scsi: qedi: Complete TMF works before disconnect

We need to make sure that abort and reset completion work has completed
before ep_disconnect returns. After ep_disconnect we can't manipulate
cmds because libiscsi will call conn_stop and take onwership.

We are trying to make sure abort work and reset completion work has
completed before we do the cmd clean up in ep_disconnect. The problem is
that:

 1. the work function sets the QEDI_CONN_FW_CLEANUP bit, so if the work was
    still pending we would not see the bit set. We need to do this before
    the work is queued.

 2. If we had multiple works queued then we could break from the loop in
    qedi_ep_disconnect early because when abort work 1 completes it could
    clear QEDI_CONN_FW_CLEANUP. qedi_ep_disconnect could then see that
    before work 2 has run.

 3. A TMF reset completion work could run after ep_disconnect starts
    cleaning up cmds via qedi_clearsq. ep_disconnect's call to qedi_clearsq
    -> qedi_cleanup_all_io would might think it's done cleaning up cmds,
    but the reset completion work could still be running. We then return
    from ep_disconnect while still doing cleanup.

This replaces the bit with a counter to track the number of queued TMF
works, and adds a bool to prevent new works from starting from the
completion path once a ep_disconnect starts.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Pass send_iscsi_tmf task to abort

qedi_abort_work knows what task to abort so just pass it to send_iscsi_tmf.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix cleanup session block/unblock use

Drivers shouldn't be calling block/unblock session for cmd cleanup because
the functions can change the session state from under libiscsi.  This adds
a new a driver level bit so it can block all I/O the host while it drains
the card.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix TMF session block/unblock use

Drivers shouldn't be calling block/unblock session for tmf handling because
the functions can change the session state from under libiscsi.
iscsi_queuecommand's call to iscsi_prep_scsi_cmd_pdu->
iscsi_check_tmf_restrictions will prevent new cmds from being sent to qedi
after we've started handling a TMF. So we don't need to try and block it in
the driver, and we can remove these block calls.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Use GFP_NOIO for TMF allocation

We run from a workqueue with no locks held so use GFP_NOIO.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix TMF tid allocation

qedi_iscsi_abort_work and qedi_tmf_work both allocate a tid then call
qedi_send_iscsi_tmf which also allocates a tid. This removes the tid
allocation from the callers.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix use after free during abort cleanup

If qedi_tmf_work's qedi_wait_for_cleanup_request call times out we will
also force the clean up of the qedi_work_map but
qedi_process_cmd_cleanup_resp could still be accessing the qedi_cmd.

To fix this issue we extend where we hold the tmf_work_lock and back_lock
so the qedi_process_cmd_cleanup_resp access is serialized with the cleanup
done in qedi_tmf_work and any completion handling for the iscsi_task.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix race during abort timeouts

If the SCSI cmd completes after qedi_tmf_work calls iscsi_itt_to_task then
the qedi qedi_cmd->task_id could be freed and used for another cmd. If we
then call qedi_iscsi_cleanup_task with that task_id we will be cleaning up
the wrong cmd.

Wait to release the task_id until the last put has been done on the
iscsi_task. Because libiscsi grabs a ref to the task when sending the
abort, we know that for the non-abort timeout case that the task_id we are
referencing is for the cmd that was supposed to be aborted.

A latter commit will fix the case where the abort times out while we are
running qedi_tmf_work.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedi: Fix null ref during abort handling

If qedi_process_cmd_cleanup_resp finds the cmd it frees the work and sets
list_tmf_work to NULL, so qedi_tmf_work should check if list_tmf_work is
non-NULL when it wants to force cleanup.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Manish Rangankar <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Move pool freeing

This doesn't fix any bugs, but it makes more sense to free the pool after
we have removed the session. At that time we know nothing is touching any
of the session fields, because all devices have been removed and scans are
stopped.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Hold task ref during TMF timeout handling

For aborts, qedi needs to cleanup the FW then send the TMF from a worker
thread. While it's doing these the cmd could complete normally and the TMF
could time out. libiscsi would then complete the iscsi_task which will call
into the driver to cleanup the driver level resources while it still might
be accessing them for the cleanup/abort.

This has iscsi_eh_abort keep the iscsi_task ref if the TMF times out, so
qedi does not have to worry about if the task is being freed while in use
and does not need to get its own ref.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Flush block work before unblock

We set the max_active iSCSI EH works to 1, so all work is going to execute
in order by default. However, userspace can now override this in sysfs. If
max_active > 1, we can end up with the block_work on CPU1 and
iscsi_unblock_session running the unblock_work on CPU2 and the session and
target/device state will end up out of sync with each other.

This adds a flush of the block_work in iscsi_unblock_session.

Link: https://lore.kernel.org/r/[email protected]
Fixes: 1d726aa6ef57 ("scsi: iscsi: Optimize work queue flush use")
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Fix completion check during abort races

We have a ref to the task being aborted, so SCp.ptr will never be NULL. We
need to use iscsi_task_is_completed to check for the completed state.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Fix shost->max_id use

The iscsi offload drivers are setting the shost->max_id to the max number
of sessions they support. The problem is that max_id is not the max number
of targets but the highest identifier the targets can have. To use it to
limit the number of targets we need to set it to max sessions - 1, or we
can end up with a session we might not have preallocated resources for.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Fix conn use after free during resets

If we haven't done a unbind target call we can race where
iscsi_conn_teardown wakes up the EH thread and then frees the conn while
those threads are still accessing the conn ehwait.

We can only do one TMF per session so this just moves the TMF fields from
the conn to the session. We can then rely on the
iscsi_session_teardown->iscsi_remove_session->__iscsi_unbind_session call
to remove the target and it's devices, and know after that point there is
no device or scsi-ml callout trying to access the session.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Get ref to conn during reset handling

The comment in iscsi_eh_session_reset is wrong and we don't wait for the
EH to complete before tearing down the conn. This has us get a ref to the
conn when we are not holding the eh_mutex/frwd_lock so it does not get
freed from under us.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Have abort handler get ref to conn

If SCSI midlayer is aborting a task when we are tearing down the conn we
could free the conn while the abort thread is accessing the conn. This has
the abort handler get a ref to the conn so it won't be freed from under it.

Note: this is not needed for device/target reset because we are holding the
eh_mutex when accessing the conn.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Add iscsi_cls_conn refcount helpers

There are a couple places where we could free the iscsi_cls_conn while it's
still in use. This adds some helpers to get/put a refcount on the struct
and converts an exiting user. Subsequent commits will then use the helpers
to fix 2 bugs in the eh code.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: iscsi_tcp: Start socket shutdown during conn stop

Make sure the conn socket shutdown starts before we start the timer to fail
commands to upper layers.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: iscsi_tcp: Set no linger

Userspace (open-iscsi based tools at least) sets no linger on the socket to
prevent stale data from being sent. However, with the in-kernel cleanup if
userspace is not up the sockfd_put will release the socket without having
set that sockopt.

iscsid sets that opt at socket close time, but it seems ok to set this at
setup time in the kernel for all tools.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Fix in-kernel conn failure handling

Commit 0ab710458da1 ("scsi: iscsi: Perform connection failure entirely in
kernel space") has the following regressions/bugs that this patch fixes:

1. It can return cmds to upper layers like dm-multipath where that can
retry them. After they are successful the fs/app can send new I/O to the
same sectors, but we've left the cmds running in FW or in the net layer.
We need to be calling ep_disconnect if userspace is not up.

This patch only fixes the issue for offload drivers. iscsi_tcp will be
fixed in separate commit because it doesn't have a ep_disconnect call.

2. The drivers that implement ep_disconnect expect that it's called before
conn_stop. Besides crashes, if the cleanup_task callout is called before
ep_disconnect it might free up driver/card resources for session1 then they
could be allocated for session2. But because the driver's ep_disconnect is
not called it has not cleaned up the firmware so the card is still using
the resources for the original cmd.

3. The stop_conn_work_fn can run after userspace has done its recovery and
we are happily using the session. We will then end up with various bugs
depending on what is going on at the time.

We may also run stop_conn_work_fn late after userspace has called stop_conn
and ep_disconnect and is now going to call start/bind conn. If
stop_conn_work_fn runs after bind but before start, we would leave the conn
in a unbound but sort of started state where IO might be allowed even
though the drivers have been set in a state where they no longer expect
I/O.

4. Returning -EAGAIN in iscsi_if_destroy_conn if we haven't yet run the in
kernel stop_conn function is breaking userspace. We should have been doing
this for the caller.

Link: https://lore.kernel.org/r/[email protected]
Fixes: 0ab710458da1 ("scsi: iscsi: Perform connection failure entirely in kernel space")
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Rel ref after iscsi_lookup_endpoint()

Subsequent commits allow the kernel to do ep_disconnect. In that case we
will have to get a proper refcount on the ep so one thread does not delete
it from under another.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Use system_unbound_wq for destroy_work

Use the system_unbound_wq for async session destruction. We don't need a
dedicated workqueue for async session destruction because:

 1. perf does not seem to be an issue since we only allow 1 active work.

 2. it does not have deps with other system works and we can run them in
    parallel with each other.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Force immediate failure during shutdown

If the system is not up, we can just fail immediately since iscsid is not
going to ever answer our netlink events. We are already setting the
recovery_tmo to 0, but by passing stop_conn STOP_CONN_TERM we never will
block the session and start the recovery timer, because for that flag
userspace will do the unbind and destroy events which would remove the
devices and wake up and kill the eh.

Since the conn is dead and the system is going dowm this just has us use
STOP_CONN_RECOVER with recovery_tmo=0 so we fail immediately. However, if
the user has set the recovery_tmo=-1 we let the system hang like they
requested since they might have used that setting for specific reasons
(one known reason is for buggy cluster software).

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Drop suspend calls from ep_disconnect

libiscsi will now suspend the send/tx queue for the drivers so we can drop
it from the drivers ep_disconnect.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Stop queueing during ep_disconnect

During ep_disconnect we have been doing iscsi_suspend_tx/queue to block new
I/O but every driver except cxgbi and iscsi_tcp can still get I/O from
__iscsi_conn_send_pdu() if we haven't called iscsi_conn_failure() before
ep_disconnect. This could happen if we were terminating the session, and
the logout timed out before it was even sent to libiscsi.

Fix the issue by adding a helper which reverses the bind_conn call that
allows new I/O to be queued. Drivers implementing ep_disconnect can use this
to make sure new I/O is not queued to them when handling the disconnect.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: iscsi: Add task completion helper

This adds a helper to detect if a cmd has completed but is not yet freed.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Lee Duncan <[email protected]>
Signed-off-by: Mike Christie <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: megaraid_sas: Update driver version to 07.717.02.00-rc1

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Chandrakanth Patil <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: megaraid_sas: Handle missing interrupts while re-enabling IRQs

While reenabling the IRQ after IRQ poll there may be a small window for the
firmware to post the replies with interrupts raised. In that case the
driver will not see the interrupts which leads to I/O timeout.

This issue only happens when there are many I/O completions on a single
reply queue. This forces the driver to switch between the interrupt and IRQ
context.

Make the driver process the reply queue one more time after enabling the
IRQ.

Link: https://lore.kernel.org/linux-scsi/[email protected]/
Link: https://lore.kernel.org/r/[email protected]
Cc: Tomas Henzl <[email protected]>
Reported-by: kernel test robot <[email protected]>
Signed-off-by: Chandrakanth Patil <[email protected]>
Signed-off-by: Sumit Saxena <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: megaraid_sas: Early detection of VD deletion through RaidMap update

Consider the case where a VD is deleted and the targetID of that VD is
assigned to a newly created VD. If the sequence of deletion/addition of VD
happens very quickly there is a possibility that second event (VD add)
occurs even before the driver processes the first event (VD delete).  As
event processing is done in deferred context the device list remains the
same (but targetID is re-used) so driver will not learn the VD
deletion/additon. I/Os meant for the older VD will be directed to new VD
which may lead to data corruption.

Make driver detect the deleted VD as soon as possible based on the RaidMap
update and block further I/O to that device.

Link: https://lore.kernel.org/r/[email protected]
Reported-by: kernel test robot <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Chandrakanth Patil <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: megaraid_sas: Fix resource leak in case of probe failure

The driver doesn't clean up all the allocated resources properly when
scsi_add_host(), megasas_start_aen() function fails during the PCI device
probe.

Clean up all those resources.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Chandrakanth Patil <[email protected]>
Signed-off-by: Sumit Saxena <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: megaraid_sas: Send all non-RW I/Os for TYPE_ENCLOSURE device through firmware

The driver issues all non-ReadWrite I/Os for TYPE_ENCLOSURE devices through
the fast path with invalid dev handle. Fast path in turn directs all the
I/Os to the firmware. As firmware stopped handling those I/Os from SAS3.5
generation of controllers (Ventura generation and onwards) this will lead
to I/O failures.

Switch the driver to issue all the non-ReadWrite I/Os for TYPE_ENCLOSURE
devices directly to firmware for SAS3.5 generation of controllers and
later.

Link: https://lore.kernel.org/r/[email protected]
Cc: <[email protected]> # v5.11+
Signed-off-by: Chandrakanth Patil <[email protected]>
Signed-off-by: Sumit Saxena <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add event handling debug prints

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add EEDP DIF DIX support

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for DSN secure firmware check

Read PCI_EXT_CAP_ID_DSN to query security status.

The driver will throw a warning message when a non-secure type controller
is detected. The purpose of this interface is to avoid interacting with any
firmware which is not secured/signed by Broadcom.  Any tampering on
firmware component will be detected by hardware and it will be communicated
to the driver to avoid any further interaction with that component.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for PM suspend and resume

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Wait for pending I/O completions upon detection of VD I/O timeout

Wait for host I/O completion (default 180 seconds) if I/O timeout is
detected on VDs.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Print pending host I/Os for debugging

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Complete support for soft reset

Unlock the host diagnostic register, write the specific reset type to that
and wait for reset acknowledgment from the controller. If the reset is not
successful retry for the predefined number of times

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for threaded ISR

Register driver for threaded interrupts.

By default the driver will attempt I/O completion from interrupt context
(primary handler). Since the driver tracks per reply queue outstanding
I/Os, it will schedule threaded ISR if there are any outstanding I/Os
expected on that particular reply queue.

Threaded ISR (secondary handler) will loop for I/O completion as long as
there are outstanding I/Os (speculative method using same per reply queue
outstanding counter) or it has completed some X amount of commands
(something like budget).

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Hardware workaround for UNMAP commands to NVMe drives

The controller hardware can not handle certain UNMAP commands for NVMe
drives. Add support in the driver for checking those commands and handle
them appropriately.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Allow certain commands during pci-remove hook

Instead of driver returning DID_NO_CONNECT during driver unload allow SSU
and Sync Cache commands to be sent to the controller to flush any cached
data from the drive.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add change queue depth support

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Implement SCSI error handler hooks

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add bios_param SCSI host template hook

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Print IOC info for debugging

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for timestamp sync with firmware

This operation requests that the IOC update the TimeStamp.

When the I/O Unit is powered on it sets the TimeStamp field value to
0x0000_0000_0000_0000 and increments the current value every millisecond.
A host driver sets the TimeStamp field to the current time by using an
IOCInit request. The TimeStamp field is periodically updated by the host
driver.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for recovering controller

Detection of firmware fault or any kind of unresponsiveness in the
controller (any admin command which times out) results in resetting the
controller. The primary reset mechanisms used are either soft reset or diag
fault reset. A reset is performed if the host sets the ResetAction field in
the HostDiagnostic register to either 001b (soft reset) or 007b (diag fault
reset). After successfully resetting the controller the driver
reinitializes the controller by going through start of the day
initialization procedure. Pending I/Os during the reset are returned back
to the SCSI midlayer for retry.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Additional event handling

Implement support for handling the following MPI events:

 - MPI3_EVENT_SAS_BROADCAST_PRIMITIVE
 - MPI3_EVENT_CABLE_MGMT
 - MPI3_EVENT_ENERGY_PACK_CHANGE

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for PCIe device event handling

Implement support for the following PCIe-related MPI events:

 - MPI3_EVENT_PCIE_TOPOLOGY_CHANGE_LIST
 - MPI3_EVENT_PCIE_ENUMERATION

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for device add/remove event handling

Firmware can report various MPI Events. Enable support for processing the
following events related to device addition/removal to the driver:

 - MPI3_EVENT_DEVICE_ADDED
 - MPI3_EVENT_DEVICE_INFO_CHANGED
 - MPI3_EVENT_DEVICE_STATUS_CHANGE
 - MPI3_EVENT_ENCL_DEVICE_STATUS_CHANGE
 - MPI3_EVENT_SAS_TOPOLOGY_CHANGE_LIST
 - MPI3_EVENT_SAS_DISCOVERY
 - MPI3_EVENT_SAS_DEVICE_DISCOVERY_ERROR

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reported-by: kernel test robot <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for internal watchdog thread

The watchdog thread is the driver's internal thread which does a few things
such as detecting firmware faults, resetting the controller, performing
timestamp sync, etc.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add support for queue command processing

Send Port Enable Request to FW for Device Discovery.  As part of port
enable completion driver calls scan_start and scan_finished hooks.  SCSI
layer references like sdev, starget, etc. are added but actual device
discovery will be supported once driver adds complete event process
handling.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Cc: [email protected]
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Create operational request and reply queue pair

Create operational request and reply queue pair.

The MPI3 transport interface consists of an Administrative Request Queue,
an Administrative Reply Queue, and Operational Messaging Queues.  The
Operational Messaging Queues are the primary communication mechanism
between the host and the I/O Controller (IOC).  Request messages, allocated
in host memory, identify I/O operations to be performed by the IOC. These
operations are queued on an Operational Request Queue by the host driver.
Reply descriptors track I/O operations as they complete.  The IOC queues
these completions in an Operational Reply Queue.

To fulfil large contiguous memory requirement, driver creates multiple
segments and provide the list of segments. Each segment size should be 4K
which is a hardware requirement. An element array is contiguous or
segmented.  A contiguous element array is located in contiguous physical
memory.  A contiguous element array must be aligned on an element size
boundary.  An element's physical address within the array may be directly
calculated from the base address, the Producer/Consumer index, and the
element size.

Expected phased identifier bit is used to find out valid entry on reply
queue. Driver sets <ephase> bit and IOC inverts the value of this bit on
each pass.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Base driver code

Implement basic pci device driver requirements: Device probing, memory
allocation, mapping system registers, allocate irq lines, etc.

Source is managed in mainly three different files:

 - mpi3mr_fw.c:  Common code which interacts with underlying fw/hw.

 - mpi3mr_os.c:  Common code which interacts with SCSI midlayer.

 - mpi3mr_app.c: Common code which interacts with application/ioctl.
 This is currently work in progress.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpi3mr: Add mpi30 Rev-R headers and Kconfig

This adds the Kconfig and mpi30 headers.

Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Reviewed-by: Hannes Reinecke <[email protected]>
Reviewed-by: Tomas Henzl <[email protected]>
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Kashyap Desai <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: Fix a kernel-doc related formatting issue

Fix the following W=1 kernel build warning:

drivers/scsi/ufs/ufshcd.c:9773: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst

[mkp: upcase abbreviations]

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Alim Akhtar <[email protected]>
Signed-off-by: Bean Huo <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: isci: Use correctly sized target buffer for memcpy()

In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memcpy(), avoid intentionally writing across
neighboring array fields.

Switch from rsp_ui to resp_buf, since resp_ui isn't SSP_RESP_IU_MAX_SIZE
bytes in length. This avoids future compile-time warnings.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Kees Cook <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: esas2r: Switch to flexible array member

In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memcpy(), avoid intentionally writing across
neighboring array fields.

Remove old-style 1-byte array in favor of a flexible array[1] to avoid
future false-positive cross-field memcpy() warning in:

esas2r_vda.c:
memcpy(vi->cmd.gsv.version_info, esas2r_vdaioctl_versions, ...)

The change in struct size doesn't change other structure sizes (it is
already maxed out to 256 bytes, for example here:

        union {
                struct atto_ioctl_vda_scsi_cmd scsi;
                struct atto_ioctl_vda_flash_cmd flash;
                struct atto_ioctl_vda_diag_cmd diag;
                struct atto_ioctl_vda_cli_cmd cli;
                struct atto_ioctl_vda_smp_cmd smp;
                struct atto_ioctl_vda_cfg_cmd cfg;
                struct atto_ioctl_vda_mgt_cmd mgt;
                struct atto_ioctl_vda_gsv_cmd gsv;
                u8 cmd_info[256];
        } cmd;

No sizes are calculated using the enclosing structure, so no other
updates are needed.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Kees Cook <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: FlashPoint: Rename si_flags field

The BusLogic driver has build errors on ia64 due to a name collision (in
the #included FlashPoint.c file). Rename the struct field in struct
sccb_mgr_info from si_flags to si_mflags (manager flags) to mend the build.

This is the first problem. There are 50+ others after this one:

In file included from ../include/uapi/linux/signal.h:6,
                 from ../include/linux/signal_types.h:10,
                 from ../include/linux/sched.h:29,
                 from ../include/linux/hardirq.h:9,
                 from ../include/linux/interrupt.h:11,
                 from ../drivers/scsi/BusLogic.c:27:
../arch/ia64/include/uapi/asm/siginfo.h:15:27: error: expected ':', ',', ';', '}' or '__attribute__' before '.' token
   15 | #define si_flags _sifields._sigfault._flags
      |                           ^
../drivers/scsi/FlashPoint.c:43:6: note: in expansion of macro 'si_flags'
   43 |  u16 si_flags;
      |      ^~~~~~~~
In file included from ../drivers/scsi/BusLogic.c:51:
../drivers/scsi/FlashPoint.c: In function 'FlashPoint_ProbeHostAdapter':
../drivers/scsi/FlashPoint.c:1076:11: error: 'struct sccb_mgr_info' has no member named '_sifields'
 1076 |  pCardInfo->si_flags = 0x0000;
      |           ^~
../drivers/scsi/FlashPoint.c:1079:12: error: 'struct sccb_mgr_info' has no member named '_sifields'

Link: https://lore.kernel.org/r/[email protected]
Fixes: 391e2f25601e ("[SCSI] BusLogic: Port driver to 64-bit.")
Cc: "James E.J. Bottomley" <[email protected]>
Cc: "Martin K. Petersen" <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Jens Axboe <[email protected]>
Cc: Hannes Reinecke <[email protected]>
Cc: Khalid Aziz <[email protected]>
Cc: Khalid Aziz <[email protected]>
Reported-by: kernel test robot <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Signed-off-by: Randy Dunlap <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpt3sas: Fix fall-through warnings for Clang

In preparation to enable -Wimplicit-fallthrough for Clang, fix a couple
of warnings by explicitly adding break statements instead of just letting
the code fall through to the next case.

Link: https://github.com/KSPP/linux/issues/115
Link: https://lore.kernel.org/r/20210528200828.GA39349@embeddedor
Reviewed-by: Kees Cook <[email protected]>
Signed-off-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qla2xxx: Log PCI address in qla_nvme_unregister_remote_port()

Pass in fcport->vha to ql_log() in order to add the PCI address to the log.

Currently NULL is passed in which gives this confusing log entry:

> qla2xxx [0000:00:00.0]-2112: : qla_nvme_unregister_remote_port: unregister remoteport on 0000000009d6a2e9 50000973981648c7

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Daniel Wagner <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: ufs-mediatek: Disable HCI before HW reset

MediaTek ufshci needs to be disabled before HW reset to avoid potential
issues.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Stanley Chu <[email protected]>
Signed-off-by: Alice.Chao <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: core: Export ufshcd_hba_stop()

Export ufshcd_hba_stop() to allow vendors to disable HCI in variant ops.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Stanley Chu <[email protected]>
Signed-off-by: Alice.Chao <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: Suppress false positive unhandled interrupt messages

From ufshcd_transfer_req_compl():

    Resetting interrupt aggregation counters first and reading the
    DOOR_BELL afterward allows us to handle all the completed requests.  In
    order to prevent other interrupts starvation the DB is read once after
    reset. The down side of this solution is the possibility of false
    interrupt if device completes another request after resetting
    aggregation and before reading the DB.

Prevent that ufshcd_intr() reports a false positive "Unhandled interrupt"
message if the above scenario is triggered.

Link: https://lore.kernel.org/r/[email protected]
Cc: Stanley Chu <[email protected]>
Cc: Can Guo <[email protected]>
Cc: Bean Huo <[email protected]>
Cc: Jaegeuk Kim <[email protected]>
Cc: Asutosh Das <[email protected]>
Suggested-by: Jaegeuk Kim <[email protected]>
Signed-off-by: Bart Van Assche <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpt3sas: Handle firmware faults during second half of IOC init

If a firmware fault occurs while scanning the devices during IOC
initialization then the driver issues the hard reset operation to recover
the IOC. However, the driver is not issuing a Port enable request
message as part of hard reset operation during IOC initialization.  Due to
this, the driver will not receive get any device discovery-related events
and hence devices will not be accessible.

Teach the driver to gracefully handle firmware faults while scanning for
target devices during IOC initialization. Make the driver issue a port
enable request message as part of hard reset operation. This permits
receiving device discovery-related events from the firmware after the hard
reset operation completes.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Suganath Prabu S <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpt3sas: Handle firmware faults during first half of IOC init

During first half of IOC initialization (i.e.  before going for device
scanning), if any firmware fault occurs then driver is aborting the IOC
initialization operation.

Modify the driver to issue a diag reset operation to recover IOC from fault
state and reinitialize the IOC.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Suganath Prabu S <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: mpt3sas: Fix deadlock while cancelling the running firmware event

Do not cancel current running firmware event work if the event type is
different from MPT3SAS_REMOVE_UNRESPONDING_DEVICES.  Otherwise a deadlock
can be observed while cancelling the current firmware event work if a hard
reset operation is called as part of processing the current event.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Suganath Prabu S <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: core: Cap scsi_host cmd_per_lun at can_queue

The sysfs handling function sdev_store_queue_depth() enforces that the sdev
queue depth cannot exceed shost can_queue. The initial sdev queue depth
comes from shost cmd_per_lun. However, the LLDD may manually set
cmd_per_lun to be larger than can_queue, which leads to an initial sdev
queue depth greater than can_queue.

Such an issue was reported in [0], which caused a hang. That has since been
fixed in commit fc09acb7de31 ("scsi: scsi_debug: Fix cmd_per_lun, set to
max_queue").

Stop this possibly happening for other drivers by capping shost cmd_per_lun
at shost can_queue.

[0] https://lore.kernel.org/linux-scsi/YHaez6iN2HHYxYOh@T590/

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Ming Lei <[email protected]>
Reviewed-by: Bart Van Assche <[email protected]>
Signed-off-by: John Garry <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Update lpfc version to 12.8.0.10

Update lpfc version to 12.8.0.10

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Reregister FPIN types if ELS_RDF is received from fabric controller

FC-LS-5 specifies that a received RDF implies a possible change to fabric
supported diagnostic functions. Endpoints are to re-perform the RDF
exchange with the fabric to enable possible new features or adapt to
changes in values.

This patch adds the logic to RDF receive to re-perform the RDF exchange
with the switch.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Add a option to enable interlocked ABTS before job completion

Default behavior for the driver, when aborting an I/O, is to terminate the
I/O with the adapter. The adapter will initiate an ABTS to terminate the
exchange on the link and mark the exchange is terminated so that no further
use of the sgl or any traffic for the exchange is worked on. Completion on
the Abort is then posted to the driver, which as the I/O is terminated can
complete the I/O to the OS. This completion may occur prior to the ABTS
handshake completing on the wire. The ABTS handshake can take a long time
to complete with timeouts and retries reaching 60+ seconds. Note: if
retries fail, LOGO occurs.

Some devices want to ensure that the ABTS handshake fully completes (this
device has fully ack'd it) before the I/O completion is posted back to the
OS, where a failed I/O may be retried via a different path.

To support this behavior, an option was added to the driver to change I/O
completion from the Abort cmd completion to the Exchange termination (aka
ABTS) completion.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix crash when lpfc_sli4_hba_setup() fails to initialize the SGLs

The driver is encountering a crash in lpfc_free_iocb_list() while
performing initial attachment.

Code review found this to be an errant failure path that was taken, jumping
to a tag that then referenced structures that were uninitialized.

Fix the failure path.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Ignore GID-FT response that may be received after a link flip

When a link bounce happens, there is a possibility that responses to
requests posted prior to the link bounce could be received. This is
problematic as the counter to track reglogin completion after link up can
become out of sync with the real state.

As there is no reason to process a request made in a prior link up context,
eliminate all the disturbance by tagging the request with the event_tag
maintained by the SLI Port for the link. The event_tag will change on every
link state transition.  As long as the tag matches the current event_tag,
the response can be processed. If it doesn't match, just discard the
response.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix node handling for Fabric Controller and Domain Controller

During link bounce testing, RPI counts were seen to differ from the number
of nodes. For fabric and domain controllers, a temporary RPI is assigned,
but the code isn't registering it. If the nodes do go away, such as on link
down, the temporary RPI isn't being released.

Change the way these two fabric services are managed, make them behave like
any other remote port. Register the RPI and register with the transport.
Never leave the nodes in a NPR or UNUSED state where their RPI is in limbo.
This allows them to follow normal dev_loss_tmo handling, RPI refcounting,
and normal removal rules. It also allows fabric I/Os to use the RPI for
traffic requests.

Note: There is some logic that still has a couple of exceptions when the
Domain controller (0xfffcXX). There are cases where the fabric won't have a
valid login but will send RDP. Other times, it will it send a LOGO then an
RDP. It makes for ad-hoc behavior to manage the node. Exceptions are
documented in the code.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix Node recovery when driver is handling simultaneous PLOGIs

When lpfc is handling a solicited and unsolicited PLOGI with another
initiator, the remote initiator is never recovered. The node for the
initiator is erroneouosly removed and all resources released.

In lpfc_cmpl_els_plogi(), when lpfc_els_retry() returns a failure code, the
driver is calling the state machine with a device remove event because the
remote port is not currently registered with the SCSI or NVMe
transports. The issue is that on a PLOGI "collision" the driver correctly
aborts the solicited PLOGI and allows the unsolicited PLOGI to complete the
process, but this process is interrupted with a device_rm event.

Introduce logic in the PLOGI completion to capture the PLOGI collision
event and jump out of the routine.  This will avoid removal of the node.
If there is no collision, the normal node removal will occur.

Fixes: 52edb2caf675 ("scsi: lpfc: Remove ndlp when a PLOGI/ADISC/PRLI/REG_RPI ultimately fails")
Cc: <[email protected]> # v5.11+
Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Add ndlp kref accounting for resume RPI path

The driver is crashing due to a bad pointer during driver load due in an
adisc acc receive routine. The driver is missing node get/put in the
mbx_resume_rpi paths.

Fix by adding the proper gets and puts into the resume_rpi path.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix "Unexpected timeout" error in direct attach topology

An 'unexpected timeout' message may be seen in a point-2-point topology.
The message occurs when a PLOGI is received before the driver is notified
of FLOGI completion. The FLOGI completion failure causes discovery to be
triggered for a second time. The discovery timer is restarted but no new
discovery activity is initiated, thus the timeout message eventually
appears.

In point-2-point, when discovery has progressed before the FLOGI completion
is processed, it is not a failure. Add code to FLOGI completion to detect
that discovery has progressed and exit the FLOGI handling (noop'ing it).

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix non-optimized ERSP handling

When processing an NVMe ERSP IU which didn't match the optimized CQE-only
path, the status was being left to the WQE status. WQE status is non-zero
as it is indicating a non-optimized completion that needs to be handled by
the driver.

Fix by clearing the status field when falling into the non-optimized
case. Log message added to track optimized vs non-optimized debug.

Link: https://lore.kernel.org/r/[email protected]
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: lpfc: Fix unreleased RPIs when NPIV ports are created

While testing NPIV and watching logins and used RPI levels, it was seen the
used RPI count was much higher than the number of remote ports discovered.

Code inspection showed that remote port removals on any NPIV instance are
releasing the RPI, but not performing an UNREG_RPI with the adapter thus
the reference counting never fully drops and the RPI is never fully
released. This was happening on NPIV nodes due to a log of fabric ELS's to
fabric addresses. This lack of UNREG_RPI was introduced by a prior node
rework patch that performed the UNREG_RPI as part of node cleanup.

To resolve the issue, do the following:

 - Restore the RPI release code, but move the location to so that it is in
   line with the new node cleanup design.

 - NPIV ports now release the RPI and drop the node when the caller sets
   the NLP_RELEASE_RPI flag.

 - Set the NLP_RELEASE_RPI flag in node cleanup which will trigger a
   release of RPI to free pool.

 - Ensure there's an UNREG_RPI at LOGO completion so that RPI release is
   completed.

 - Stop offline_prep from skipping nodes that are UNUSED. The RPI may
   not have been released.

 - Stop the default RPI handling in lpfc_cmpl_els_rsp() for SLI4.

 - Fixed up debugfs RPI displays for better debugging.

Fixes: a70e63eee1c1 ("scsi: lpfc: Fix NPIV Fabric Node reference counting")
Link: https://lore.kernel.org/r/[email protected]
Cc: <[email protected]> # v5.11+
Co-developed-by: Justin Tee <[email protected]>
Signed-off-by: Justin Tee <[email protected]>
Signed-off-by: James Smart <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: scsi_dh_alua: Retry RTPG on a different path after failure

If an RTPG fails, we can't infer anything wrt. the state of the ports in
the port group except that we were unable to reach the one port on which
the RTPG had failed. "offline" is just a secondary port state, which means
that we can't infer the state of any port in the PG from the failure (in
fact, even the failed port might still be in "active/optimized" primary
port access state).

Therefore, when we encounter an RTPG failure, we should retry the RTPG on a
different port. This avoids falsely setting port states to offline for
unreachable ports. To do this, ports on which an RTPG has failed are
temporarily set to "disabled" to avoid repeating the failed I/O on the same
target port. Once the RTPG has either succeeded on one port or failed on
all ports of the PG, the ports are enabled again.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Martin Wilck <[email protected]>
Signed-off-by: Hannes Reinecke <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qla2xxx: Remove redundant assignment to rval

Variable rval is set to QLA_SUCCESS but this value is never read as it is
overwritten later on. Hence it is a redundant assignment and can be
removed.

Clean up the following clang-analyzer warning:

drivers/scsi/qla2xxx/qla_init.c:4359:2: warning: Value stored to 'rval'
is never read [clang-analyzer-deadcode.DeadStores].

Link: https://lore.kernel.org/r/1620643206-127930-1-git-send-email-jiapeng.chong@linux.alibaba.com
Reported-by: Abaci Robot <[email protected]>
Signed-off-by: Jiapeng Chong <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: ufs-exynos: Make a const array static, makes object smaller

Don't populate the const array granularity_tbl on the stack but instead
make it static. Makes the object code smaller by 190 bytes:

Before:
   text    data     bss     dec     hex filename
  25563    6908       0   32471    7ed7 ./drivers/scsi/ufs/ufs-exynos.o

After:
   text    data     bss     dec     hex filename
  25213    7068       0   32281    7e19 ./drivers/scsi/ufs/ufs-exynos.o

(gcc version 10.3.0)

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Colin Ian King <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: fas216: Use fallthrough pseudo-keyword

Replace /*FALLTHROUGH*/ comment with pseudo-keyword macro 'fallthrough'.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Wei Ming Chen <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: core: Clean up whitespace

checkpatch reports the following errors:

ERROR: space prohibited before that ',' (ctx:WxW)
#945: FILE: drivers/scsi/ufs/ufshcd.h:945:
+int ufshcd_init(struct ufs_hba * , void __iomem * , unsigned int);
                                  ^

ERROR: space prohibited before that ',' (ctx:WxW)
#945: FILE: drivers/scsi/ufs/ufshcd.h:945:
+int ufshcd_init(struct ufs_hba * , void __iomem * , unsigned int);
                                                   ^
Remove unnecessary whitespace in ufshcd.h.

Link: https://lore.kernel.org/r/2038148563.21621340102306.JavaMail.epsvc@epcpadp3
Signed-off-by: Keoseong Park <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: Fix spelling mistakes in header files

Fix some spelling mistakes in comments:

  pathes ==> paths
  Resouce ==> Resource
  retreived ==> retrieved
  recevied ==> received
  interruped ==> interrupted

[mkp: kept 'keep-alives' and 'busses']

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Zhen Lei <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: core: Remove leading spaces in Kconfig

Remove leading spaces before tabs in Kconfig file(s) by running the
following command:

  $ find drivers/scsi -name 'Kconfig*' | xargs sed -r -i 's/^[ ]+\t/\t/'

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Juerg Haefliger <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: target: tcmu: Fix boolreturn.cocci warnings

drivers/target/target_core_user.c:1424:9-10: WARNING: return of 0/1 in function 'tcmu_handle_completions' with return type bool

 Return statements in functions returning bool should use
 true/false instead of 1/0.

Generated by: scripts/coccinelle/misc/boolreturn.cocci

Link: https://lore.kernel.org/r/20210515230358.GA97544@60d1edce16e0
Fixes: 9814b55cde05 ("scsi: target: tcmu: Return from tcmu_handle_completions() if cmd_id not found")
CC: Bodo Stroesser <[email protected]>
Reported-by: kernel test robot <[email protected]>
Acked-by: Bodo Stroesser <[email protected]>
Signed-off-by: kernel test robot <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: Use designated initializers in ufs_pm_lvl_states[]

The comments in the enum ufs_pm_level definition are redundant. Remove the
comments from the ufs_pm_level enum and use designated initializers in the
ufs_pm_lvl_states[] definition instead.

Link: https://lore.kernel.org/r/[email protected]
Cc: Stanley Chu <[email protected]>
Cc: Can Guo <[email protected]>
Cc: Bean Huo <[email protected]>
Cc: Jaegeuk Kim <[email protected]>
Cc: Asutosh Das <[email protected]>
Signed-off-by: Bart Van Assche <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: hisi_sas: Propagate errors in interrupt_init_v1_hw()

After commit 6c11dc060427 ("scsi: hisi_sas: Fix IRQ checks") we have the
error codes returned by platform_get_irq() ready for the propagation
upsream in interrupt_init_v1_hw() -- that will fix still broken deferred
probing. Let's propagate the error codes from devm_request_irq() as well
since I don't see the reason to override them with -ENOENT...

Link: https://lore.kernel.org/r/[email protected]
Acked-by: John Garry <[email protected]>
Signed-off-by: Sergey Shtylyov <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: bfa: Fix inconsistent indenting

Eliminate the follow smatch warning:

drivers/scsi/bfa/bfa_svc.c:3176 bfa_fcport_send_enable() warn:
inconsistent indenting.

Link: https://lore.kernel.org/r/1621590368-72041-1-git-send-email-jiapeng.chong@linux.alibaba.com
Reported-by: Abaci Robot <[email protected]>
Signed-off-by: Jiapeng Chong <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: bfa: Fix typo

Change 'chnage' to 'change'.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: zuoqilin <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: pmcraid: Fix typos

Change "avaibale" and "avaible" to "available".

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: zuoqilin <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: scsi_transport_fc: Remove double FC_FPORT_DELETED in mask creation

Remove the double listed FC_FPORT_DELETING from the mask creation.

Commit 260f4aeddb48 ("scsi: scsi_transport_fc: return -EBUSY for deleted
vport") added VC_VPORT_DELETING to the flag masks. This is not necessary as
FC_FPORT_DEL is defined as VC_FPORT_DELETED | FC_FPORT_DELETING.

Link: https://lore.kernel.org/r/[email protected]
Cc: Hannes Reinecke <[email protected]>
Signed-off-by: Daniel Wagner <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: arcmsr: Update driver version to v1.50.00.05-20210429

Update driver version to v1.50.00.05-20210429.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: ching Huang <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: arcmsr: Fix doorbell status being updated late on ARC-1886

It is possible for the IOP to be delayed in updating the doorbell
status. The doorbell status should not be 0 so loop until the value
changes.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: ching Huang <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: qedf: Use vzalloc() instead of vmalloc()/memset(0)

Use vzalloc() instead of vmalloc() and memset(0) to simpify the code.

Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Himanshu Madhani <[email protected]>
Signed-off-by: Yang Yingliang <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: ufs: ufs-exynos: Move definitions from .h to .c

In the Linux kernel definitions of data structures should occur in .c
files. Hence move the exynos7_uic_attr definition from a .h into a .c
file. Additionally, declare exynos_ufs_drvs static. This patch fixes the
following two sparse warnings:

drivers/scsi/ufs/ufs-exynos.h:248:28: warning: symbol 'exynos_ufs_drvs' was not declared. Should it be static?
drivers/scsi/ufs/ufs-exynos.h:250:28: warning: symbol 'exynos7_uic_attr' was not declared. Should it be static?

Link: https://lore.kernel.org/r/[email protected]
Cc: Alim Akhtar <[email protected]>
Cc: Kiwoong Kim <[email protected]>
Reviewed-by: Alim Akhtar <[email protected]>
Signed-off-by: Bart Van Assche <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: 3w-9xxx: Fix endianness issues in command packets

The controller expects all data it sends/receives to be little-endian.
Therefore, the packet struct definitions should use the __le16/32/64
types. Once those are correct, sparse reports several issues with the
driver code, which are fixed here as well.

The main issue observed was at the call to scsi_set_resid(), where the
byteswapped parameter would eventually trigger the alignment check at
drivers/scsi/sd.c:2009. At that point, the kernel would continuously
complain about an "Unaligned partial completion", and no further I/O could
occur.

This gets the controller working on big endian powerpc64.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Samuel Holland <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

scsi: 3w-9xxx: Reduce scope of structure packing

Currently, all command packet structs used by this driver are packed.
However, only one (TW_SG_Entry) actually needs to be packed, because it
uses 64-bit addresses at 32-bit alignment. To improve the quality of
generated code, stop packing all of the other command packet structs.  This
requires adjusting the type of one misaligned "reserved" member.

After this change, pahole reports that only one type had its layout change:
the tw_compat_info member of TW_Device_Extension is now naturally aligned.

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Samuel Holland <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>