in random plases that really have no periph related
to CCB, for example, bus scanning. NULL value is fine in such cases and it
is correctly logged in debug messages as "noperiph". If at some point we
need some real XPT periphs (alike to pmpX now), quite likely they will be
per-bus, and not a single global instance as xpt_periph now.
unmapped I/O. That one exception is access to INQUIRY VPD request result.
Those requests are never unmapped now, but to be safe add respective check
there and allow unmapped I/O for the SIM by setting PIM_UNMAPPED flag.
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>)
- remove special handling of zero length transfers in mpi_pre_fw_upload();
- add missing MPS_CM_FLAGS_DATAIN flag in mpi_pre_fw_upload();
- move mps_user_setup_request() call into proper place;
- increase user command timeout from 30 to 60 seconds;
- avoid NULL dereference panic in case of firmware crash.
Set max DMA segment size to 24bit, as MPI SGE supports it.
Use mps_add_dmaseg() to add empty SGE instead of custom code.
Tune endianness safety.
Reviewed by: Desai, Kashyap <Kashyap.Desai@lsi.com>
Sponsored by: iXsystems, Inc.
14.00.00.01-fbsd.
Their description of the changes is as follows:
1. Copyright contents has been changed in all respective .c
and .h files
2. Support for WRITE12 and READ12 for direct-io (warpdrive only)
has been added.
3. Driver has added checks to see if Drive has READ_CAP_16
support before sending it down to the device.
If SPC3_SID_PROTECT flag is set in the inquiry data, the
device supports protection information, and must support
the 16 byte read capacity command, otherwise continue without
sending read cap 16. This will optimize driver performance,
since it will not send READ_CAP_16 to the drive which does
not have support of READ_CAP_16.
4. With new approach, "MPTIOCTL_RESET_ADAPTER" IOCTL will not
use DELAY() which is busy loop implementation.
It will use <msleep> (Better way to sleep without busy
loop). Also from the HBA reset code path and some other
places, DELAY() is replaced with msleep() or "pause()",
which is based on sleep/wakeup style calls. Driver use
msleep()/pause() instead of DELAY based on CAN_SLEEP/NO_SLEEP
flags to avoid busy loop which is not required all the
time.e.a
a. While driver is getting loaded, driver calls most of the
commands with NO_SLEEP.
b. When Driver is functional and it needs Reinit of HBA,
CAN_SLEEP flag is used.
5. <mpslsi> driver is not Endian safe. It will not work on Big
Endian machines like Sparc and PowerPC platforms because it
assumes it is running on a Little Endian machine.
Driver code is modified such way that it does not assume CPU
arch is Little Endian.
a. All places where Driver interacts from HBA to Host, it
converts Little Endian format to CPU format.
b. All places where Driver interacts from Host to HBA, it
converts CPU format to Little Endian.
6. Findout memory leaks in FreeBSD Driver and resolve those,
such as memory leak in targ's luns creation/deletion.
Also added additional checks to see memory allocation
success/fail.
7. Add loginfo prints as debug message, i.e. When FW sends any
loginfo, Driver should print those as debug message.
This will help for debugging purpose.
8. There is possibility to get config request timeout. Current
driver is able to detect config request timetout, but it does
not do anything on config_request timeout. Driver should
call mps_reinit() if any request_poll (which is called as
part of config_request) is time out.
9. cdb length check is required for 32 byte CDB. Add correct mpi
control value for 32 bit CDB as below while submitting SCSI IO
Request to controller.
mpi_control |= 4 << MPI2_SCSIIO_CONTROL_ADDCDBLEN_SHIFT;
10. Check the actual status of Message unit reset
(mps_message_unit_reset).Previously FreeBSD Driver just writes
MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET and never check the ack
(it just wait for 50 millisecond). So, Driver now check the
status of "MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET" after writing
it to the FW.
Now it also checking for whether doorbell ack uses msleep with
proper sleep flags, instead of <DELAY>.
11. Previously CAM does not detect Multi-Lun Devices. In order to
detect Multi-Lun Devices by CAM the driver needs following change
set:
a. There is "max_lun" field which Driver need to set based on
hw/fw support. Currently LSI released driver does not set
this field.
b. Default of "max_lun" should not be 0 in OS, but it is
currently set to 0 in CAM layer.
c. Export max_lun capacity to 255
12. Driver will not reset target info after port enable complete and
also do Device removal when Device remove from FW. The detail
description is as follows
a. When Driver receive WD PD add events, it will add all
information in driver local data structure.
b. Only for WD, we have below checks after port enable
completes, where driver clear off all information retrieved
at #1.
if ((sc->WD_available &&
(sc->WD_hide_expose == MPS_WD_HIDE_ALWAYS)) ||
(sc->WD_valid_config && (sc->WD_hide_expose ==
MPS_WD_HIDE_IF_VOLUME)) {
// clear off target data structure.
}
It is mainly not to attach PDs to OS.
FreeBSD does bus rescan as older Parallel scsi style. So Driver
needs to handle which Drive is visible to OS. That is a reason
we have to clear off targ information for PDs.
Again, above logic was implemented long time ago. Similar concept
we have for non-wd also. For that, LSI have introduced different
logic to hide PDs.
Eventually, because of above gap, when Phy goes offline, we
observe below failure. That is what Driver is not doing complete
removal of device with FW. (which was pointed by Scott)
Apr 5 02:39:24 Freebsd7 kernel: mpslsi0: mpssas_prepare_remove
Apr 5 02:39:24 Freebsd7 kernel: mpssas_prepare_remove 497 : invalid handle 0xe
Now Driver will not reset target info after port enable complete
and also will do Device removal when Device remove from FW.
13. Returning "CAM_SEL_TIMEOUT" instead of "CAM_TID_INVALID"
error code on request to the Target IDs that have no devices
conected at that moment. As if "CAM_TID_INVALID" error code
is returned to the CAM Layaer then it results in a huge chain
of errors in verbose kernel messages on boot and every
hot-plug event.
Submitted by: Sreekanth Reddy <Sreekanth.Reddy@lsi.com>
MFC after: 3 days
This allows LUNs greater than 0 to be probed. It can be increased later if
need be.
This brings back SVN rev 224973, which was inadvertently removed with the
import of the LSI driver.
Reported by: dwhite
MFC after: 3 days
1. Fixed timeout specification for the msleep in mps_wait_command().
Added 30 second timeout for mps_wait_command() calls in mps_user.c.
2. Make sure we call mps_detach_user() from the kldunload path.
3. Raid Hotplug behavior change.
The driver now removes a volume when it goes to a failed state,
so we also need to add volume back to the OS when it goes to
opitimal/degraded/online from failed/missing.
Handle raid volume add and remove from the IR_Volume event.
4. Added some more debugging information.
5. Replace xpt_async(AC_LOST_DEVICE, path, NULL) with
mpssas_rescan_target().
This is to work around a panic in CAM that shows up when adding a
drive with a rescan and removing another device from the driver thread
with an AC_LOST_DEVICE async notification.
This problem was encountered in testing with the LSI sas2ircu utility,
which was used to create a RAID volume from physical disks. The driver
has to create the RAID volume target and remove the physical disk
targets, and triggered a panic in the process.
The CAM issue needs to be fully diagnosed and fixed, but this works
around the issue for now.
6. Fix some memory initialization issues in mps_free_command().
7. Resolve the "devq freeze forever" issue. This was caused by the
internal read capacity command issued in the non-head version of the
driver. When the command completed with an error, the driver wasn't
unfreezing thd device queue.
The version in head uses the CAM infrastructure for getting the read
capacity information, and therefore doesn't have the same issue.
8. Bump the version to 13.00.00.00-fbsd. (this is very close to LSI's
internal stable driver 13.00.00.00)
Submitted by: Kashyap Desai <Kashyap.Desai@lsi.com>
MFC after: 3 days
This involves significant changes to the mps(4) driver, but is not a
complete rewrite.
Some of the changes in this version of the driver:
- Integrated RAID (IR) support.
- Support for WarpDrive controllers.
- Support for SCSI protection information (EEDP).
- Support for TLR (Transport Level Retries), needed for tape drives.
- Improved error recovery code.
- ioctl interface compatible with LSI utilities.
mps.4: Update the mps(4) driver man page somewhat for the driver
changes. The list of supported hardware still needs to be
updated to reflect the full list of supported cards.
conf/files: Add the new driver files.
mps/mpi/*: Updated version of the MPI header files, with a BSD style
copyright.
mps/*: See above for a description of the new driver features.
modules/mps/Makefile:
Add the new mps(4) driver files.
Submitted by: Kashyap Desai <Kashyap.Desai@lsi.com>
Reviewed by: ken
MFC after: 1 week
no reason why it should be limited to 64K of DFLTPHYS. DMA data tag is any
way set to allow MAXPHYS, S/G lists (chain elements) are sufficient and
overflows are also handled. On my tests even 1MB I/Os are working fine.
Reviewed by: ken@
one. Interestingly, these are actually the default for quite some time
(bus_generic_driver_added(9) since r52045 and bus_generic_print_child(9)
since r52045) but even recently added device drivers do this unnecessarily.
Discussed with: jhb, marcel
- While at it, use DEVMETHOD_END.
Discussed with: jhb
- Also while at it, use __FBSDID.
Zero any sense not transferred by the device as the SCSI specification
mandates that any untransferred data should be assumed to be zero.
Reviewed by: ken
CAM.
Desriptor sense is a new sense data format that originated in SPC-3. Among
other things, it allows for an 8-byte info field, which is necessary to
pass back block numbers larger than 4 bytes.
This change adds a number of new functions to scsi_all.c (and therefore
libcam) that abstract out most access to sense data.
This includes a bump of CAM_VERSION, because the CCB ABI has changed.
Userland programs that use the CAM pass(4) driver will need to be
recompiled.
camcontrol.c: Change uses of scsi_extract_sense() to use
scsi_extract_sense_len().
Use scsi_get_sks() instead of accessing sense key specific
data directly.
scsi_modes: Update the control mode page to the latest version (SPC-4).
scsi_cmds.c,
scsi_target.c: Change references to struct scsi_sense_data to struct
scsi_sense_data_fixed. This should be changed to allow the
user to specify fixed or descriptor sense, and then use
scsi_set_sense_data() to build the sense data.
ps3cdrom.c: Use scsi_set_sense_data() instead of setting sense data
manually.
cam_periph.c: Use scsi_extract_sense_len() instead of using
scsi_extract_sense() or accessing sense data directly.
cam_ccb.h: Bump the CAM_VERSION from 0x15 to 0x16. The change of
struct scsi_sense_data from 32 to 252 bytes changes the
size of struct ccb_scsiio, but not the size of union ccb.
So the version must be bumped to prevent structure
mis-matches.
scsi_all.h: Lots of updated SCSI sense data and other structures.
Add function prototypes for the new sense data functions.
Take out the inline implementation of scsi_extract_sense().
It is now too large to put in a header file.
Add macros to calculate whether fields are present and
filled in fixed and descriptor sense data
scsi_all.c: In scsi_op_desc(), allow the user to pass in NULL inquiry
data, and we'll assume a direct access device in that case.
Changed the SCSI RESERVED sense key name and description
to COMPLETED, as it is now defined in the spec.
Change the error recovery action for a number of read errors
to prevent lots of retries when the drive has said that the
block isn't accessible. This speeds up reconstruction of
the block by any RAID software running on top of the drive
(e.g. ZFS).
In scsi_sense_desc(), allow for invalid sense key numbers.
This allows calling this routine without checking the input
values first.
Change scsi_error_action() to use scsi_extract_sense_len(),
and handle things when invalid asc/ascq values are
encountered.
Add a new routine, scsi_desc_iterate(), that will call the
supplied function for every descriptor in descriptor format
sense data.
Add scsi_set_sense_data(), and scsi_set_sense_data_va(),
which build descriptor and fixed format sense data. They
currently default to fixed format sense data.
Add a number of scsi_get_*() functions, which get different
types of sense data fields from either fixed or descriptor
format sense data, if the data is present.
Add a number of scsi_*_sbuf() functions, which print
formatted versions of various sense data fields. These
functions work for either fixed or descriptor sense.
Add a number of scsi_sense_*_sbuf() functions, which have a
standard calling interface and print the indicated field.
These functions take descriptors only.
Add scsi_sense_desc_sbuf(), which will print a formatted
version of the given sense descriptor.
Pull out a majority of the scsi_sense_sbuf() function and
put it into scsi_sense_only_sbuf(). This allows callers
that don't use struct ccb_scsiio to easily utilize the
printing routines. Revamp that function to handle
descriptor sense and use the new sense fetching and
printing routines.
Move scsi_extract_sense() into scsi_all.c, and implement it
in terms of the new function, scsi_extract_sense_len().
The _len() version takes a length (which should be the
sense length - residual) and can indicate which fields are
present and valid in the sense data.
Add a couple of new scsi_get_*() routines to get the sense
key, asc, and ascq only.
mly.c: Rename struct scsi_sense_data to struct
scsi_sense_data_fixed.
sbp_targ.c: Use the new sense fetching routines to get sense data
instead of accessing it directly.
sbp.c: Change the firewire/SCSI sense data transformation code to
use struct scsi_sense_data_fixed instead of struct
scsi_sense_data. This should be changed later to use
scsi_set_sense_data().
ciss.c: Calculate the sense residual properly. Use
scsi_get_sense_key() to fetch the sense key.
mps_sas.c,
mpt_cam.c: Set the sense residual properly.
iir.c: Use scsi_set_sense_data() instead of building sense data by
hand.
iscsi_subr.c: Use scsi_extract_sense_len() instead of grabbing sense data
directly.
umass.c: Use scsi_set_sense_data() to build sense data.
Grab the sense key using scsi_get_sense_key().
Calculate the sense residual properly.
isp_freebsd.h: Use scsi_get_*() routines to grab asc, ascq, and sense key
values.
Calculate and set the sense residual.
MFC after: 3 days
Sponsored by: Spectra Logic Corporation
mps.c: Hide the 'out of chain frames' warning behind MPS_INFO.
mps_sas.c: Hide the SIM queue freeze/unfreeze messages behind MPS_INFO.
mpsvar.h: Bump the number of chain frames from 1024 to 2048. From
testing, it looks like this makes it less likely that we'll
run out of chain frames, and it doesn't cost much memory
(32K).
MFC after: 3 days
When the driver ran out of DMA chaining buffers, it kept the timeout for
the I/O, and I/O would stall.
The driver was not freezing the device queue on errors.
mps.c: Pull command completion logic into a separate
function, and call the callback/wakeup for commands
that are never sent due to lack of chain buffers.
Add a number of extra diagnostic sysctl variables.
Handle pre-hardware errors for configuration I/O.
This doesn't panic the system, but it will fail the
configuration I/O and there is no retry mechanism.
So the device probe will not succeed. This should
be a very uncommon situation, however.
mps_sas.c: Freeze the SIM queue when we run out of chain
buffers, and unfreeze it when more commands
complete.
Freeze the device queue when errors occur, so that
CAM can insure proper command ordering.
Report pre-hardware errors for task management
commands. In general, that shouldn't be possible
because task management commands don't have S/G
lists, and that is currently the only error path
before we get to the hardware.
Handle pre-hardware errors (like out of chain
elements) for SMP requests. That shouldn't happen
either, since we should have enough space for two
S/G elements in the standard request.
For commands that end with
MPI2_IOCSTATUS_SCSI_IOC_TERMINATED and
MPI2_IOCSTATUS_SCSI_EXT_TERMINATED, return them
with CAM_REQUEUE_REQ to retry them unconditionally.
These seem to be related to back end, transport
related problems that are hopefully transient. We
don't want to go through the retry count for
something that is not a permanent error.
Keep track of the number of outstanding I/Os.
mpsvar.h: Track the number of free chain elements.
Add variables for the number of outstanding I/Os,
and I/O high water mark.
Add variables to track the number of free chain
buffers and the chain low water mark, as well as
the number of chain allocation failures.
Add I/O state flags and an attach done flag.
MFC after: 3 days
the controller firmware will return all of our commands. Instead, keep
track of outstanding I/Os and return them to CAM once device removal
processing completes.
mpsvar.h: Declare the new "io_list" in the mps_softc.
mps.c: Initialize the new "io_list" in the mps softc.
mps_sas.c: o Track SCSI I/O requests on the io_list from the
time of mpssas_action() through mpssas_scsiio_complete().
o Zero out the request structures used for device
removal commands prior to filling them out.
o Once the target reset task management function completes
during device removal processing, assume any SCSI I/O
commands that are still oustanding will never return
from the controller, and process them manually.
Submitted by: gibbs
MFC after: 3 days
Prior to this change, the addressing method wasn't getting set, and
so the LUN field could be set incorrectly in some instances.
This fix should allow for LUN numbers up to 16777215 (and return an error
for anything larger, which wouldn't fit into the flat addressing model).
Submitted by: scottl (in part)
This bug manifested itself after repeated device arrivals and
departures. The root of the problem was that the last entry in the
reply array wasn't initialized/allocated. So every time we got
around to that event, we had a bogus address.
There were a couple more problems with the code that are also fixed:
- The reply mechanism was being treated as sequential (indexed by
sc->replycurindex) even though the spec says that the driver
should use the ReplyFrameAddress field of the post queue
descriptor to figure out where the reply is. There is no
guarantee that the reply descriptors will be used in sequential
order.
- The second word of the reply post queue descriptor wasn't being
checked in mps_intr_locked() to make sure that it wasn't
0xffffffff. So the driver could potentially come across a
partially DMAed descriptor.
- The number of replies allocated was one less than the actual
size of the queue. Instead, it was the size of the number of
replies that can be used at one time. (Which is one less than
the size of the queue.)
mps.c: When initializing the entries in the reply free
queue, make sure we initialize the full number that
we tell the chip we have (sc->fqdepth), not the
number that can be used at any one time (sc->num_replies).
When allocating replies, make sure we allocate the
number of replies that we've told the chip exist,
not just the number that can be used simultaneously.
Use the ReplyFrameAddress field of the post queue
descriptor to figure out which reply is being
referenced. This is what the spec says to do, and
the spec doesn't guarantee that the replies will be
used in order.
Put a check in to verify that the reply address passed
back from the card is valid. (Panic if it isn't, we'll
panic when we try to deference the reply pointer in any
case.)
In mps_intr_locked(), verify that the second word of the
post queue descriptor is not 0xffffffff in addition to
verifying that the unused flag is not set, so we can
make sure we didn't get a partially DMAed descriptor.
Remove references to sc->replycurindex, it isn't needed
now.
mpsvar.h: Remove replycurindex from the softc, it isn't needed now.
Reviewed by: scottl
