scsi: lpfc: Fix nvme target mode ABTSing a received ABTS

If an unsolicited ABTS was received, the driver looks up the exchange it
references. It it does various searches looking for the exchange
context. When one is eventually matched and it is associated with an XRI
context, the driver sends an ABORT WQE to terminate the exchange. Current
code looks at whether the transport had taken action on the XRI yet or not
(no action if set to LPFC_NVMET_STE_RCV; action if non-LPFC_NVMET_STE_RCV).
Based on action or not one of two (sol vs unsol) issue abort routines are
called. The unsol version cheats and transmits a sequence containing an
ABTS with no interaction with the adapter.  The sol version issues an Abort
WQE and lets the adapter manage whether the ABTS is sent to not.

The issue is the unsol version is sending ABTS unconditionally for the
exchange that received the ABTS. It's unnecessary.

Remove the conditional and just call the adapter command-based routine to
let the adapter manage the ABTS.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix hang when downloading fw on port enabled for nvme

As part of firmware download, the adapter is reset. On the adapter the
reset causes the function to stop and all outstanding io is terminated
(without responses). The reset path then starts teardown of the adapter,
starting with deregistration of the remote ports with the nvme-fc
transport. The local port is then deregistered and the driver waits for
local port deregistration. This never finishes.

The remote port deregistrations terminated the nvme controllers, causing
them to send aborts for all the outstanding io. The aborts were serviced in
the driver, but stalled due to its state. The nvme layer then stops to
reclaim it's outstanding io before continuing.  The io must be returned
before the reset on the controller is deemed complete and the controller
delete performed.  The remote port deregistration won't complete until all
the controllers are terminated. And the local port deregistration won't
complete until all controllers and remote ports are terminated. Thus things
hang.

The issue is the reset which stopped the adapter also stopped all the
responses that would drive i/o completions, and the aborts were also
stopped that stopped i/o completions. The driver, when resetting the
adapter like this, needs to be generating the completions as part of the
adapter reset so that I/O complete (in error), and any aborts are not
queued.

Fix by adding flush routines whenever the adapter port has been reset or
discovered in error. The flush routines will generate the completions for
the scsi and nvme outstanding io. The abort ios, if waiting, will be caught
and flushed as well.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix too many sg segments spamming in kernel log

This issue is specific to SLI-3 adapters, specifically when DIF is used.

Once seen, this message floods the logs:
9064 BLKGRD: lpfc_scsi_prep_dma_buf_s3: Too many sg segments from
  dma_map_sg

The driver, upon detecting an error such as too many elements in an sglist,
misrepresents the error by treating it as a temporary resource issue by
returning MLQUEUE_HOST_BUSY.  In these cases, no retry will fix it and it
should have been a hard error. The repeated retry was causing the spamming
of the log.

As for the initial reason of why an I/O encountered this issue at all is
not clear as parameters set by the driver should have avoided this.  The
dm multipath maintainer has been notified of the issue.

Fix by changing the return code for the dma mapping routines to indicate
cases that are not retryable and return DID_ERROR on those cases.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix crash due to port reset racing vs adapter error handling

If the adapter encounters a condition which causes the adapter to fail
(driver must detect the failure) simultaneously to a request to the driver
to reset the adapter (such as a host_reset), the reset path will be racing
with the asynchronously-detect adapter failure path.  In the failing
situation, one path has started to tear down the adapter data structures
(io_wq's) while the other path has initiated a repeat of the teardown and
is in the lpfc_sli_flush_xxx_rings path and attempting to access the
just-freed data structures.

Fix by the following:

 - In cases where an adapter failure is detected, rather than explicitly
   calling offline_eratt() to start the teardown, change the adapter state
   and let the later calls of posted work to the slowpath thread invoke the
   adapter recovery.  In essence, this means all requests to reset are
   serialized on the slowpath thread.

 - Clean up the routine that restarts the adapter. If there is a failure
   from brdreset, don't immediately error and leave things in a partial
   state. Instead, ensure the adapter state is set and finish the teardown
   of structures before returning.

 - If in the scsi host reset handler and the board fails to reset and
   restart (which can be due to parallel reset/recovery paths), instead of
   hard failing and explicitly calling offline_eratt() (which gets into the
   redundant path), just fail out and let the asynchronous path resolve the
   adapter state.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix deadlock on host_lock during cable pulls

During cable pull testing a deadlock was seen between lpfc_nlp_counters()
vs lpfc_mbox_process_link_up() vs lpfc_work_list_done(). They are all
waiting on the shost->host_lock.

Issue is all of these cases raise irq when taking out the lock but use
spin_unlock_irq() when unlocking. The unlock path is will unconditionally
re-enable interrupts in cases where irq state should be preserved. The
re-enablement allowed the other paths to execute which then causes the
deadlock.

Fix by converting the lock/unlock to irqsave/irqrestore.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix error in remote port address change

In a test with high nvme remote port counts connected via a multi-hop FC
switch config where switches were systematically reset (e.g. fabric
partitioning and re-establishment), the nvme remote ports would switch
addresses based on the switch reconfiguration events. The driver would get
into a situation where the nvme port changed address, PLOGI and PRLI would
succeed nvme transport registration occurred, but subsequent LS requests by
the nvme subsystem failed due to a bad ndlp state and connectivity to the
device failed.

The driver hit a race condition on multiple devices that address swapped
simultaneously. In cases where the driver notices the remote port structure
came back as the same value as previously (meaning a nvme_rport structure
was re-enabled and did not go through devloss_tmo/connect_tmo_failures on
all controllers) the driver would unconditionally exit assuming the ndlp
information was correct. But, if the ndlp's had been swapped, the ndlp had
stale port state information, which when used by the LS request commands,
would fail the commands.

Fix by checking whether a node swap had occurred, and only exit if no ndlp
swap had occurred.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix driver nvme rescan logging

In situations where zoning is not being used, thus NVMe initiators see
other NVMe initiators as well as NVMe targets, a link bounce on an
initiator will cause the NVMe initiators to spew "6169" State Error
messages.

The driver is not qualifying whether the remote port is a NVMe targer or
not before calling the lpfc_nvme_rescan_port(), which validates the role
and prints the message if its only an NVMe initiator.

Fix by the following:

 - Before calling lpfc_nvme_rescan_port() ensure that the node is a NVMe
   storage target or a NVMe discovery controller.

 - Clean up implementation of lpfc_nvme_rescan_port. remoteport pointer
   will always be NULL if a NVMe initiator only. But, grabbing of
   remoteport pointer should be done under lock to coincide with the
   registering of the remote port with the fc transport.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix sg_seg_cnt for HBAs that don't support NVME

On an SLI-3 adapter which does not support NVMe, but with the driver global
attribute to enable nvme on any adapter if it does support NVMe
(e.g. module parameter lpfc_enable_fc4_type=3), the SGL and total SGE
values are being munged by the protocol enablement when it shouldn't be.

Correct by changing the location of where the NVME sgl information is being
applied, which will avoid any SLI-3-based adapter.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix propagation of devloss_tmo setting to nvme transport

If admin changes the devloss_tmo on an rport via the fc_remote_port rport
dev_loss_tmo attribute, the value is on set on scsi stack.  The change is
not propagated to NVMe.

The set routine in the lldd lacks the call to
nvme_fc_set_remoteport_devloss() to set the value.

Fix by adding the call to the lldd set routine.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix loss of remote port after devloss due to lack of RPIs

In tests with remote ports contantly logging out/logging coupled with
occassional local link bounce, if a remote port is disocnnected for longer
than devloss_tmo and then subsequently reconnected, eventually the test
will fail to login with the remote port and remote port connectivity is
lost.

When devloss_tmo expires, the driver does not free the node struct until
the port or npiv instances is being deleted. The node is left allocated but
the state set to UNUSED. If the node was in the process of logging in when
the local link drop occurred, meaning the RPI was allocated for the node in
order to send the ELS, but not yet registered which comes after successful
login, the node is moved to the NPR state, and if devloss expires, to
UNUSED state.  If the remote port comes back, the node associated with it
is restarted and this path happens to allocate a new RPI and overwrites the
prior RPI value. In the cases where the port was logged in and loggs out,
the path did release the RPI but did not set the node rpi value.  In the
cases where the remote port never finished logging in, the path never did
the call to release the rpi. In this latter case, when the node is
subsequently restore, the new rpi allocation overwrites the rpi that was
not released, and the rpi is now leaked.  Eventually the port will run out
of RPI resources to log into new remote ports.

Fix by following changes:

 - When an rpi is released, do so under locks and ensure the node rpi value
   is set to a non-allocated value (LPFC_RPI_ALLOC_ERROR).  Note:
   refactored to a small service routine to avoid indentation issues.

 - When re-enabling a node, check the rpi value to determine if a new
   allocation is necessary. If already set, use the prior rpi.

Enhanced logging to help in the future.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix devices that don't return after devloss followed by rediscovery

If a remote port is removed and remains removed for devloss_tmo, if an RSCN
is subsequently received indicating the presence of the remte port, the
driver does not login to and rediscovery the remote port.

Currently, in order to for a port to be rediscovered post an RSCN, the node
state must be NPR to reflect not logged in. When devloss expires, the node
state is marked UNUSED. When an RSCN occurs, the nodes referenced by the
RSCN will have a NPR_2B_DISC flag set, but the re-login will only be
attempted if the node is in NPR_NODE state.  Thus the node is skipped over.

Fix by recognizing the NPR_2B_DISC and UNUSED and transition the node back
to NPR state to allow the re-login to take place.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix null ptr oops updating lpfc_devloss_tmo via sysfs attribute

If an admin updates lpfc's devloss_tmo sysfs attribute, the kernel will
oops.

Coding of a loop allowed a new value (rport) to be set/checked for null
followed by an older value (remoteport) checked for null to allow progress
where the new value, even though null, will be referenced.

Rework the logic to validate and prevent any reference to the null ptr.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix FLOGI handling across multiple link up/down conditions

It's possible for the driver to initiate an FLOGI and before it completes,
another link down/up transition occurs requiring a new FLOGI. Currently,
nothing is done to abort/noop the older FLOGI request to the adapter, so if
this transition occurs and the FLOGI completion is received after the link
down/up transition, the driver may erroneously act on the older FLOGI. In
most cases, the adapter properly terminates/fails the FLOGI, but there is a
timing condition where the FLOGI may complete on the wire prior to the
transition, but the response may not be seen/processed by the driver before
the driver sees the link transition.

Fix by having the link down handler in the driver run through any
outstanding ELS's and change the completion handler of the ELS so that it
will be no-op'd and released.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix oops when fewer hdwqs than cpus

When tearing down the adapter for a reset, online/offline, or driver
unload, the queue free routine would hit a GPF oops.  This only occurs on
conditions where the number of hardware queues created is fewer than the
number of cpus in the system. In this condition cpus share a hardware
queue. And of course, it's the 2nd cpu that shares a hardware that
attempted to free it a second time and hit the oops.

Fix by reworking the cpu to hardware queue mapping such that:
Assignment of hardware queues to cpus occur in two passes:
first pass: is first time assignment of a hardware queue to a cpu.
  This will set the LPFC_CPU_FIRST_IRQ flag for the cpu.
second pass: for cpus that did not get a hardware queue they will
  be assigned one from a primary cpu (one set in first pass).

Deletion of hardware queues is driven by cpu itteration, and queues
will only be deleted if the LPFC_CPU_FIRST_IRQ flag is set.

Also contains a few small cleanup fixes and a little better logging.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix irq raising in lpfc_sli_hba_down

The adapter reset path (lpfc_sli_hba_down) is taking/releasing a lock with
irq. But, the path is already under the hbalock which raised irq so it's
unnecessary.

Convert to simple lock/unlock.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix Oops in nvme_register with target logout/login

lpfc_nvme_register_port hit a null prev_ndlp pointer in a test with lots of
target ports swapping addresses. The oldport value was stale, thus it's
ndlp (prev_ndlp set to it) was used.

Fix by moving oldrport pointer checks, and if used prev_ndlp pointer
assignment, to be done while the lock is held.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix issuing init_vpi mbox on SLI-3 card

The driver is inadvertently trying to issue an INIT_VPI mailbox command on
an SLI-3 driver. The command is specific to SLI-4. When the call is made to
send the command, if on an SLI-3 adapter, an array pointer is NULL and the
driver will oops.

Fix by restricting the command to SLI-4 adapters only.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix ADISC reception terminating login state if a NVME target

If a target issues an ADISC to the port and the target is a NVME target,
the driver is inadvertantly invalidating the login and marking the remote
port as logged out. Communication with the target is lost.

Revise the ADISC check so that FCP or NVME targets will be marked valid at
the end of ADISC processing.  Enhance logging to recognize condition
better.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix discovery when target has no GID_FT information

Some remote ports may be slow in registering their GID_FT protocol
information with the fabric. If the remote port is an initiator, it may
send PLOGI to the port before the GID_FT logic is complete. Meaning, after
accepting the PLOGI, when the driver may see no response to the GID_FT that
is issued after the login to determine the protocols supported so that
proper PRLI's may be transmit. If the driver has no fc4 information, it
currently stops and the remote port is not discovered.

Fix by issuing a LOGO when there is no GID_FT information.  The LOGO
completion handling will attempt to re-login if the nport_id is still
present.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix port relogin failure due to GID_FT interaction

In cases of remote-port-side cable pull/replug, there happens to be a
target that upon replug will send the port a PLOGI, a PRLI, and a LOGO.
When this sequence is received by the driver, the PLOGI accepted and a
GFT_ID is issued to find the protocol support for the remote port. While
the GFT_ID is outstanding, a LOGO is received. The driver logs the remote
port out and unregisters the RPI and schedules a new PLOGI transmission.
However, the GFT_ID was not terminated. When it completed, the driver
attempted to transition the remote port to PRLI transmission, which cancels
the PLOGI scheduling. The PRLI transmit attempt is rejected by the adapter
as the remote port is not logged in. No retry is attempted as it's expected
the logout is noted and the supposedly scheduled PLOGI should address the
state. As there is no PLOGI, the remote port does not get re-discovered.

Fix by aborting the outstanding GFT_ID if the related remote port is logged
out.

Ensure a PRLI transmit attempt only occurs if the remote port is logging
in. This avoids the incorrect attempt while logged out.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix leak of ELS completions on adapter reset

If the adapter is reset while there are outstanding ELS's, subsequent
reinitialization of the adapter will fail as it has not recovered all of
the io contexts relative to the ELS's.

If an ELS timed out or otherwise failed and an the ELS was attempted to be
aborted (which changes the ELS completion context), in causes where the
driver generates completions for the outstanding IO as the adapter would
not due to being reset, the driver released only the ELS context and failed
to release the abort context.  When the adapter went to reinit, as it had
not received all of the contexts, it failed to reinit.

Fix by having the ELS completion handler identify the driver-generated
completion status and release the abort context.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix failure to clear non-zero eq_delay after io rate reduction

Unusually high IO latency can be observed with little IO in progress. The
latency may remain high regardless of amount of IO and can only be cleared
by forcing lpfc_fcp_imax values to non-zero and then back to zero.

The driver's eq_delay mechanism that scales the interrupt coalescing based
on io completion load failed to reduce or turn off coalescing when load
decreased. Specifically, if no io completed on a cpu within an eq_delay
polling window, the eq delay processing was skipped and no change was made
to the coalescing values. This left the coalescing values set when they
were no longer applicable.

Fix by always clearing the percpu counters for each time period and always
run the eq_delay calculations if an eq has a non-zero coalescing value.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix crash on driver unload in wq free

If a timer routine uses workqueues, it could fire before the workqueue is
allocated.

Fix by allocating the workqueue before the timer routines are setup

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix ELS field alignments

After seeing some interoperability issues with ADISC, it was determined the
ELS definitions in lpfc were using types that allowed the compiler to add
pad to the structure, causing the structure to no longer be per spec. The
offending structures are ADISC, FAN, and RNID.

This patch implements the simple fix of eliminating the pad by forcing the
compiler to pack the structure. Care was taken to ensure field accesses
won't be by operations that would hit a bad field alignment.

The better solution would be to convert to the uapi fc header definitions,
but the number of changes required to do is rather intrusive so this course
of action was deferred.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix PLOGI failure with high remoteport count

When connected to a high number of remote ports, the driver is encountering
PLOGI errors.  The errors are due to adapter detected failures indicating
illegal field values.

Turns out the driver was prematurely clearing an RPI bitmask before waiting
for an UNREG_RPI mailbox completion. This allowed the RPI to be reused
before it was actually available.

Fix by clearing RPI bitmask only after UNREG_RPI mailbox completion.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Limit xri count for kdump environment

scsi-mq operation inherently performs pre-allocation of resources for
blk-mq request queues. Even though the kdump environment reduces the
configuration to a single CPU, thus 1 hardware queue, which helps
significantly, the resources are still rather large due to the per request
allocations. blk-mq pre-allocations can be over 4KB per request.  With
adapter can_queue values in the 4k or 8k range, this can easily be 32MBs
before any other driver memory is factored in.  Driver SGL DMA buffer
allocation can be up to 8KB per request as well adding an additional
64MB. Totals are well over 100MB for a single shost.  Given kdump memory
auto-sizing utilities don't accommodate this amount of memory well, it's
very possible for kdump to fail due to lack of memory.

Fix by having the driver recognize that it is booting within a kdump
context and reduce the number of requests it will support to a more
reasonable value.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: remove NULL check before some freeing functions

As dma_pool_destroy and mempool_destroy functions has NULL check. We may
not need NULL check before calling them.

Fix below warnings reported by coccicheck
./drivers/scsi/lpfc/lpfc_mem.c:252:2-18: WARNING: NULL check before some
freeing functions is not needed.
./drivers/scsi/lpfc/lpfc_mem.c:255:2-18: WARNING: NULL check before some
freeing functions is not needed.
./drivers/scsi/lpfc/lpfc_mem.c:258:2-18: WARNING: NULL check before some
freeing functions is not needed.
./drivers/scsi/lpfc/lpfc_mem.c:261:2-18: WARNING: NULL check before some
freeing functions is not needed.
./drivers/scsi/lpfc/lpfc_mem.c:265:2-18: WARNING: NULL check before some
freeing functions is not needed.
./drivers/scsi/lpfc/lpfc_mem.c:269:2-17: WARNING: NULL check before some
freeing functions is not needed.

Signed-off-by: Hariprasad Kelam <hariprasad.kelam@gmail.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs-qcom: Make structure ufs_hba_qcom_vops constant

Static structure ufs_hba_qcom_vops, of type ufs_hba_variant_ops, is used
only once, when it is passed as the second argument to function
ufshcd_pltfrm_init(). In the definition of ufshcd_pltfrm_init(), its second
parameter (corresponding to ufs_hba_qcom_vops) is declared as
constant. Hence declare ufs_hba_qcom_vops itself constant as well to
protect it from unintended modification.  Issue found with Coccinelle.

Signed-off-by: Nishka Dasgupta <nishkadg.linux@gmail.com>
Reviewed-by: Vivek Gautam <vivek.gautam@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: use spin_lock_irqsave in IRQ context

As spin_unlock_irq will enable interrupts.
Function lpfc_findnode_rpi is called from
    lpfc_sli_abts_err_handler (./drivers/scsi/lpfc/lpfc_sli.c)
 <- lpfc_sli_async_event_handler
 <- lpfc_sli_process_unsol_iocb
 <- lpfc_sli_handle_fast_ring_event
 <- lpfc_sli_fp_intr_handler
 <- lpfc_sli_intr_handler
 and lpfc_sli_intr_handler is an interrupt handler.

Interrupts are enabled in interrupt handler.  Use
spin_lock_irqsave/spin_unlock_irqrestore instead of spin_(un)lock_irq in
IRQ context to avoid this.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: remove redundant code

Remove the redundant initialization code.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: remove redundant assignment of variable rc

Variable ret is initialized to a value that is never read and it is
re-assigned later and immediately returns. Clean up the code by removing
rc and just returning 0.

[mkp: typo]

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: cleanup trace buffer initialization

Avoid code duplication between qla2x00_alloc_offload_mem() and
qla2x00_alloc_fw_dump() by moving the FCE and EFT buffer allocation and
initialization to separate functions. Cleanly track failure and success by
making sure that the ha->eft, ha->fce and respective eft_dma, fce_dma
members are set if and only if the buffers are properly allocated and
initialized. Avoid pointless buffer reallocation.  Eliminate some goto
statements. Make sure the fce_enabled flag is cleared when the FCE buffer
is freed.

Fixes: ad0a0b01f088 ("scsi: qla2xxx: Fix Firmware dump size for Extended login and Exchange Offload")
Fixes: a28d9e4ef997 ("scsi: qla2xxx: Add support for multiple fwdump templates/segments")
Cc: Joe Carnuccio <joe.carnuccio@cavium.com>
Cc: Quinn Tran <qutran@marvell.com>
Cc: Himanshu Madhani <hmadhani@marvell.com>
Cc: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin Wilck <mwilck@suse.com>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: qla2x00_alloc_fw_dump: set ha->eft

In qla2x00_alloc_fw_dump(), an existing EFT buffer (e.g. from previous
invocation of qla2x00_alloc_offload_mem()) is freed.  The buffer is then
re-allocated, but without setting the eft and eft_dma fields to the new
values.

Fixes: a28d9e4ef997 ("scsi: qla2xxx: Add support for multiple fwdump templates/segments")
Cc: Joe Carnuccio <joe.carnuccio@cavium.com>
Cc: Quinn Tran <qutran@marvell.com>
Cc: Himanshu Madhani <hmadhani@marvell.com>
Cc: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin Wilck <mwilck@suse.com>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: use __u{8,16,32,64} instead of uint{8,16,32,64}_t in uapi headers

When CONFIG_UAPI_HEADER_TEST=y, exported headers are compile-tested to make
sure they can be included from user-space.

Currently, scsi_bsg_fc.h, scsi_netlink.h, and scsi_netlink_fc.h are
excluded from the test coverage. To make them join the compile-test, we
need to fix the build errors attached below.

For a case like this, we decided to use __u{8,16,32,64} variable types in
this discussion:

  https://lkml.org/lkml/2019/6/5/18

Build log:

  CC      usr/include/scsi/scsi_netlink_fc.h.s
  CC      usr/include/scsi/scsi_netlink.h.s
  CC      usr/include/scsi/scsi_bsg_fc.h.s
In file included from ./usr/include/scsi/scsi_netlink_fc.h:10:0,
                 from <command-line>:32:
./usr/include/scsi/scsi_netlink.h:29:2: error: unknown type name  uint8_t
  uint8_t version;
  ^~~~~~~
./usr/include/scsi/scsi_netlink.h:30:2: error: unknown type name  uint8_t
  uint8_t transport;
  ^~~~~~~
./usr/include/scsi/scsi_netlink.h:31:2: error: unknown type name  uint16_t
  uint16_t magic;
  ^~~~~~~~
./usr/include/scsi/scsi_netlink.h:32:2: error: unknown type name  uint16_t
  uint16_t msgtype;
  ^~~~~~~~
  CC      usr/include/rdma/vmw_pvrdma-abi.h.s
./usr/include/scsi/scsi_netlink.h:33:2: error: unknown type name  uint16_t
  uint16_t msglen;
  ^~~~~~~~
./usr/include/scsi/scsi_netlink.h:34:33: error:  uint64_t  undeclared here (not in a function); did you mean  __uint128_t ?
 } __attribute__((aligned(sizeof(uint64_t))));
                                 ^~~~~~~~
                                 __uint128_t
./usr/include/scsi/scsi_netlink.h:78:2: error: expected specifier-qualifier-list before  uint64_t
  uint64_t vendor_id;
  ^~~~~~~~
In file included from <command-line>:32:0:
./usr/include/scsi/scsi_netlink_fc.h:46:2: error: expected specifier-qualifier-list before  uint64_t
  uint64_t seconds;
  ^~~~~~~~
make[2]: *** [scripts/Makefile.build;302: usr/include/scsi/scsi_netlink_fc.h.s] Error 1
make[2]: *** Waiting for unfinished jobs....
In file included from <command-line>:32:0:
./usr/include/scsi/scsi_netlink.h:29:2: error: unknown type name  uint8_t
  uint8_t version;
  ^~~~~~~
./usr/include/scsi/scsi_netlink.h:30:2: error: unknown type name  uint8_t
  uint8_t transport;
  ^~~~~~~
./usr/include/scsi/scsi_netlink.h:31:2: error: unknown type name  uint16_t
  uint16_t magic;
  ^~~~~~~~
./usr/include/scsi/scsi_netlink.h:32:2: error: unknown type name  uint16_t
  uint16_t msgtype;
  ^~~~~~~~
./usr/include/scsi/scsi_netlink.h:33:2: error: unknown type name  uint16_t
  uint16_t msglen;
  ^~~~~~~~
./usr/include/scsi/scsi_netlink.h:34:33: error:  uint64_t  undeclared here (not in a function); did you mean  __uint128_t ?
 } __attribute__((aligned(sizeof(uint64_t))));
                                 ^~~~~~~~
                                 __uint128_t
./usr/include/scsi/scsi_netlink.h:78:2: error: expected specifier-qualifier-list before  uint64_t
  uint64_t vendor_id;
  ^~~~~~~~
make[2]: *** [scripts/Makefile.build;302: usr/include/scsi/scsi_netlink.h.s] Error 1
In file included from <command-line>:32:0:
./usr/include/scsi/scsi_bsg_fc.h:69:2: error: unknown type name  uint8_t
  uint8_t  reserved;
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:72:2: error: unknown type name  uint8_t
  uint8_t  port_id[3];
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:90:2: error: unknown type name  uint8_t
  uint8_t  reserved;
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:93:2: error: unknown type name  uint8_t
  uint8_t  port_id[3];
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:114:2: error: unknown type name  uint8_t
  uint8_t  command_code;
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:117:2: error: unknown type name  uint8_t
  uint8_t  port_id[3];
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:154:2: error: unknown type name  uint32_t
  uint32_t status;  /* See FC_CTELS_STATUS_xxx */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:158:3: error: unknown type name  uint8_t
   uint8_t action;  /* fragment_id for CT REJECT */
   ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:159:3: error: unknown type name  uint8_t
   uint8_t reason_code;
   ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:160:3: error: unknown type name  uint8_t
   uint8_t reason_explanation;
   ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:161:3: error: unknown type name  uint8_t
   uint8_t vendor_unique;
   ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:177:2: error: unknown type name  uint8_t
  uint8_t  reserved;
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:180:2: error: unknown type name  uint8_t
  uint8_t  port_id[3];
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:185:2: error: unknown type name  uint32_t
  uint32_t preamble_word0; /* revision & IN_ID */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:186:2: error: unknown type name  uint32_t
  uint32_t preamble_word1; /* GS_Type, GS_SubType, Options, Rsvd */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:187:2: error: unknown type name  uint32_t
  uint32_t preamble_word2; /* Cmd Code, Max Size */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:207:2: error: unknown type name  uint64_t
  uint64_t vendor_id;
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:210:2: error: unknown type name  uint32_t
  uint32_t vendor_cmd[0];
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:217:2: error: unknown type name  uint32_t
  uint32_t vendor_rsp[0];
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:236:2: error: unknown type name  uint8_t
  uint8_t els_code;
  ^~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:254:2: error: unknown type name  uint32_t
  uint32_t preamble_word0; /* revision & IN_ID */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:255:2: error: unknown type name  uint32_t
  uint32_t preamble_word1; /* GS_Type, GS_SubType, Options, Rsvd */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:256:2: error: unknown type name  uint32_t
  uint32_t preamble_word2; /* Cmd Code, Max Size */
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:268:2: error: unknown type name  uint32_t
  uint32_t msgcode;
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:292:2: error: unknown type name  uint32_t
  uint32_t result;
  ^~~~~~~~
./usr/include/scsi/scsi_bsg_fc.h:295:2: error: unknown type name  uint32_t
  uint32_t reply_payload_rcv_len;
  ^~~~~~~~

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: esas2r: Prefer pcie_capability_read_word()

Commit 8c0d3a02c130 ("PCI: Add accessors for PCI Express Capability") added
accessors for the PCI Express Capability so that drivers didn't need to be
aware of differences between v1 and v2 of the PCI Express Capability.

Replace pci_read_config_word() and pci_write_config_word() calls with
pcie_capability_read_word() and pcie_capability_write_word().

Signed-off-by: Frederick Lawler <fred@fredlawl.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: csiostor: Prefer pcie_capability_read_word()

Commit 8c0d3a02c130 ("PCI: Add accessors for PCI Express Capability") added
accessors for the PCI Express Capability so that drivers didn't need to be
aware of differences between v1 and v2 of the PCI Express Capability.

Replace pci_read_config_word() and pci_write_config_word() calls with
pcie_capability_read_word() and pcie_capability_write_word().

Signed-off-by: Frederick Lawler <fred@fredlawl.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: sym53c8xx_2: remove redundant assignment to retv

Variable retv is initialized to a value that is never read and it is
re-assigned later. The initialization is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pmcraid: Fix a typo - pcmraid --> pmcraid

This should be 'pmcraid', not 'pcmraid'

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fas216: Mark expected switch fall-throughs

Mark switch cases where we are expecting to fall through.

Fix the following warnings (Building: rpc_defconfig arm):

drivers/scsi/arm/fas216.c: In function  fas216_disconnect_intr :
drivers/scsi/arm/fas216.c:913:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
   if (fas216_get_last_msg(info, info->scsi.msgin_fifo) == ABORT) {
      ^
drivers/scsi/arm/fas216.c:919:2: note: here
  default:    /* huh?     */
  ^~~~~~~
drivers/scsi/arm/fas216.c: In function  fas216_kick :
drivers/scsi/arm/fas216.c:1959:3: warning: this statement may fall through [-Wimplicit-fallthrough=]
   fas216_allocate_tag(info, SCpnt);
   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/scsi/arm/fas216.c:1960:2: note: here
  case TYPE_OTHER:
  ^~~~
drivers/scsi/arm/fas216.c: In function  fas216_busservice_intr :
drivers/scsi/arm/fas216.c:1413:3: warning: this statement may fall through [-Wimplicit-fallthrough=]
   fas216_stoptransfer(info);
   ^~~~~~~~~~~~~~~~~~~~~~~~~
drivers/scsi/arm/fas216.c:1414:2: note: here
  case STATE(STAT_STATUS, PHASE_SELSTEPS):/* Sel w/ steps -> Status       */
  ^~~~
drivers/scsi/arm/fas216.c:1424:3: warning: this statement may fall through [-Wimplicit-fallthrough=]
   fas216_stoptransfer(info);
   ^~~~~~~~~~~~~~~~~~~~~~~~~
drivers/scsi/arm/fas216.c:1425:2: note: here
  case STATE(STAT_MESGIN, PHASE_COMMAND): /* Command -> Message In */
  ^~~~
drivers/scsi/arm/fas216.c: In function  fas216_funcdone_intr :
drivers/scsi/arm/fas216.c:1573:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
   if ((stat & STAT_BUSMASK) == STAT_MESGIN) {
      ^
drivers/scsi/arm/fas216.c:1579:2: note: here
  default:
  ^~~~~~~
drivers/scsi/arm/fas216.c: In function  fas216_handlesync :
drivers/scsi/arm/fas216.c:605:20: warning: this statement may fall through [-Wimplicit-fallthrough=]
   info->scsi.phase = PHASE_MSGOUT_EXPECT;
   ~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~
drivers/scsi/arm/fas216.c:607:2: note: here
  case async:
  ^~~~

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: megaraid_sas: Introduce module parameter for default queue depth

This patch provides a module parameter and sysfs interface to select
whether the queue depth for each device should be based on the value
suggested by firmware (the default) or the maximum supported by the
controller (can_queue).

Although we have a sysfs interface per sdev to change the queue depth of
individual scsi devices, this implementation provides a single sysfs entry
per shost to switch between the controller max and the value reported by
firmware. The module parameter can provide an interface for one time grub
settings and provides persistent settings across the boot.

[mkp: tweaked commit desc]

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Chandrakanth Patil <chandrakanth.patil@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Fix a NULL pointer dereference

BUG: KASAN: null-ptr-deref in qla24xx_handle_plogi_done_event+0x134/0x9f0 [qla2xxx]
Read of size 4 at addr 00000000000000a0 by task swapper/2/0

CPU: 2 PID: 0 Comm: swapper/2 Not tainted 5.2.0-dbg+ #1
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
Call Trace:
 <IRQ>
 dump_stack+0x8a/0xd6
 __kasan_report.cold+0x5/0x41
 kasan_report+0x16/0x20
 __asan_load4+0x7e/0x80
 qla24xx_handle_plogi_done_event+0x134/0x9f0 [qla2xxx]
 qla2x00_els_dcmd2_sp_done+0x15f/0x230 [qla2xxx]
 qla24xx_els_ct_entry+0x3b3/0x610 [qla2xxx]
 qla24xx_process_response_queue+0x514/0x10e0 [qla2xxx]
 qla24xx_msix_rsp_q+0x80/0x100 [qla2xxx]
 __handle_irq_event_percpu+0x72/0x450
 handle_irq_event_percpu+0x74/0xf0
 handle_irq_event+0x5e/0x8f
 handle_edge_irq+0x13a/0x320
 handle_irq+0x30/0x40
 do_IRQ+0x91/0x190
 common_interrupt+0xf/0xf
 </IRQ>
RIP: 0010:default_idle+0x31/0x230

Fixes: 8777e4314d39 ("scsi: qla2xxx: Migrate NVME N2N handling into state machine") # v4.19.
Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Simplify qla24xx_async_abort_cmd()

Make the code easier to read by converting 'goto' statements into
'return' statements.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove two superfluous if-tests

This patch fixes the following Coverity complaint:

Null-checking sp->u.iocb_cmd.u.ctarg.rsp suggests that it may be null, but
it has already been dereferenced on all paths leading to the check.

See also commit e374f9f59281 ("scsi: qla2xxx: Migrate switch registration commands away from mailbox interface") # v4.16.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Introduce qla2x00_els_dcmd2_free()

This patch reduces code duplication.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Inline the qla2x00_fcport_event_handler() function

Instead of calling qla2x00_fcport_event_handler() and letting the switch
statement inside that function decide which other function to call, call
the latter function directly. Remove the event member from the event_arg
structure because it is no longer needed. Remove the
qla_handle_els_plogi_done() function because it is never called.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Report invalid mailbox status codes

It is easy to mix up the QLA_* and the MBS_* status codes. Complain loudly
if that happens.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove superfluous sts_entry_* casts

The C language supports implicit casting of void pointers to non-void
pointers. Remove explicit sts_entry_* casts that are not necessary.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Let the compiler check the type of the SCSI command context pointer

Split srb_cmd.ctx into two pointers such that the compiler can check the
type of that pointer.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Complain if sp->done() is not called from the completion path

Not calling sp->done() from the command completion path is a severe bug.
Hence complain loudly if that happens.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Make sure that aborted commands are freed

The LIO core requires that the target driver callback functions
.queue_data_in() and .queue_status() call target_put_sess_cmd() or
transport_generic_free_cmd(). These calls may happen synchronously or
asynchronously. Make sure that one of these LIO functions is called in case
a command has been aborted. This patch avoids that the code for removing a
session hangs due to commands that do not make progress.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: 694833ee00c4 ("scsi: tcm_qla2xxx: Do not allow aborted cmd to advance.") # v4.13.
Fixes: a07100e00ac4 ("qla2xxx: Fix TMR ABORT interaction issue between qla2xxx and TCM") # v4.5.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Introduce qla2xxx_get_next_handle()

This patch reduces code duplication.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Modify NVMe include directives

Since struct sg_table is used in nvme-fc-driver.h, include
<linux/scatterlist.h> from that header file.

Since no definitions or declarations from <linux/blk-mq.h> are used in the
qla_nvme.h header file, do not include <linux/blk-mq.h> from that header
file.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Make qlt_handle_abts_completion() more robust

Avoid that this function crashes if mcmd == NULL.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Fix a race condition between aborting and completing a SCSI command

Instead of allocating a struct srb dynamically from inside .queuecommand(),
set qla2xxx_driver_template.cmd_size such that struct scsi_cmnd and struct
srb are contiguous. Do not call QLA_QPAIR_MARK_BUSY() /
QLA_QPAIR_MARK_NOT_BUSY() for SRBs associated with SCSI commands. That is
safe because scsi_remove_host() is called before queue pairs are deleted
and scsi_remove_host() waits for all outstanding SCSI commands to finish.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Introduce the function qla2xxx_init_sp()

This patch does not change any functionality but makes the next patch
easier to read.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Enable type checking for the SRB free and done callback functions

Since all pointers passed to the srb_t.done() and srb_t.free() functions
have type srb_t, change the type of the first argument of these functions
from void * into struct srb *. This allows the compiler to verify the
argument types for these functions. This patch does not change any
functionality.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Rework key encoding in qlt_find_host_by_d_id()

Use the same approach for encoding the destination ID as the approach used
by qlt_update_vp_map().

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Set the responder mode if appropriate for ELS pass-through IOCBs

According to the firmware documentation responder mode must be set for ELS
pass-through IOCBs if a response is expected.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Make it explicit that ELS pass-through IOCBs use little endian

According to the firmware documentation the firmware expects all ELS
pass-through IOCB parameters in little endian format. Make this explicit.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Check secondary image if reading the primary image fails

This patch fixes several Coverity complaints about reading data that has
not been initialized.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Change the return type of qla24xx_read_flash_data()

This change makes it easier to detect qla24xx_read_flash_data() failures
and also to handle such failures. This change does not modify the behavior
of the driver since all callers ignore the qla24xx_read_flash_data()
return value.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Introduce the be_id_t and le_id_t data types for FC src/dst IDs

Introduce the be_id_t and le_id_t data types for Fibre Channel source and
destination ID formats supported by the firmware instead of using an
uint8_t[3] array. Introduce functions for converting from and to the
port_id_t data types. This patch does not change the behavior of the
qla2xxx driver but improves source code readability and also allows the
compiler to verify the endianness of Fibre Channel IDs.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Complain if a soft reset fails

Failure of a soft reset is a severe failure. Hence report such failures.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Use memcpy() and strlcpy() instead of strcpy() and strncpy()

This patch makes the string manipulation code easier to verify.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Check the PCI info string output buffer size

Pass the output buffer size to the code that generates a PCI info string
and check the output buffer size while generating a PCI info string.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Complain if waiting for pending commands times out

Such a timeout is a severe issue. Hence complain if waiting for pending
commands times out. This patch fixes a small bug: it modifies
qla82xx_chip_reset_cleanup() such that the "Done waiting" message is
reported if qla82xx_chip_reset_cleanup() succeeded instead of if that
function failed.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Declare fourth qla2x00_set_model_info() argument const

Make it clear to humans and also to the compiler that the string passed
as fourth argument is not modified.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Always check the qla2x00_wait_for_hba_online() return value

This patch fixes several Coverity complaints about not always checking
the qla2x00_wait_for_hba_online() return value.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Suppress multiple Coverity complaint about out-of-bounds accesses

This patch does not change any functionality.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Suppress a Coveritiy complaint about integer overflow

This patch fixes the following Coverity complaint:

Unintentional integer overflow (OVERFLOW_BEFORE_WIDEN)
overflow_before_widen: Potentially overflowing expression dma_eng_num *
65536U with type unsigned int (32 bits, unsigned) is evaluated using 32-bit
arithmetic, and then used in a context that expects an expression of type
uint64_t (64 bits, unsigned).

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove unreachable code from qla83xx_idc_lock()

This was detected by Coverity.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Fix qla24xx_process_bidir_cmd()

Set the r??_data_len variables before using these instead of after.

This patch fixes the following Coverity complaint:

const: At condition req_data_len != rsp_data_len, the value of req_data_len
must be equal to 0.
const: At condition req_data_len != rsp_data_len, the value of rsp_data_len
must be equal to 0.
dead_error_condition: The condition req_data_len != rsp_data_len cannot be
true.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: a9b6f722f62d ("[SCSI] qla2xxx: Implementation of bidirectional.") # v3.7.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Simplify a debug statement

Read the FC port state once instead of twice. This patch fixes the
following Coverity complaint:

Unchecked return value (CHECKED_RETURN)
check_return: Calling atomic_read without checking return value (as is
done elsewhere 80 out of 92 times).

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove dead code

Since sess == NULL before 'goto out_term2' is executed, the code under 'if
(sess)' cannot be reached. Hence remove that code. This was detected by
Coverity.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Complain if parsing the version string fails

This patch fixes a Coverity complaint about not checking the sscanf()
return value.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Complain if a mailbox command times out

This patch fixes the following Coverity complaint:

Unchecked return value (CHECKED_RETURN)
check_return: Calling wait_for_completion_timeout without checking return
value (as is done elsewhere 14 out of 17 times).

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Use strlcpy() instead of strncpy()

This patch fixes several gcc complaints about string truncation.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Do not corrupt vha->plogi_ack_list

Delete the PLOGIN ACK data structure from the vha->plogi_ack_list before
freeing that data structure to avoid that that list gets corrupted.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Report the firmware status code if a mailbox command fails

It is helpful when debugging this driver to have the firmware status code
available if a mailbox command fails. Hence report that firmware status
code.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Fix session lookup in qlt_abort_work()

Pass the correct session ID to find_sess_by_s_id() instead of passing an
uninitialized variable.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: 2d70c103fd2a ("[SCSI] qla2xxx: Add LLD target-mode infrastructure for >= 24xx series") # v3.5.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Simplify qla24xx_abort_sp_done()

Instead of explicitly checking whether a timeout has occurred, ignore the
del_timer() return value.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove two superfluous tests

Since qlt_remove_target() only calls qlt_release() if
vha->vha_tgt.qla_tgt != NULL, checking that pointer inside qlt_release()
is not necessary. This patch fixes the following Coverity complaint:

CID 188348 (#1 of 1): Dereference after null check (FORWARD_NULL)
var_deref_model: Passing null pointer &vha->vha_tgt.qla_tgt->tgt_list_entry
to list_del, which dereferences it.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove a superfluous pointer check

Checking a pointer after it has been dereferenced is not useful. This was
detected by Coverity.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Simplify qlt_lport_dump()

Simplify the implementation of this function by using the %phC format
specifier instead of using explicit for-loops.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Reduce the number of casts in GID list code

This patch makes the code that parses the GID list easier to read without
changing the behavior of the code.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Verify locking assumptions at runtime

Make sure that locking assumptions are verified at runtime if kernel
debugging is enabled.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Change data_dsd into an array

This patch does not change any functionality but fixes a Coverity complaint
about using a scalar as an array.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Declare qla_tgt_cmd.cdb const

Make it clear that the CDB is not modified after processing of a SCSI
command has started.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Reduce the scope of three local variables in qla2xxx_queuecommand()

This patch makes it clear that the tag, hwq and qpair variables are only
used in the mq path.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Change the return type of qla2x00_update_ms_fdmi_iocb() into void

The value returned by this function is not used. Hence change the return
type of this function into 'void' and remove the return statement.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Declare the fourth ql_dump_buffer() argument const

This patch makes it clear to humans and also to the compiler that
ql_dump_buffer() does not modify the memory the @buf argument points at.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove a superfluous forward declaration

Since qlt_make_local_sess() is defined before it is called, remove the
forward declaration of that function.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Remove an include directive from qla_mr.c

There is no bsg code in the qla_mr.c source file. Hence do not include
the <linux/bsg-lib.h> header file from qla_mr.c.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Include the <asm/unaligned.h> header file from qla_dsd.h

Since the put_unaligned_*() macros are used in this header file, include
the header file that defines these macros.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: 15b7a68c1d03 ("scsi: qla2xxx: Introduce the dsd32 and dsd64 data structures") # v5.2-rc1.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Use tabs instead of spaces for indentation

This patch only modifies whitespace.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Improve Linux kernel coding style conformance

Insert a space where required, surround complex expressions in macros with
parentheses, use the UL suffix instead of the (unsigned long) cast, do not
use line continuations when not necessary and do not explicitly initialize
static variables to zero.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Really fix qla2xxx_eh_abort()

I'm not sure how this happened but the patch that was intended to fix abort
handling was incomplete. This patch fixes that patch as follows:

 - If aborting the SCSI command failed, wait until the SCSI command
   completes.

 - Return SUCCESS instead of FAILED if an abort attempt races with SCSI
   command completion.

 - Since qla2xxx_eh_abort() increments the sp reference count by calling
   sp_get(), decrement the sp reference count before returning.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: 219d27d7147e ("scsi: qla2xxx: Fix race conditions in the code for aborting SCSI commands")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: Make qla2x00_abort_srb() again decrease the sp reference count

Since qla2x00_abort_srb() starts with increasing the reference count of
@sp, decrease that same reference count before returning.

Cc: Himanshu Madhani <hmadhani@marvell.com>
Fixes: 219d27d7147e ("scsi: qla2xxx: Fix race conditions in the code for aborting SCSI commands") # v5.2.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mpt3sas: Update driver version to 31.100.00.00

Updated driver version from 29.100.00.00 to 31.100.00.00 which is
equivalent to Phase 12 OOB.

Signed-off-by: Suganath Prabu <suganath-prabu.subramani@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mpt3sas: Run SAS DEVICE STATUS CHANGE EVENT from ISR

In some cases, like while performing extensive expander reset or phy reset,
user may observe that drives are not visible in OS. Driver's
firmware-worker thread is blocked for more than 120 seconds resulting in a
call trace.

1. Received target add event for Device A and hence driver has registered
this device to SML by calling sas_rphy_add(). SML has half added this
device and returned the control to the driver by quitting from
sas_rphy_add() API, and started some background scanning on this device A.

2. While background scanning is going on device A, driver has received SAS
DEVICE STATUS CHANGE EVENT with RC code "Internal device reset" event and
hence driver has set tm_busy flag for this Device A from FW worker thread
context. When tm_busy flag is set then driver return scsi commands with
device busy status asking the kernel to retry the command after some time.
So background scanning for device A will be waiting for this tm_busy to be
cleared.

3. Meanwhile driver has received a target add event for Device B and hence
driver called sas_rphy_add() API to register this device with SML. But
since background scanning for Device A is still pending and SML is not
quitting from sas_rphy_add(), the driver’s firmware worker thread got
blocked.

4. Now driver has received SAS DEVICE STATUS CHANGE EVENT with RC code
"Internal device reset complete" event. But as driver’s firmware worker
thread got blocked in Step3, it can’t process this event and it was not
clearing the tm_busy flag and deadlock occurred (where SML was waiting for
tm_busy flag to be cleared and our FW worker thread is waiting for SML to
quit from sas_device_rphy_add() API).

Same deadlock will be observed even if device B is getting removed in
step3. So to limit these types of deadlocks driver will process the SAS
DEVICE STATUS CHANGE EVENT events from ISR context instead of processing
this event from worker thread context.  This improvement avoids above
deadlock.

Signed-off-by: Suganath Prabu <suganath-prabu.subramani@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>