Conditional statements are faster than indirect calls. Use a structure
member to track the SCSI command submitter such that later patches can call
scsi_done(scmd) instead of scmd->scsi_done(scmd).
The asymmetric behavior that scsi_send_eh_cmnd() sets the submission
context to the SCSI error handler and that it does not restore the
submission context to the SCSI core is retained.
Link: https://lore.kernel.org/r/20211007202923.2174984-2-bvanassche@acm.org
Cc: Hannes Reinecke <hare@suse.com>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The following query shows which drivers define callbacks that are called by
the power management support code in the SCSI core (scsi_pm.c):
$ git grep -nHEwA16 "$(echo $(git grep -h 'scsi_register_driver(&' |
sed 's/.*&//;s/\..*//') | sed 's/ /|/g')" |
grep '\.pm[[:blank:]]*=[[:blank:]]'
drivers/scsi/sd.c-620- .pm = &sd_pm_ops,
drivers/scsi/sr.c-100- .pm = &sr_pm_ops,
drivers/scsi/ufs/ufshcd.c-9765- .pm = &ufshcd_wl_pm_ops,
Since unconditionally runtime resuming a device during system resume is not
necessary, remove that code. Modify the SCSI disk (sd) driver such that it
follows the same approach as the UFS driver, namely to skip system suspend
and resume for devices that are runtime suspended. The CD-ROM code does not
need to be updated since its PM callbacks do not affect the device power
state.
This patch has been tested as follows:
[ shell 1 ]
cd /sys/kernel/debug/tracing
grep -E 'blk_(pre|post)_runtime|runtime_(suspend|resume)|autosuspend_delay|pm_runtime_(get|put)' available_filter_functions |
while read a b; do echo "$a"; done |
grep -v __pm_runtime_resume >set_ftrace_filter
echo function > current_tracer
echo 1 > tracing_on
cat trace_pipe
[ shell 2 ]
cd /sys/block/sr0
# Increase the event poll interval to make it easier to derive from the
# tracing output whether runtime power actions are the result of sg_inq.
echo 30000 > events_poll_msecs
cd device/power
# Enable runtime power management.
echo auto > control
echo 1000 > autosuspend_delay_ms
sleep 1
# Verify in shell 1 that sr0 has been runtime suspended
sg_inq /dev/sr0
eject /dev/sr0
sg_inq /dev/sr0
# Disable runtime power management.
echo on > control
cd /sys/block/sda/device/power
echo auto > control
echo 1000 > autosuspend_delay_ms
sleep 1
# Verify in shell 1 that sr0 has been runtime suspended
sg_inq /dev/sda
Link: https://lore.kernel.org/r/20211006215453.3318929-4-bvanassche@acm.org
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Martin Kepplinger <martin.kepplinger@puri.sm>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If the softreset fails in the I_T reset, libsas will then continue to issue
a controller reset to try to recover.
However a faulty disk may cause the softreset to fail, and resetting the
controller will not help this scenario. Indeed, we will just continue the
cycle of error handle handling to try to recover.
So if the softreset fails upon certain conditions, just disable the phy
associated with the disk. The user needs to handle this problem.
Link: https://lore.kernel.org/r/1634041588-74824-5-git-send-email-john.garry@huawei.com
Signed-off-by: Luo Jiaxing <luojiaxing@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When issuing a hardreset/linkreset/phy_set_linkrate from sysfs, the phy
will be disabled and re-enabled for the directly attached scenario.
It takes some time for the phy to come back up after re-enabling the phy.
If the controller becomes suspended while waiting for the phy to come back,
the phy up may be lost (along with the disk).
To solve this problem, wait for the phy up to occur with a timeout. Indeed
this is already done in hisi_sas_debug_I_T_nexus_reset() for local phys, so
just relocate the functionality to hisi_sas_control_phy().
Since the HA workqueue is drained when suspending the controller, and the
phy control function is called from the same workqueue, we can guarantee
that the controller will not be suspended during this period.
Link: https://lore.kernel.org/r/1634041588-74824-3-git-send-email-john.garry@huawei.com
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The SCSI error handler calls scsi_unjam_host() which can call the queue
function ufshcd_queuecommand() indirectly. The error handler changes the
state to UFSHCD_STATE_RESET while running, but error interrupts that
happen while the error handler is running could change the state to
UFSHCD_STATE_EH_SCHEDULED_NON_FATAL which would allow requests to go
through ufshcd_queuecommand() even though the error handler is running.
Block that hole by checking whether the error handler is in progress.
Link: https://lore.kernel.org/r/20211008084048.257498-1-adrian.hunter@intel.com
Reviewed-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This variable is just a temporary variable, used to do an endian
conversion. The problem is that the last byte is not initialized. After
the conversion is completely done, the last byte is discarded so it doesn't
cause a problem. But static checkers and the KMSan runtime checker can
detect the uninitialized read and will complain about it.
Link: https://lore.kernel.org/r/20211006073242.GA8404@kili
Fixes: 5036f0a0ec ("[SCSI] csiostor: Fix sparse warnings.")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Merge the 5.15/scsi-fixes branch into the staging tree to resolve UFS
conflict reported by sfr.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Stop the OS from re-discovering multiple LUNs for tape drive and medium
changer.
Duplicate device nodes for Ultrium tape drive and medium changer are being
created.
The Ultrium tape drive is a multi-LUN SCSI target. It presents a LUN for
the tape drive and a 2nd LUN for the medium changer. Our controller FW
lists both LUNs in the RPL results.
As a result, the smartpqi driver exposes both devices to the OS. Then the
OS does its normal device discovery via the SCSI REPORT LUNS command, which
causes it to re-discover both devices a 2nd time, which results in the
duplicate device nodes.
Link: https://lore.kernel.org/r/20210928235442.201875-10-don.brace@microchip.com
Reviewed-by: Scott Benesh <scott.benesh@microchip.com>
Reviewed-by: Scott Teel <scott.teel@microchip.com>
Reviewed-by: Mike McGowen <mike.mcgowen@microchip.com>
Acked-by: John Donnelly <john.p.donnelly@oracle.com>
Signed-off-by: Kevin Barnett <kevin.barnett@microchip.com>
Signed-off-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Send a TEST UNIT READY to HBA disks and do not present them to the OS if
0x02/0x04/0x1b (SANITIZE IN PROGRESS) is returned.
During boot-up, some OSes appear to hang when there are one or more disks
undergoing a sanitize operation.
According to SCSI SBC4 specification section 4.11.2 "Commands allowed
during SANITIZE", some SCSI commands are permitted, but read/write
operations are not.
When the OS attempts to read the disk partition table a CHECK CONDITION ASC
0x04 ASCQ 0x1b is returned which causes the OS to retry the read until
SANITIZE has completed. This can take hours.
According to document HPE Smart Storage Administrator User Guide, during
the sanitize erase operation, the drive is unusable. I.e. the expected
behavior for SANITIZE is the that disk remains offline even after SANITIZE
has completed. The customer is expected to re-enable the disk using the
management utility.
Link: https://lore.kernel.org/r/20210928235442.201875-6-don.brace@microchip.com
Reviewed-by: Scott Benesh <scott.benesh@microchip.com>
Reviewed-by: Scott Teel <scott.teel@microchip.com>
Reviewed-by: Mike McGowen <mike.mcgowen@microchip.com>
Acked-by: John Donnelly <john.p.donnelly@oracle.com>
Signed-off-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Enhance check for commands queued to the controller. Add new function
pqi_nonempty_inbound_queue_count() that will wait for all I/O queued for
submission to controller across all queue groups to drain. Add helper
functions to obtain queue command counts for each queue group. These
queues should drain quickly as they are already staged to be submitted down
to the controller's IB queue.
Enhance check for outstanding command completion. Update the count of
outstanding commands while waiting. This value was not re-obtained and was
potentially causing infinite wait for all completions.
Link: https://lore.kernel.org/r/20210928235442.201875-5-don.brace@microchip.com
Reviewed-by: Scott Benesh <scott.benesh@microchip.com>
Reviewed-by: Scott Teel <scott.teel@microchip.com>
Reviewed-by: Mike McGowen <mike.mcgowen@microchip.com>
Acked-by: John Donnelly <john.p.donnelly@oracle.com>
Signed-off-by: Kevin Barnett <kevin.barnett@microchip.com>
Signed-off-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Correct kdump hangs when controller is locked up.
There are occasions when a controller reboot (controller soft reset) is
issued when a controller firmware crash dump is in progress.
This leads to incomplete controller firmware crash dump:
- When the controller crash dump is in progress, and a kdump is initiated,
the driver issues inbound doorbell reset to bring back the controller in
SIS mode.
- If the controller is in locked up state, the inbound doorbell reset does
not work causing controller initialization failures. This results in the
driver hanging waiting for SIS mode.
To avoid an incomplete controller crash dump, add in a controller crash
dump handshake:
- Controller will indicate start and end of the controller crash dump by
setting some register bits.
- Driver will look these bits when a kdump is initiated. If a controller
crash dump is in progress, the driver will wait for the controller crash
dump to complete before issuing the controller soft reset then complete
driver initialization.
Link: https://lore.kernel.org/r/20210928235442.201875-3-don.brace@microchip.com
Reviewed-by: Scott Benesh <scott.benesh@microchip.com>
Reviewed-by: Scott Teel <scott.teel@microchip.com>
Reviewed-by: Mike McGowen <mike.mcgowen@microchip.com>
Acked-by: John Donnelly <john.p.donnelly@oracle.com>
Signed-off-by: Mahesh Rajashekhara <mahesh.rajashekhara@microchip.com>
Signed-off-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This function is more complicated than necessary.
If we change from scnprintf() to snprintf() that lets us remove the if
bytes_wrote < sizeof(protocol) checks. Also, we can use bytes_wrote ? ","
: "" to print the comma and remove the separate if statement and the
"is_string_nonempty" variable.
[mkp: a few formatting cleanups and s/wrote/written/]
Link: https://lore.kernel.org/r/20210916132605.GF25094@kili
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Commit d39df15851 ("scsi: iscsi: Have abort handler get ref to conn")
added iscsi_get_conn()/iscsi_put_conn() calls during abort handling but
then also changed the handling of the case where we detect an already
completed task where we now end up doing a goto to the common put/cleanup
code. This results in a iscsi_task use after free, because the common
cleanup code will do a put on the iscsi_task.
This reverts the goto and moves the iscsi_get_conn() to after we've checked
if the iscsi_task is valid.
Link: https://lore.kernel.org/r/20211004210608.9962-1-michael.christie@oracle.com
Fixes: d39df15851 ("scsi: iscsi: Have abort handler get ref to conn")
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When an FC-GS I/O is aborted by lpfc, the driver requires a node pointer
for a dereference operation. In the abort I/O routine, the driver miscasts
a context pointer to the wrong data type and overwrites a single byte
outside of the allocated space. This miscast is done in the abort I/O
function handler because the handler works on both FC-GS and FC-LS
commands. However, the code neglected to get the correct job location for
the node.
Fix this by acquiring the necessary node pointer from the correct job
structure depending on the I/O type.
Link: https://lore.kernel.org/r/20211004231210.35524-1-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
IRQ polling thread calls ISR after enable_irq() to handle any missed I/O
completion. The atomic flag "in_used" was added to have the synchronization
between the IRQ polling thread and the interrupt context. There is a bug
around it leading to a race condition.
Below is the sequence:
- IRQ polling thread accesses ISR, fetches the reply descriptor.
- Real interrupt arrives and pre-empts polling thread (enable_irq() is
already called).
- Interrupt context picks the same reply descriptor as fetched by polling
thread, processes it, and exits.
- Polling thread resumes and processes the descriptor which is already
processed by interrupt thread leads to kernel crash.
Setting the "in_used" flag before fetching the reply descriptor ensures
synchronized access to ISR.
Link: https://www.spinics.net/lists/linux-scsi/msg159440.html
Link: https://lore.kernel.org/r/20210929124022.24605-2-sumit.saxena@broadcom.com
Fixes: 9bedd36e91 ("scsi: megaraid_sas: Handle missing interrupts while re-enabling IRQs")
Signed-off-by: Sumit Saxena <sumit.saxena@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Change the log level of the following message to debug:
Unsupported SCSI Opcode 0xXX, sending CHECK_CONDITION.
This message is mostly helpful during debugging sessions in order to
understand errors on the initiator side. But most of the time it's just
useless and makes reading logs much harder.
It gets particularly annoying if there are many initiators that come and go
or if an initiator runs a program that does not care whether the command is
supported and just keeps sending it.
Link: https://lore.kernel.org/r/20210929114959.705852-1-k.shelekhin@yadro.com
Reviewed-by: Roman Bolshakov <r.bolshakov@yadro.com>
Reviewed-by: Lee Duncan <lduncan@suse.com>
Signed-off-by: Konstantin Shelekhin <k.shelekhin@yadro.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Callers of ufshcd_reset_and_restore() expect it to return in an operational
state. However, the code only checks direct errors and so the ufshcd_state
may not be UFSHCD_STATE_OPERATIONAL due to error interrupts.
Fix by also checking ufshcd_state, still allowing non-fatal errors which
are left for the error handler to deal with.
Link: https://lore.kernel.org/r/20211002154550.128511-2-adrian.hunter@intel.com
Reviewed-by: Avri altman <avri.altman@wdc.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Commit 57d104c153 ("ufs: add UFS power management support") made the UFS
driver submit a REQUEST SENSE command before submitting a power management
command to a WLUN to clear the POWER ON unit attention. Instead of
submitting a REQUEST SENSE command before submitting a power management
command, retry the power management command until it succeeds.
This is the preparation to get rid of all UNIT ATTENTION code which should
be handled by users.
Link: https://lore.kernel.org/r/20211001182015.1347587-2-jaegeuk@kernel.org
Cc: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
