The controller may frequently enter and exit suspend for each I/O which we
need to deal with. This is inefficient and may cause too much suspend and
resume activity for the controller. To avoid this, use a default 5s
autosuspend for the controller to stop frequently suspending and
resuming. This value may still be modified via sysfs interfaces.
Link: https://lore.kernel.org/r/1639999298-244569-16-git-send-email-chenxiang66@hisilicon.com
Acked-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Processing events such as PORTE_BROADCAST_RCVD may cause dependency issues
for runtime power management support. Such a problem would be that
handling a PORTE_BROADCAST_RCVD event requires that the host is resumed to
send SMP commands. However, in resuming the host, the phyup events
generated from re-enabling the phys are processed in the same workqueue as
the original PORTE_BROADCAST_RCVD event. As such, the host will never
finish resuming (as it waits for the phyup event processing), and then the
PORTE_BROADCAST_RCVD event can't be processed as the SMP commands are
blocked, and so we have a deadlock. Solve this problem by ensuring that
libsas keeps the host active until completely finished phy or port events,
such as PORTE_BYTES_DMAED. As such, we don't have to worry about resuming
the host for processing individual SMP commands in this example.
Link: https://lore.kernel.org/r/1639999298-244569-15-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
It is possible that controller may become suspended between processing a
phyup interrupt and the event being processed by libsas. As such, we can't
ensure the controller is active when processing the phyup event - this may
cause the phyup event to be lost or other issues. To avoid any possible
issues, add pm_runtime_get_noresume() in phyup interrupt handler and
pm_runtime_put_sync() in the work handler exit to ensure that we stay
always active. Since we only want to call pm_runtime_get_noresume() for v3
hw, signal this will a new event, HISI_PHYE_PHY_UP_PM.
Link: https://lore.kernel.org/r/1639999298-244569-14-git-send-email-chenxiang66@hisilicon.com
Acked-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
During the processing of event PORT_BYTES_DMAED, the driver queues work
DISCE_DISCOVER_DOMAIN and then flushes workqueue ha->disco_q. If a new
phyup event occurs during resuming the controller, the work
PORTE_BYTES_DMAED of new phy occurs before suspended phy's. The work
DISCE_DISCOVER_DOMAIN of new phy requires an active SAS controller (it
needs to resume SAS controller by function scsi_sysfs_add_sdev() and some
other functions such as function add_device_link()). However, the
activation of the SAS controller requires completion of work
PORTE_BYTES_DMAED of suspended phys while it is blocked by new phy's work
on ha->event_q. So there is a deadlock and it is released only after resume
timeout.
To solve the issue, defer works of new phys during suspend and queue those
defer works after SAS controller becomes active.
Link: https://lore.kernel.org/r/1639999298-244569-13-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In the second part of function __sas_drain_work(), deferred work is queued.
This functionality is required other places so factor it out into the
function sas_queue_deferred_work().
Link: https://lore.kernel.org/r/1639999298-244569-12-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Add a flag SAS_HA_RESUMING and use it to indicate the state of resuming the
host controller.
Link: https://lore.kernel.org/r/1639999298-244569-11-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When sending SMP I/Os to the host we need to ensure that the host is not
suspended and can process the commands. This is a better approach than
replying on the host to resume itself to handle such commands. Use
pm_runtime_get_sync() and pm_runtime_put_sync() calls for the host when
executing SMP I/Os.
Link: https://lore.kernel.org/r/1639999298-244569-10-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If a new disk is inserted through an expander when the host was suspended,
it will not necessarily be detected as the topology is not re-scanned
during resume. To detect possible changes in topology during suspension,
insert a PORTE_BROADCAST_RCVD event per port when resuming to trigger a
revalidation.
Link: https://lore.kernel.org/r/1639999298-244569-8-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Most places that use asd_sas_port->phy_list are protected by spinlock
asd_sas_port->phy_list_lock, however there are still some places which miss
grabbing the lock. Add it in function hisi_sas_refresh_port_id() when
accessing asd_sas_port->phy_list. This carries a risk that list mutates
while at the same time dropping the lock in function
hisi_sas_send_ata_reset_each_phy(). Read asd_sas_port->phy_mask instead of
accessing asd_sas_port->phy_list to avoid this risk.
Link: https://lore.kernel.org/r/1639999298-244569-6-git-send-email-chenxiang66@hisilicon.com
Acked-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Most places that use asd_sas_port->phy_list in libsas are protected by
spinlock asd_sas_port->phy_list_lock. However, there are still a few places
which miss the lock. Add it in those places.
Link: https://lore.kernel.org/r/1639999298-244569-5-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
John Garry reported a deadlock that occurs when trying to access a
runtime-suspended SATA device. For obscure reasons, the rescan procedure
causes the link to be hard-reset, which disconnects the device.
The rescan tries to carry out a runtime resume when accessing the device.
scsi_rescan_device() holds the SCSI device lock and won't release it until
it can put commands onto the device's block queue. This can't happen until
the queue is successfully runtime-resumed or the device is unregistered.
But the runtime resume fails because the device is disconnected, and
__scsi_remove_device() can't do the unregistration because it can't get the
device lock.
The best way to resolve this deadlock appears to be to allow the block
queue to start running again even after an unsuccessful runtime resume.
The idea is that the driver or the SCSI error handler will need to be able
to use the queue to resolve the runtime resume failure.
This patch removes the err argument to blk_post_runtime_resume() and makes
the routine act as though the resume was successful always. This fixes the
deadlock.
Link: https://lore.kernel.org/r/1639999298-244569-4-git-send-email-chenxiang66@hisilicon.com
Fixes: e27829dc92 ("scsi: serialize ->rescan against ->remove")
Reported-and-tested-by: John Garry <john.garry@huawei.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This reverts commit b14a37e011.
In that commit, we had to filter out phy-up events during suspend, as it
work cause a deadlock between processing the phyup event and the resume HA
function try to drain the HA event workqueue to complete the resume
process.
Now that we no longer try to drain the HA event queue during the HA resume
processor, the deadlock would not occur, so remove the special handling for
it.
Link: https://lore.kernel.org/r/1639999298-244569-3-git-send-email-chenxiang66@hisilicon.com
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
For the hisi_sas driver, if a directly attached disk is removed during
suspend, a hang will occur in the resume process:
The background is that in commit 16fd4a7c59 ("scsi: hisi_sas: Add device
link between SCSI devices and hisi_hba"), it is ensured that the HBA device
cannot be runtime suspended when any SCSI device associated is active.
Other drivers which use libsas don't worry about this as none support
runtime suspend.
The mentioned hang occurs when an disk is removed during suspend. In the
removal process - from PHYE_RESUME_TIMEOUT event processing - we call into
scsi_remove_device(), which is being processed in the HA event workqueue.
Here we wait for all suppliers of the SCSI device to resume, which includes
the HBA device (from the above commit). However the HBA device cannot
resume, as it is waiting for the PHYE_RESUME_TIMEOUT to be processed (from
calling sas_resume_ha() -> sas_drain_work()). This is the deadlock.
There does not appear to be any need for the sas_drain_work() to be called
at all in sas_resume_ha() as it is not syncing against anything, so allow
LLDDs to avoid this by providing a variant of sas_resume_ha() which does
"sync", i.e. doesn't drain the event workqueue.
Link: https://lore.kernel.org/r/1639999298-244569-2-git-send-email-chenxiang66@hisilicon.com
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Value 0 is used for SAM status and libsas exec_status bytes codes in
sas_end_task() - use defined macros instead. In addition, change to proper
enum types.
Also replace SAM_STAT_CHECK_CONDITION with SAS_SAM_STAT_CHECK_CONDITION,
the former being a proper member of enum exec_status.
Link: https://lore.kernel.org/r/1639579061-179473-9-git-send-email-john.garry@huawei.com
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The OOB interrupt and phyup interrupt handlers may run out-of-order in high
CPU usage scenarios. Since the hisi_sas_phy.timer is added in
hisi_sas_phy_oob_ready() and disarmed in phy_up_v3_hw(), this out-of-order
execution will cause hisi_sas_phy.timer timeout to trigger.
To solve, protect hisi_sas_phy.timer and .attached with a lock, and ensure
that the timer won't be added after phyup handler completes.
Link: https://lore.kernel.org/r/1639579061-179473-8-git-send-email-john.garry@huawei.com
Signed-off-by: Qi Liu <liuqi115@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If we issue a controller reset command during executing a FLR a hung task
may be found:
Call trace:
__switch_to+0x158/0x1cc
__schedule+0x2e8/0x85c
schedule+0x7c/0x110
schedule_timeout+0x190/0x1cc
__down+0x7c/0xd4
down+0x5c/0x7c
hisi_sas_task_exec+0x510/0x680 [hisi_sas_main]
hisi_sas_queue_command+0x24/0x30 [hisi_sas_main]
smp_execute_task_sg+0xf4/0x23c [libsas]
sas_smp_phy_control+0x110/0x1e0 [libsas]
transport_sas_phy_reset+0xc8/0x190 [libsas]
phy_reset_work+0x2c/0x40 [libsas]
process_one_work+0x1dc/0x48c
worker_thread+0x15c/0x464
kthread+0x160/0x170
ret_from_fork+0x10/0x18
This is a race condition which occurs when the FLR completes first.
Here the host HISI_SAS_RESETTING_BIT flag out gets of sync as
HISI_SAS_RESETTING_BIT is not always cleared with the hisi_hba.sem held, so
now only set/unset HISI_SAS_RESETTING_BIT under hisi_hba.sem .
Link: https://lore.kernel.org/r/1639579061-179473-7-git-send-email-john.garry@huawei.com
Signed-off-by: Qi Liu <liuqi115@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
A user may issue a control phy command from sysfs at any time, even if the
controller is resetting.
If a phy is disabled by hardreset/linkreset command before calling
get_phys_state() in the reset path, the saved phy state may be incorrect.
To avoid incorrectly recording the phy state, use hisi_hba.sem to ensure
that the controller reset may not run at the same time as when the phy
control function is running.
Link: https://lore.kernel.org/r/1639579061-179473-6-git-send-email-john.garry@huawei.com
Signed-off-by: Qi Liu <liuqi115@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The task prep code is the same between the normal path (in
hisi_sas_task_prep()) and the internal abort path, so factor is out into a
common function.
Link: https://lore.kernel.org/r/1639579061-179473-5-git-send-email-john.garry@huawei.com
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
To help factor out code in future, it's useful to know if we're executing
an internal abort, so pass a pointer to the structure. The idea is that a
NULL pointer means not an internal abort.
Link: https://lore.kernel.org/r/1639579061-179473-4-git-send-email-john.garry@huawei.com
Reviewed-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>
For an internal abort, the task does not have a protocol, so set to none.
This will make it easier to differentiate internal abort tasks in future.
Link: https://lore.kernel.org/r/1639579061-179473-3-git-send-email-john.garry@huawei.com
Reviewed-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>
Currently we start delivery of commands to the DQ after returning from
hisi_sas_task_exec() with success.
Let's just start delivery directly in that function without having to check
if some local variable is set.
Link: https://lore.kernel.org/r/1639579061-179473-2-git-send-email-john.garry@huawei.com
Reviewed-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>
dma_alloc_coherent() ignores the zone specifiers so this is pointless and
confusing.
Link: https://lore.kernel.org/r/20211214163605.416288-1-hch@lst.de
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Call shost_for_each_device() with holding host->host_lock will cause a
deadlock situation, which will cause the system to stall (the log as
follow). Fix this issue by using __shost_for_each_device() in
ufshcd_pending_cmds().
stalls on CPUs/tasks:
all trace:
__switch_to+0x120/0x170
0xffff800011643998
ask dump for CPU 5:
ask:kworker/u16:2 state:R running task stack: 0 pid: 80 ppid: 2 flags:0x0000000a
orkqueue: events_unbound async_run_entry_fn
all trace:
__switch_to+0x120/0x170
0x0
ask dump for CPU 6:
ask:kworker/u16:6 state:R running task stack: 0 pid: 164 ppid: 2 flags:0x0000000a
orkqueue: events_unbound async_run_entry_fn
all trace:
__switch_to+0x120/0x170
0xffff54e7c4429f80
ask dump for CPU 7:
ask:kworker/u16:4 state:R running task stack: 0 pid: 153 ppid: 2 flags:0x0000000a
orkqueue: events_unbound async_run_entry_fn
all trace:
__switch_to+0x120/0x170
blk_mq_run_hw_queue+0x34/0x110
blk_mq_sched_insert_request+0xb0/0x120
blk_execute_rq_nowait+0x68/0x88
blk_execute_rq+0x4c/0xd8
__scsi_execute+0xec/0x1d0
scsi_vpd_inquiry+0x84/0xf0
scsi_get_vpd_buf+0x34/0xb8
scsi_attach_vpd+0x34/0x140
scsi_probe_and_add_lun+0xa6c/0xab8
__scsi_scan_target+0x438/0x4f8
scsi_scan_channel+0x6c/0xa8
scsi_scan_host_selected+0xf0/0x150
do_scsi_scan_host+0x88/0x90
scsi_scan_host+0x1b4/0x1d0
ufshcd_async_scan+0x248/0x310
async_run_entry_fn+0x30/0x178
process_one_work+0x1e8/0x368
worker_thread+0x40/0x478
kthread+0x174/0x180
ret_from_fork+0x10/0x20
Link: https://lore.kernel.org/r/20211214120537.531628-1-huobean@gmail.com
Fixes: 8d077ede48 ("scsi: ufs: Optimize the command queueing code")
Reported-by: YongQin Liu <yongqin.liu@linaro.org>
Reported-by: Amit Pundir <amit.pundir@linaro.org>
Tested-by: John Stultz <john.stultz@linaro.org>
Tested-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Co-developed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently, the dev_loss_tmo setting is only ever used for SCSI
devices. This patch reshuffles initialisation such that the SCSI remote
ports are registered before the NVMe ones, allowing the dev_loss_tmo
setting to be synchronized between SCSI and NVMe.
Link: https://lore.kernel.org/r/20211214111139.52503-1-dwagner@suse.de
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Pull in the 5.16 fixes branch to resolve a conflict in the UFS driver
core.
Conflicts:
drivers/scsi/ufs/ufshcd.c
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When building under -Warray-bounds, a warning is generated when casting a
u32 into MAILBOX_t (which is larger). This warning is conservative, but
it's not an unreasonable change to make to improve future robustness. Use a
tagged struct_group that can refer to either the specific fields or the
first u32 separately, silencing this warning:
drivers/scsi/lpfc/lpfc_sli.c: In function 'lpfc_reset_barrier':
drivers/scsi/lpfc/lpfc_sli.c:4787:29: error: array subscript 'MAILBOX_t[0]' is partly outside array bounds of 'volatile uint32_t[1]' {aka 'volatile unsigned int[1]'} [-Werror=array-bounds]
4787 | ((MAILBOX_t *)&mbox)->mbxCommand = MBX_KILL_BOARD;
| ^~
drivers/scsi/lpfc/lpfc_sli.c:4752:27: note: while referencing 'mbox'
4752 | volatile uint32_t mbox;
| ^~~~
There is no change to the resulting executable instruction code.
Link: https://lore.kernel.org/r/20211203223351.107323-1-keescook@chromium.org
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memset(), avoid intentionally writing across
neighboring fields.
Add struct_group() to mark "stat" region of struct lpfc_cgn_info that
should be initialized to zero, and refactor the "data" region memset()
to wipe everything up to the cgn_stats region.
Link: https://lore.kernel.org/r/20211208195957.1603092-1-keescook@chromium.org
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Commit 598a90f200 ("scsi: qla2xxx: add ring buffer for tracing debug
logs") introduced unconditional log string formatting to ql_dbg() even if
ql_dbg_log event is disabled. It harms performance because some strings are
formatted in fastpath and/or interrupt context.
Link: https://lore.kernel.org/r/20211112145446.51210-1-r.bolshakov@yadro.com
Fixes: 598a90f200 ("scsi: qla2xxx: add ring buffer for tracing debug logs")
Cc: Rajan Shanmugavelu <rajan.shanmugavelu@oracle.com>
Cc: stable@vger.kernel.org
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Update lpfc version to 14.0.0.4.
Link: https://lore.kernel.org/r/20211204002644.116455-10-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>
Dump raw CMF parameter information in debugfs cgn_buffer.
Link: https://lore.kernel.org/r/20211204002644.116455-9-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>
Ensure read bytes data does not go over MBPI for CMF timer intervals that
are purposely shortened.
Link: https://lore.kernel.org/r/20211204002644.116455-8-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>
Calculate any extra bytes needed to account for timer accuracy. If we are
less than LPFC_CMF_INTERVAL, then calculate the adjustment needed for total
to reflect a full LPFC_CMF_INTERVAL.
Add additional info to rxmonitor, and adjust some log formatting.
Link: https://lore.kernel.org/r/20211204002644.116455-7-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>
Extraneous teardown routines are present in the firmware dump path causing
altered states in firmware captures.
When a firmware dump is requested via sysfs, trigger the dump immediately
without tearing down structures and changing adapter state.
The driver shall rely on pre-existing firmware error state clean up
handlers to restore the adapter.
Link: https://lore.kernel.org/r/20211204002644.116455-6-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>
The driver is calling schedule_timeout after the DA_ID nameserver request
and LOGO commands are issued to the fabric by the initiator virtual
endport. These fixed delay functions are causing long delays in the
driver's worker thread when processing discovery I/Os in a serialized
fashion, which is then triggering mailbox timeout errors artificially.
To fix this, don't wait on the DA_ID request to complete and call
wait_event_timeout to allow the vport delete thread to make progress on an
event driven basis rather than fixing the wait time.
Link: https://lore.kernel.org/r/20211204002644.116455-5-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>
Issuing lpfc_force_rscn twice results in an ndlp kref use-after-free call
trace.
A prior patch reworked the get/put handling by ensuring nlp_get was done
before WQE submission and a put was done in the completion path.
Unfortunately, the issue_els_rscn path had a piece of legacy code that did
a nlp_put, causing an imbalance on the ref counts.
Fixed by removing the unnecessary legacy code snippet.
Link: https://lore.kernel.org/r/20211204002644.116455-4-jsmart2021@gmail.com
Fixes: 4430f7fd09 ("scsi: lpfc: Rework locations of ndlp reference taking")
Cc: <stable@vger.kernel.org> # v5.11+
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>
During heavy I/O testing with issue_lip to bounce the link, occasionally
I/O is terminated with status 3 result 9, which means the RPI is suspended.
The I/O is completed and this type of error will result in immediate retry
by the SCSI layer. The retry count expires and the I/O fails and returns
error to the application.
To avoid these quick retry/retries exhausted scenarios change the return
code given to the midlayer to DID_REQUEUE rather than DID_ERROR. This gets
them retried, and eventually succeed when the link recovers.
Link: https://lore.kernel.org/r/20211204002644.116455-3-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>
During rmmod testing, messages appeared indicating lpfc_mbuf_pool entries
were still busy. This situation was only seen doing rmmod after at least 1
vport (NPIV) instance was created and destroyed. The number of messages
scaled with the number of vports created.
When a vport is created, it can receive a PLOGI from another initiator
Nport. When this happens, the driver prepares to ack the PLOGI and
prepares an RPI for registration (via mbx cmd) which includes an mbuf
allocation. During the unsolicited PLOGI processing and after the RPI
preparation, the driver recognizes it is one of the vport instances and
decides to reject the PLOGI. During the LS_RJT preparation for the PLOGI,
the mailbox struct allocated for RPI registration is freed, but the mbuf
that was also allocated is not released.
Fix by freeing the mbuf with the mailbox struct in the LS_RJT path.
As part of the code review to figure the issue out a couple of other areas
where found that also would not have released the mbuf. Those are cleaned
up as well.
Link: https://lore.kernel.org/r/20211204002644.116455-2-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>
The time spent in io_schedule() and also the interrupt latency are
significant when submitting direct I/O to a UFS device. Hence this patch
that implements polling support. User space software can enable polling by
passing the RWF_HIPRI flag to the preadv2() system call or the
IORING_SETUP_IOPOLL flag to the io_uring interface.
Although the block layer supports to partition the tag space for
interrupt-based completions (HCTX_TYPE_DEFAULT) purposes and polling
(HCTX_TYPE_POLL), the choice has been made to use the same hardware queue
for both hctx types because partitioning the tag space would negatively
affect performance.
On my test setup this patch increases IOPS from 2736 to 22000 (8x) for the
following test:
for hipri in 0 1; do
fio --ioengine=io_uring --iodepth=1 --rw=randread \
--runtime=60 --time_based=1 --direct=1 --name=qd1 \
--filename=/dev/block/sda --ioscheduler=none --gtod_reduce=1 \
--norandommap --hipri=$hipri
done
Link: https://lore.kernel.org/r/20211203231950.193369-18-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.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>
Remove the clock scaling lock from ufshcd_queuecommand() since it is a
performance bottleneck. Instead check the SCSI device budget bitmaps in the
code that waits for ongoing ufshcd_queuecommand() calls. A bit is set in
sdev->budget_map just before scsi_queue_rq() is called and a bit is cleared
from that bitmap if scsi_queue_rq() does not submit the request or after
the request has finished. See also the blk_mq_{get,put}_dispatch_budget()
calls in the block layer.
There is no risk for a livelock since the block layer delays queue reruns
if queueing a request fails because the SCSI host has been blocked.
Link: https://lore.kernel.org/r/20211203231950.193369-17-bvanassche@acm.org
Cc: Asutosh Das (asd) <asutoshd@codeaurora.org>
Reviewed-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Instead of locking and unlocking the clock scaling lock, surround the
command queueing code with an RCU reader lock and call synchronize_rcu().
This patch prepares for removal of the clock scaling lock.
Link: https://lore.kernel.org/r/20211203231950.193369-16-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.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>
Fix the following kernel crash:
Unable to handle kernel paging request at virtual address ffffffc91e735000
Call trace:
__queue_work+0x26c/0x624
queue_work_on+0x6c/0xf0
ufshcd_hold+0x12c/0x210
__ufshcd_wl_suspend+0xc0/0x400
ufshcd_wl_shutdown+0xb8/0xcc
device_shutdown+0x184/0x224
kernel_restart+0x4c/0x124
__arm64_sys_reboot+0x194/0x264
el0_svc_common+0xc8/0x1d4
do_el0_svc+0x30/0x8c
el0_svc+0x20/0x30
el0_sync_handler+0x84/0xe4
el0_sync+0x1bc/0x1c0
Fix this crash by ungating the clock before destroying the work queue on
which clock gating work is queued.
Link: https://lore.kernel.org/r/20211203231950.193369-15-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.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>
Release resources when aborting a command. Make sure that aborted commands
are completed once by clearing the corresponding tag bit from
hba->outstanding_reqs. This patch is an improved version of commit
3ff1f6b6ba ("scsi: ufs: core: Improve SCSI abort handling").
Link: https://lore.kernel.org/r/20211203231950.193369-14-bvanassche@acm.org
Fixes: 7a3e97b0dc ("[SCSI] ufshcd: UFS Host controller driver")
Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.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 only functional change in this patch is that scsi_done() is now called
after ufshcd_release() and ufshcd_clk_scaling_update_busy() instead of
before.
The next patch in this series will introduce a call to
ufshcd_release_scsi_cmd() in the abort handler.
Link: https://lore.kernel.org/r/20211203231950.193369-13-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.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>
This patch does not change any functionality but makes the next patch in
this series easier to read.
Link: https://lore.kernel.org/r/20211203231950.193369-12-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.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 previous patch removed all code that uses hba->cmd_queue. Hence also
remove hba->cmd_queue itself.
Link: https://lore.kernel.org/r/20211203231950.193369-11-bvanassche@acm.org
Suggested-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.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 deadlock has been observed on a test setup:
- All tags allocated
- The SCSI error handler calls ufshcd_eh_host_reset_handler()
- ufshcd_eh_host_reset_handler() queues work that calls
ufshcd_err_handler()
- ufshcd_err_handler() locks up as follows:
Workqueue: ufs_eh_wq_0 ufshcd_err_handler.cfi_jt
Call trace:
__switch_to+0x298/0x5d8
__schedule+0x6cc/0xa94
schedule+0x12c/0x298
blk_mq_get_tag+0x210/0x480
__blk_mq_alloc_request+0x1c8/0x284
blk_get_request+0x74/0x134
ufshcd_exec_dev_cmd+0x68/0x640
ufshcd_verify_dev_init+0x68/0x35c
ufshcd_probe_hba+0x12c/0x1cb8
ufshcd_host_reset_and_restore+0x88/0x254
ufshcd_reset_and_restore+0xd0/0x354
ufshcd_err_handler+0x408/0xc58
process_one_work+0x24c/0x66c
worker_thread+0x3e8/0xa4c
kthread+0x150/0x1b4
ret_from_fork+0x10/0x30
Fix this lockup by making ufshcd_exec_dev_cmd() allocate a reserved
request.
Link: https://lore.kernel.org/r/20211203231950.193369-10-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.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>
Prepare for making sdev->host->can_queue less than hba->nutrs. This patch
does not change any functionality.
Link: https://lore.kernel.org/r/20211203231950.193369-9-bvanassche@acm.org
Tested-by: Bean Huo <beanhuo@micron.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>
