scsi: libsas: Switch back to original event notifiers API

libsas event notifiers required an extension where gfp_t flags must be
explicitly passed. For bisectability, a temporary _gfp() variant of such
functions were added. All call sites then got converted use the _gfp()
variants and explicitly pass GFP context. Having no callers left, the
original libsas notifiers were then modified to accept gfp_t flags by
default.

Switch back to the original event notifiers API, while still passing GFP
context.  The _gfp() notifier variants will be removed afterwards.

Link: https://lore.kernel.org/r/20210118100955.1761652-17-a.darwish@linutronix.de
Cc: Jason Yan <yanaijie@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm80xx: Switch back to original libsas event notifiers

libsas event notifiers required an extension where gfp_t flags must be
explicitly passed. For bisectability, a temporary _gfp() variant of such
functions were added. All call sites then got converted use the _gfp()
variants and explicitly pass GFP context. Having no callers left, the
original libsas notifiers were then modified to accept gfp_t flags by
default.

Switch back to the original libas API, while still passing GFP context.
The libsas _gfp() variants will be removed afterwards.

Link: https://lore.kernel.org/r/20210118100955.1761652-16-a.darwish@linutronix.de
Cc: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Reviewed-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic94xx: Switch back to original libsas event notifiers

libsas event notifiers required an extension where gfp_t flags must be
explicitly passed. For bisectability, a temporary _gfp() variant of such
functions were added. All call sites then got converted use the _gfp()
variants and explicitly pass GFP context. Having no callers left, the
original libsas notifiers were then modified to accept gfp_t flags by
default.

Switch back to the original libas API, while still passing GFP context.
The libsas _gfp() variants will be removed afterwards.

Link: https://lore.kernel.org/r/20210118100955.1761652-15-a.darwish@linutronix.de
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas: Switch back to original libsas event notifiers

libsas event notifiers required an extension where gfp_t flags must be
explicitly passed. For bisectability, a temporary _gfp() variant of such
functions were added. All call sites then got converted use the _gfp()
variants and explicitly pass GFP context. Having no callers left, the
original libsas notifiers were then modified to accept gfp_t flags by
default.

Switch back to the original libas API, while still passing GFP context.
The libsas _gfp() variants will be removed afterwards.

Link: https://lore.kernel.org/r/20210118100955.1761652-14-a.darwish@linutronix.de
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: libsas: Add gfp_t flags parameter to event notifications

All call-sites of below libsas APIs:

  - sas_alloc_event()
  - sas_notify_port_event()
  - sas_notify_phy_event()

have been converted to use the _gfp()-suffixed version.  Modify the
original APIs above to take a gfp_t flags parameter by default.

For bisectability, call-sites will be modified again to use the original
libsas APIs (while passing gfp_t). The temporary _gfp()-suffixed versions
can then be removed.

Link: https://lore.kernel.org/r/20210118100955.1761652-13-a.darwish@linutronix.de
Cc: Jason Yan <yanaijie@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas: Pass gfp_t flags to libsas event notifiers

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

Below are the context analysis for modified functions:

=> hisi_sas_bytes_dmaed():

Since it is invoked from both process and atomic contexts, let its callers
pass the gfp_t flags:

  * hisi_sas_main.c:
  ------------------

    hisi_sas_phyup_work(): workqueue context
      -> hisi_sas_bytes_dmaed(..., GFP_KERNEL)

    hisi_sas_controller_reset_done(): has an msleep()
      -> hisi_sas_rescan_topology()
        -> hisi_sas_phy_down()
          -> hisi_sas_bytes_dmaed(..., GFP_KERNEL)

    hisi_sas_debug_I_T_nexus_reset(): calls wait_for_completion_timeout()
      -> hisi_sas_phy_down()
        -> hisi_sas_bytes_dmaed(..., GFP_KERNEL)

  * hisi_sas_v1_hw.c:
  -------------------

    int_abnormal_v1_hw(): irq handler
      -> hisi_sas_phy_down()
        -> hisi_sas_bytes_dmaed(..., GFP_ATOMIC)

  * hisi_sas_v[23]_hw.c:
  ----------------------

    int_phy_updown_v[23]_hw(): irq handler
      -> phy_down_v[23]_hw()
        -> hisi_sas_phy_down()
          -> hisi_sas_bytes_dmaed(..., GFP_ATOMIC)

=> int_bcast_v1_hw() and phy_bcast_v3_hw():

Both are invoked exclusively from irq handlers. Pass GFP_ATOMIC.

Link: https://lore.kernel.org/r/20210118100955.1761652-12-a.darwish@linutronix.de
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic94xx: Pass gfp_t flags to libsas event notifiers

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

Context analysis:

  aic94xx_hwi.c: asd_dl_tasklet_handler()
    -> asd_ascb::tasklet_complete()
    == escb_tasklet_complete()
      -> aic94xx_scb.c: asd_phy_event_tasklet()
      -> aic94xx_scb.c: asd_bytes_dmaed_tasklet()
      -> aic94xx_scb.c: asd_link_reset_err_tasklet()
      -> aic94xx_scb.c: asd_primitive_rcvd_tasklet()

All functions are invoked by escb_tasklet_complete(), which is invoked by
the tasklet handler. Pass GFP_ATOMIC.

Link: https://lore.kernel.org/r/20210118100955.1761652-11-a.darwish@linutronix.de
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm80xx: Pass gfp_t flags to libsas event notifiers

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

Call chain analysis, pm8001_hwi.c:

  pm8001_interrupt_handler_msix() || pm8001_interrupt_handler_intx() || pm8001_tasklet()
    -> PM8001_CHIP_DISP->isr() = pm80xx_chip_isr()
      -> process_oq [spin_lock_irqsave(&pm8001_ha->lock, ...)]
        -> process_one_iomb()
          -> mpi_hw_event()
            -> hw_event_sas_phy_up()
              -> pm8001_bytes_dmaed()
            -> hw_event_sata_phy_up
              -> pm8001_bytes_dmaed()

All functions are invoked by process_one_iomb(), which is invoked by the
interrupt service routine and the tasklet handler. A similar call chain is
also found at pm80xx_hwi.c. Pass GFP_ATOMIC.

For pm8001_sas.c, pm8001_phy_control() runs in task context as it calls
wait_for_completion() and msleep().  Pass GFP_KERNEL.

Link: https://lore.kernel.org/r/20210118100955.1761652-10-a.darwish@linutronix.de
Cc: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Reviewed-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: libsas: Pass gfp_t flags to event notifiers

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

Context analysis:

  - sas_enable_revalidation(): process, acquires mutex
  - sas_resume_ha(): process, calls wait_event_timeout()

Link: https://lore.kernel.org/r/20210118100955.1761652-9-a.darwish@linutronix.de
Cc: Jason Yan <yanaijie@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: isci: Pass gfp_t flags in isci_port_bc_change_received()

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

libsas sas_notify_port_event() is called from
isci_port_bc_change_received(). Below is the context analysis for all of
its call chains:

host.c: sci_controller_error_handler(): atomic, irq handler     (*)
OR host.c: sci_controller_completion_handler(), atomic, tasklet (*)
  -> sci_controller_process_completions()
    -> sci_controller_event_completion()
      -> phy.c: sci_phy_event_handler()
        -> port.c: sci_port_broadcast_change_received()
          -> isci_port_bc_change_received()

host.c: isci_host_init()                                        (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_initialize(), atomic                        (*)
    -> port_config.c: sci_port_configuration_agent_initialize()
      -> sci_mpc_agent_validate_phy_configuration()
        -> port.c: sci_port_add_phy()
          -> sci_port_set_phy()
            -> phy.c: sci_phy_set_port()
              -> port.c: sci_port_broadcast_change_received()
                -> isci_port_bc_change_received()

port_config.c: apc_agent_timeout(), atomic, timer callback      (*)
  -> sci_apc_agent_configure_ports()
    -> port.c: sci_port_add_phy()
      -> sci_port_set_phy()
        -> phy.c: sci_phy_set_port()
          -> port.c: sci_port_broadcast_change_received()
            -> isci_port_bc_change_received()

phy.c: enter SCI state: *SCI_PHY_STOPPED*                       # Cont. from [1]
  -> sci_phy_stopped_state_enter()
    -> host.c: sci_controller_link_down()
      -> ->link_down_handler()
      == port_config.c: sci_apc_agent_link_down()
        -> port.c: sci_port_remove_phy()
          -> sci_port_clear_phy()
            -> phy.c: sci_phy_set_port()
              -> port.c: sci_port_broadcast_change_received()
                -> isci_port_bc_change_received()

phy.c: enter SCI state: *SCI_PHY_STARTING*                      # Cont. from [2]
  -> sci_phy_starting_state_enter()
    -> host.c: sci_controller_link_down()
      -> ->link_down_handler()
      == port_config.c: sci_apc_agent_link_down()
        -> port.c: sci_port_remove_phy()
          -> sci_port_clear_phy()
            -> phy.c: sci_phy_set_port()
              -> port.c: sci_port_broadcast_change_received()
                -> isci_port_bc_change_received()

[1] Call chains for entering state: *SCI_PHY_STOPPED*
-----------------------------------------------------

host.c: isci_host_init()                                        (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_initialize(), atomic                        (*)
      -> phy.c: sci_phy_initialize()
        -> phy.c: sci_phy_link_layer_initialization()
          -> phy.c: sci_change_state(SCI_PHY_STOPPED)

init.c: PCI ->remove() || PM_OPS ->suspend,  process context    (+)
  -> host.c: isci_host_deinit()
    -> sci_controller_stop_phys()
      -> phy.c: sci_phy_stop()
-> sci_change_state(SCI_PHY_STOPPED)

phy.c: isci_phy_control()
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_phy_stop(), atomic                                     (*)
    -> sci_change_state(SCI_PHY_STOPPED)

[2] Call chains for entering state: *SCI_PHY_STARTING*
------------------------------------------------------

phy.c: phy_sata_timeout(), atimer, timer callback               (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_change_state(SCI_PHY_STARTING)

host.c: phy_startup_timeout(), atomic, timer callback           (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_controller_start_next_phy()
    -> sci_phy_start()
      -> sci_change_state(SCI_PHY_STARTING)

host.c: isci_host_start()                                       (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_start(), atomic                             (*)
    -> sci_controller_start_next_phy()
      -> sci_phy_start()
        -> sci_change_state(SCI_PHY_STARTING)

phy.c: Enter SCI state *SCI_PHY_SUB_FINAL*                      # Cont. from [2A]
  -> sci_change_state(SCI_PHY_SUB_FINAL)
    -> sci_phy_starting_final_substate_enter()
      -> sci_change_state(SCI_PHY_READY)
        -> Enter SCI state: *SCI_PHY_READY*
          -> sci_phy_ready_state_enter()
            -> host.c: sci_controller_link_up()
              -> sci_controller_start_next_phy()
                -> sci_phy_start()
                  -> sci_change_state(SCI_PHY_STARTING)

phy.c: sci_phy_event_handler(), atomic, discussed earlier       (*)
  -> sci_change_state(SCI_PHY_STARTING), 11 instances

port.c: isci_port_perform_hard_reset()
spin_lock_irqsave(isci_host::scic_lock, )
  -> port.c: sci_port_hard_reset(), atomic                      (*)
    -> phy.c: sci_phy_reset()
      -> sci_change_state(SCI_PHY_RESETTING)
        -> enter SCI PHY state: *SCI_PHY_RESETTING*
          -> sci_phy_resetting_state_enter()
            -> sci_change_state(SCI_PHY_STARTING)

[2A] Call chains for entering SCI state: *SCI_PHY_SUB_FINAL*
------------------------------------------------------------

host.c: power_control_timeout(), atomic, timer callback         (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> phy.c: sci_phy_consume_power_handler()
    -> phy.c: sci_change_state(SCI_PHY_SUB_FINAL)

host.c: sci_controller_error_handler(): atomic, irq handler     (*)
OR host.c: sci_controller_completion_handler(), atomic, tasklet (*)
  -> sci_controller_process_completions()
    -> sci_controller_unsolicited_frame()
      -> phy.c: sci_phy_frame_handler()
        -> sci_change_state(SCI_PHY_SUB_AWAIT_SAS_POWER)
          -> sci_phy_starting_await_sas_power_substate_enter()
            -> host.c: sci_controller_power_control_queue_insert()
              -> phy.c: sci_phy_consume_power_handler()
                -> sci_change_state(SCI_PHY_SUB_FINAL)
        -> sci_change_state(SCI_PHY_SUB_FINAL)
    -> sci_controller_event_completion()
      -> phy.c: sci_phy_event_handler()
        -> sci_phy_start_sata_link_training()
          -> sci_change_state(SCI_PHY_SUB_AWAIT_SATA_POWER)
            -> sci_phy_starting_await_sata_power_substate_enter
              -> host.c: sci_controller_power_control_queue_insert()
                -> phy.c: sci_phy_consume_power_handler()
                  -> sci_change_state(SCI_PHY_SUB_FINAL)

As can be seen from the "(*)" markers above, almost all the call-chains are
atomic. The only exception, marked with "(+)", is a PCI ->remove() and
PM_OPS ->suspend() cold path. Thus, pass GFP_ATOMIC to the libsas port
event notifier.

Note, the now-replaced libsas APIs used in_interrupt() to implicitly decide
which memory allocation type to use.  This was only partially correct, as
it fails to choose the correct GFP flags when just preemption or interrupts
are disabled. Such buggy code paths are marked with "(@)" in the call
chains above.

Link: https://lore.kernel.org/r/20210118100955.1761652-8-a.darwish@linutronix.de
Fixes: 1c393b970e0f ("scsi: libsas: Use dynamic alloced work to avoid sas event lost")
Cc: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: isci: Pass gfp_t flags in isci_port_link_up()

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

libsas sas_notify_port_event() is called from isci_port_link_up().  Below
is the context analysis for all of its call chains:

host.c: isci_host_init()                                        (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_initialize(), atomic                        (*)
    -> port_config.c: sci_port_configuration_agent_initialize()
      -> sci_mpc_agent_validate_phy_configuration()
        -> port.c: sci_port_add_phy()
          -> sci_port_general_link_up_handler()
            -> sci_port_activate_phy()
              -> isci_port_link_up()

port_config.c: apc_agent_timeout(), atomic, timer callback      (*)
  -> sci_apc_agent_configure_ports()
    -> port.c: sci_port_add_phy()
      -> sci_port_general_link_up_handler()
        -> sci_port_activate_phy()
          -> isci_port_link_up()

phy.c: enter SCI state: *SCI_PHY_SUB_FINAL*                     # Cont. from [1]
  -> phy.c: sci_phy_starting_final_substate_enter()
    -> phy.c: sci_change_state(SCI_PHY_READY)
      -> enter SCI state: *SCI_PHY_READY*
        -> phy.c: sci_phy_ready_state_enter()
          -> host.c: sci_controller_link_up()
            -> .link_up_handler()
            == port_config.c: sci_apc_agent_link_up()
              -> port.c: sci_port_link_up()
                -> (continue at [A])
            == port_config.c: sci_mpc_agent_link_up()
      -> port.c: sci_port_link_up()
                -> (continue at [A])

port_config.c: mpc_agent_timeout(), atomic, timer callback      (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> ->link_up_handler()
  == port_config.c: sci_apc_agent_link_up()
    -> port.c: sci_port_link_up()
      -> (continue at [A])
  == port_config.c: sci_mpc_agent_link_up()
    -> port.c: sci_port_link_up()
      -> (continue at [A])

[A] port.c: sci_port_link_up()
  -> sci_port_activate_phy()
    -> isci_port_link_up()
  -> sci_port_general_link_up_handler()
    -> sci_port_activate_phy()
      -> isci_port_link_up()

[1] Call chains for entering SCI state: *SCI_PHY_SUB_FINAL*
-----------------------------------------------------------

host.c: power_control_timeout(), atomic, timer callback         (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> phy.c: sci_phy_consume_power_handler()
    -> phy.c: sci_change_state(SCI_PHY_SUB_FINAL)

host.c: sci_controller_error_handler(): atomic, irq handler     (*)
OR host.c: sci_controller_completion_handler(), atomic, tasklet (*)
  -> sci_controller_process_completions()
    -> sci_controller_unsolicited_frame()
      -> phy.c: sci_phy_frame_handler()
        -> sci_change_state(SCI_PHY_SUB_AWAIT_SAS_POWER)
          -> sci_phy_starting_await_sas_power_substate_enter()
            -> host.c: sci_controller_power_control_queue_insert()
              -> phy.c: sci_phy_consume_power_handler()
                -> sci_change_state(SCI_PHY_SUB_FINAL)
        -> sci_change_state(SCI_PHY_SUB_FINAL)
    -> sci_controller_event_completion()
      -> phy.c: sci_phy_event_handler()
        -> sci_phy_start_sata_link_training()
          -> sci_change_state(SCI_PHY_SUB_AWAIT_SATA_POWER)
            -> sci_phy_starting_await_sata_power_substate_enter
              -> host.c: sci_controller_power_control_queue_insert()
                -> phy.c: sci_phy_consume_power_handler()
                  -> sci_change_state(SCI_PHY_SUB_FINAL)

As can be seen from the "(*)" markers above, all the call-chains are
atomic.  Pass GFP_ATOMIC to libsas port event notifier.

Note, the now-replaced libsas APIs used in_interrupt() to implicitly decide
which memory allocation type to use.  This was only partially correct, as
it fails to choose the correct GFP flags when just preemption or interrupts
are disabled. Such buggy code paths are marked with "(@)" in the call
chains above.

Link: https://lore.kernel.org/r/20210118100955.1761652-7-a.darwish@linutronix.de
Fixes: 1c393b970e0f ("scsi: libsas: Use dynamic alloced work to avoid sas event lost")
Cc: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: isci: Pass gfp_t flags in isci_port_link_down()

Use the new libsas event notifiers API, which requires callers to
explicitly pass the gfp_t memory allocation flags.

sas_notify_phy_event() is exclusively called by isci_port_link_down().
Below is the context analysis for all of its call chains:

port.c: port_timeout(), atomic, timer callback                  (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> port_state_machine_change(..., SCI_PORT_FAILED)
    -> enter SCI port state: *SCI_PORT_FAILED*
      -> sci_port_failed_state_enter()
        -> isci_port_hard_reset_complete()
          -> isci_port_link_down()

port.c: isci_port_perform_hard_reset()
spin_lock_irqsave(isci_host::scic_lock, )
  -> port.c: sci_port_hard_reset(), atomic                      (*)
    -> phy.c: sci_phy_reset()
      -> sci_change_state(SCI_PHY_RESETTING)
        -> enter SCI PHY state: *SCI_PHY_RESETTING*
          -> sci_phy_resetting_state_enter()
            -> port.c: sci_port_deactivate_phy()
      -> isci_port_link_down()

port.c: enter SCI port state: *SCI_PORT_READY*                  # Cont. from [1]
  -> sci_port_ready_state_enter()
    -> isci_port_hard_reset_complete()
      -> isci_port_link_down()

phy.c: enter SCI state: *SCI_PHY_STOPPED*                       # Cont. from [2]
  -> sci_phy_stopped_state_enter()
    -> host.c: sci_controller_link_down()
      -> ->link_down_handler()
      == port_config.c: sci_apc_agent_link_down()
        -> port.c: sci_port_remove_phy()
          -> sci_port_deactivate_phy()
            -> isci_port_link_down()
      == port_config.c: sci_mpc_agent_link_down()
        -> port.c: sci_port_link_down()
          -> sci_port_deactivate_phy()
            -> isci_port_link_down()

phy.c: enter SCI state: *SCI_PHY_STARTING*                      # Cont. from [3]
  -> sci_phy_starting_state_enter()
    -> host.c: sci_controller_link_down()
      -> ->link_down_handler()
      == port_config.c: sci_apc_agent_link_down()
        -> port.c: sci_port_remove_phy()
          -> isci_port_link_down()
      == port_config.c: sci_mpc_agent_link_down()
        -> port.c: sci_port_link_down()
          -> sci_port_deactivate_phy()
            -> isci_port_link_down()

[1] Call chains for 'enter SCI port state: *SCI_PORT_READY*'
------------------------------------------------------------

host.c: isci_host_init()                                        (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_initialize(), atomic                        (*)
    -> port_config.c: sci_port_configuration_agent_initialize()
      -> sci_mpc_agent_validate_phy_configuration()
        -> port.c: sci_port_add_phy()
          -> sci_port_general_link_up_handler()
            -> port_state_machine_change(, SCI_PORT_READY)
              -> enter port state *SCI_PORT_READY*

host.c: isci_host_start()                                       (@)
spin_lock_irq(isci_host::scic_lock)
  -> host.c: sci_controller_start(), atomic                     (*)
    -> host.c: sci_port_start()
      -> port.c: port_state_machine_change(, SCI_PORT_READY)
        -> enter port state *SCI_PORT_READY*

port_config.c: apc_agent_timeout(), atomic, timer callback      (*)
  -> sci_apc_agent_configure_ports()
    -> port.c: sci_port_add_phy()
      -> sci_port_general_link_up_handler()
        -> port_state_machine_change(, SCI_PORT_READY)
          -> enter port state *SCI_PORT_READY*

port_config.c: mpc_agent_timeout(), atomic, timer callback      (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> ->link_up_handler()
  == port.c: sci_apc_agent_link_up()
    -> sci_port_general_link_up_handler()
      -> port_state_machine_change(, SCI_PORT_READY)
        -> enter port state *SCI_PORT_READY*
  == port.c: sci_mpc_agent_link_up()
    -> port.c: sci_port_link_up()
      -> sci_port_general_link_up_handler()
        -> port_state_machine_change(, SCI_PORT_READY)
          -> enter port state *SCI_PORT_READY*

phy.c: enter SCI state: SCI_PHY_SUB_FINAL                       # Cont. from [1A]
  -> sci_phy_starting_final_substate_enter()
    -> sci_change_state(SCI_PHY_READY)
      -> enter SCI state: *SCI_PHY_READY*
        -> sci_phy_ready_state_enter()
          -> host.c: sci_controller_link_up()
            -> port_agent.link_up_handler()
            == port_config.c: sci_apc_agent_link_up()
              -> port.c: sci_port_link_up()
                -> sci_port_general_link_up_handler()
                  -> port_state_machine_change(, SCI_PORT_READY)
                    -> enter port state *SCI_PORT_READY*
            == port_config.c: sci_mpc_agent_link_up()
              -> port.c: sci_port_link_up()
                -> sci_port_general_link_up_handler()
                  -> port_state_machine_change(, SCI_PORT_READY)
                    -> enter port state *SCI_PORT_READY*

[1A] Call chains for entering SCI state: *SCI_PHY_SUB_FINAL*
------------------------------------------------------------

host.c: power_control_timeout(), atomic, timer callback         (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> phy.c: sci_phy_consume_power_handler()
    -> phy.c: sci_change_state(SCI_PHY_SUB_FINAL)

host.c: sci_controller_error_handler(): atomic, irq handler     (*)
OR host.c: sci_controller_completion_handler(), atomic, tasklet (*)
  -> sci_controller_process_completions()
    -> sci_controller_unsolicited_frame()
      -> phy.c: sci_phy_frame_handler()
        -> sci_change_state(SCI_PHY_SUB_AWAIT_SAS_POWER)
          -> sci_phy_starting_await_sas_power_substate_enter()
            -> host.c: sci_controller_power_control_queue_insert()
              -> phy.c: sci_phy_consume_power_handler()
                -> sci_change_state(SCI_PHY_SUB_FINAL)
        -> sci_change_state(SCI_PHY_SUB_FINAL)
    -> sci_controller_event_completion()
      -> phy.c: sci_phy_event_handler()
        -> sci_phy_start_sata_link_training()
          -> sci_change_state(SCI_PHY_SUB_AWAIT_SATA_POWER)
            -> sci_phy_starting_await_sata_power_substate_enter
              -> host.c: sci_controller_power_control_queue_insert()
                -> phy.c: sci_phy_consume_power_handler()
                  -> sci_change_state(SCI_PHY_SUB_FINAL)

[2] Call chains for entering state: *SCI_PHY_STOPPED*
-----------------------------------------------------

host.c: isci_host_init()                                        (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_initialize(), atomic                        (*)
      -> phy.c: sci_phy_initialize()
        -> phy.c: sci_phy_link_layer_initialization()
          -> phy.c: sci_change_state(SCI_PHY_STOPPED)

init.c: PCI ->remove() || PM_OPS ->suspend,  process context    (+)
  -> host.c: isci_host_deinit()
    -> sci_controller_stop_phys()
      -> phy.c: sci_phy_stop()
-> sci_change_state(SCI_PHY_STOPPED)

phy.c: isci_phy_control()
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_phy_stop(), atomic                                     (*)
    -> sci_change_state(SCI_PHY_STOPPED)

[3] Call chains for entering state: *SCI_PHY_STARTING*
------------------------------------------------------

phy.c: phy_sata_timeout(), atimer, timer callback               (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_change_state(SCI_PHY_STARTING)

host.c: phy_startup_timeout(), atomic, timer callback           (*)
spin_lock_irqsave(isci_host::scic_lock, )
  -> sci_controller_start_next_phy()
    -> sci_phy_start()
      -> sci_change_state(SCI_PHY_STARTING)

host.c: isci_host_start()                                       (@)
spin_lock_irq(isci_host::scic_lock)
  -> sci_controller_start(), atomic                             (*)
    -> sci_controller_start_next_phy()
      -> sci_phy_start()
        -> sci_change_state(SCI_PHY_STARTING)

phy.c: Enter SCI state *SCI_PHY_SUB_FINAL*, atomic, check above (*)
  -> sci_change_state(SCI_PHY_SUB_FINAL)
    -> sci_phy_starting_final_substate_enter()
      -> sci_change_state(SCI_PHY_READY)
        -> Enter SCI state: *SCI_PHY_READY*
          -> sci_phy_ready_state_enter()
            -> host.c: sci_controller_link_up()
              -> sci_controller_start_next_phy()
                -> sci_phy_start()
                  -> sci_change_state(SCI_PHY_STARTING)

phy.c: sci_phy_event_handler(), atomic, discussed earlier       (*)
  -> sci_change_state(SCI_PHY_STARTING), 11 instances

phy.c: enter SCI state: *SCI_PHY_RESETTING*, atomic, discussed  (*)
  -> sci_phy_resetting_state_enter()
    -> sci_change_state(SCI_PHY_STARTING)

As can be seen from the "(*)" markers above, almost all the call-chains are
atomic. The only exception, marked with "(+)", is a PCI ->remove() and
PM_OPS ->suspend() cold path. Thus, pass GFP_ATOMIC to the libsas phy event
notifier.

Note, The now-replaced libsas APIs used in_interrupt() to implicitly decide
which memory allocation type to use.  This was only partially correct, as
it fails to choose the correct GFP flags when just preemption or interrupts
are disabled. Such buggy code paths are marked with "(@)" in the call
chains above.

Link: https://lore.kernel.org/r/20210118100955.1761652-6-a.darwish@linutronix.de
Fixes: 1c393b970e0f ("scsi: libsas: Use dynamic alloced work to avoid sas event lost")
Cc: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mvsas: Pass gfp_t flags to libsas event notifiers

mvsas calls the non _gfp version of the libsas event notifiers API, leading
to the buggy call chains below:

  mvsas/mv_sas.c: mvs_work_queue() [process context]
  spin_lock_irqsave(mvs_info::lock, )
    -> libsas/sas_event.c: sas_notify_phy_event()
      -> sas_alloc_event()
        -> in_interrupt() = false
          -> invalid GFP_KERNEL allocation
    -> libsas/sas_event.c: sas_notify_port_event()
      -> sas_alloc_event()
        -> in_interrupt() = false
          -> invalid GFP_KERNEL allocation

Use the new event notifiers API instead, which requires callers to
explicitly pass the gfp_t memory allocation flags.

Below are context analysis for the modified functions:

=> mvs_bytes_dmaed():

Since it is invoked from both process and atomic contexts, let its callers
pass the gfp_t flags. Call chains:

  scsi_scan.c: do_scsi_scan_host() [has msleep()]
    -> shost->hostt->scan_start()
    -> [mvsas/mv_init.c: Scsi_Host::scsi_host_template .scan_start = mvs_scan_start()]
    -> mvsas/mv_sas.c: mvs_scan_start()
      -> mvs_bytes_dmaed(..., GFP_KERNEL)

  mvsas/mv_sas.c: mvs_work_queue()
  spin_lock_irqsave(mvs_info::lock,)
    -> mvs_bytes_dmaed(..., GFP_ATOMIC)

  mvsas/mv_64xx.c: mvs_64xx_isr() || mvsas/mv_94xx.c: mvs_94xx_isr()
    -> mvsas/mv_chips.h: mvs_int_full()
      -> mvsas/mv_sas.c: mvs_int_port()
        -> mvs_bytes_dmaed(..., GFP_ATOMIC);

=> mvs_work_queue():

Invoked from process context, but it calls all the libsas event notifier
APIs under a spin_lock_irqsave(). Pass GFP_ATOMIC.

Link: https://lore.kernel.org/r/20210118100955.1761652-5-a.darwish@linutronix.de
Fixes: 1c393b970e0f ("scsi: libsas: Use dynamic alloced work to avoid sas event lost")
Cc: Jason Yan <yanaijie@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: libsas: Introduce a _gfp() variant of event notifiers

sas_alloc_event() uses in_interrupt() to decide which allocation should be
used.

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

The in_interrupt() check is also only partially correct, because it fails
to choose the correct code path when just preemption or interrupts are
disabled. For example, as in the following call chain:

  mvsas/mv_sas.c: mvs_work_queue() [process context]
  spin_lock_irqsave(mvs_info::lock, )
    -> libsas/sas_event.c: sas_notify_phy_event()
      -> sas_alloc_event()
        -> in_interrupt() = false
          -> invalid GFP_KERNEL allocation
    -> libsas/sas_event.c: sas_notify_port_event()
      -> sas_alloc_event()
        -> in_interrupt() = false
          -> invalid GFP_KERNEL allocation

Introduce sas_alloc_event_gfp(), sas_notify_port_event_gfp(), and
sas_notify_phy_event_gfp(), which all behave like the non _gfp() variants
but use a caller-passed GFP mask for allocations.

For bisectability, all callers will be modified first to pass GFP context,
then the non _gfp() libsas API variants will be modified to take a gfp_t by
default.

Link: https://lore.kernel.org/r/20210118100955.1761652-4-a.darwish@linutronix.de
Fixes: 1c393b970e0f ("scsi: libsas: Use dynamic alloced work to avoid sas event lost")
Cc: Jason Yan <yanaijie@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: libsas: Remove notifier indirection

LLDDs report events to libsas with .notify_port_event and .notify_phy_event
callbacks.

These callbacks are fixed and so there is no reason why the functions
cannot be called directly, so do that.

This neatens the code slightly, makes it more obvious, and reduces function
pointer usage, which is generally a good thing. Downside is that there are
2x more symbol exports.

[a.darwish@linutronix.de: Remove the now unused "sas_ha" local variables]

Link: https://lore.kernel.org/r/20210118100955.1761652-3-a.darwish@linutronix.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: libsas: docs: Remove notify_ha_event()

The ->notify_ha_event() hook has long been removed from the libsas event
interface.

Remove it from documentation.

Link: https://lore.kernel.org/r/20210118100955.1761652-2-a.darwish@linutronix.de
Fixes: 042ebd293b86 ("scsi: libsas: kill useless ha_event and do some cleanup")
Cc: stable@vger.kernel.org
Reviewed-by: John Garry <john.garry@huawei.com>
Reviewed-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ncr53c8xx: Use SAM status values

Use SAM status values instead of the driver-defined ones.  This also fixes
a potential bug as the driver-defined values declare 'COMMAND TERMINATED'
with a value of 0x20, whereas SCSI-II defines it with a value of 0x22.

Link: https://lore.kernel.org/r/20210113090500.129644-36-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: advansys: Kill driver-defined status byte accessors

Replace the driver-defined status byte accessors with the mid-layer defined
ones.

Link: https://lore.kernel.org/r/20210113090500.129644-35-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla2xxx: fc_remote_port_chkready() returns a SCSI result value

fc_remote_port_chkready() returns a SCSI result value, not the port
status. Fix the value returned when the remote port isn't set.

Link: https://lore.kernel.org/r/20210113090500.129644-34-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: storvsc: Return DID_ERROR for invalid commands

ILLEGAL_COMMAND is a sense code, not a driver byte.

Link: https://lore.kernel.org/r/20210113090500.129644-33-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ips: Use correct command completion on error

A non-zero queuecommand() return code means 'busy', i.e. the command hasn't
been submitted. So any command which should be failed need to be completed
via the ->scsi_done() callback with the appropriate result code set.

Link: https://lore.kernel.org/r/20210113090500.129644-32-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: wd33c93: Use SCSI status

Use standard SCSI status and drop usage of the linux-specific ones.

Link: https://lore.kernel.org/r/20210113090500.129644-31-hare@suse.de
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: esp_scsi: Do not set SCSI message byte

The message byte setting always devolves to COMMAND_COMPLETE so we can drop
setting the message byte in the SCSI result.

Link: https://lore.kernel.org/r/20210113090500.129644-30-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: esp_scsi: Use host byte as last argument to esp_cmd_is_done()

Just pass in the host byte to esp_cmd_is_done() and set the status or
message bytes if the host byte is DID_OK.

Link: https://lore.kernel.org/r/20210113090500.129644-29-hare@suse.de
Acked-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: core: Add 'set_status_byte()' accessor

Add the missing 'set_status_byte()' accessor function.

Link: https://lore.kernel.org/r/20210113090500.129644-28-hare@suse.de
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: dpt_i2o: Use DID_ERROR instead of INITIATOR_ERROR message

Change the error code for an invalid SCSI opcode to DID_ERROR.
INITIATOR_ERROR is a scsi parallel message which doesn't apply for RAID
HBAs.

Link: https://lore.kernel.org/r/20210113090500.129644-27-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mac53c94: Do not set invalid command result

CMD_ACCEPT_MSG is an internal definition and most certainly not a SCSI
status. As the latter gets set during command completion we can drop the
assignment here.

Link: https://lore.kernel.org/r/20210113090500.129644-26-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: atp870u: Use standard definitions

Use standard definitions for SCSI commands and return status instead of the
hardcoded values.

Link: https://lore.kernel.org/r/20210113090500.129644-25-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufshcd: Do not set COMMAND_COMPLETE

COMMAND_COMPLETE is defined as '0', so setting it is quite pointless.

Link: https://lore.kernel.org/r/20210113090500.129644-24-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: scsi_debug: Do not set COMMAND_COMPLETE

COMMAND_COMPLETE is defined as '0', so setting it is quite pointless.

Link: https://lore.kernel.org/r/20210113090500.129644-23-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: initio: Drop internal SCSI message definition

Use the standard SCSI message definitions instead of the driver-internal
ones.

Link: https://lore.kernel.org/r/20210113090500.129644-22-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: dc395x: Drop internal SCSI message definitions

Drop the internal SCSI message definitions and use the functions provided
by the SPI transport class.

Link: https://lore.kernel.org/r/20210113090500.129644-21-hare@suse.de
Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic7xxx: aic79xx: Drop internal SCSI message definition

Use the standard SCSI message definitions instead of the driver-internal
ones.

Link: https://lore.kernel.org/r/20210113090500.129644-20-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: nsp_cs: Drop internal SCSI message definition

Use the standard SCSI message definitions instead of the driver-internal
ones.

Link: https://lore.kernel.org/r/20210113090500.129644-19-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: stex: Do not set COMMAND_COMPLETE

COMMAND_COMPLETE is defined as '0', so setting it is quite pointless.

Link: https://lore.kernel.org/r/20210113090500.129644-18-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hpsa: Do not set COMMAND_COMPLETE

COMMAND_COMPLETE is defined as '0', and it is a SCSI parallel message to
boot. Drop the call to set_msg_byte().

Link: https://lore.kernel.org/r/20210113090500.129644-17-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aacraid: Avoid setting message byte on completion

The aacraid controller is a RAID controller and the driver will never see
any SCSI messages. Plus it's quite pointless to set the message byte if the
host byte is already set, as the latter takes precedence during error
recovery.  Drop the message byte values for the final result.

Link: https://lore.kernel.org/r/20210113090500.129644-16-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: zfcp: Do not set COMMAND_COMPLETE

COMMAND_COMPLETE is defined as '0', and it is a SCSI parallel message to
boot. So drop the call to set_msg_byte().

Link: https://lore.kernel.org/r/20210113090500.129644-15-hare@suse.de
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla4xxx: Use standard SAM status definitions

Use standard SAM status definitions and drop the driver-defined ones.

Link: https://lore.kernel.org/r/20210113090500.129644-14-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: dc395: Drop private SAM status code definitions

We don't need to duplicate definitions from the common include files.

Link: https://lore.kernel.org/r/20210113090500.129644-13-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: nsp32: Fixup status handling

SCp.status is always the SAM-defined status value, not the Linux
ones. Fixup the one wrong definition.

Link: https://lore.kernel.org/r/20210113090500.129644-12-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: acornscsi: Use standard defines

Use midlayer-defined values and drop the non-existing QUEUE_FULL case; we
are checking the SCSI messages in the switch statement, and QUEUE_FULL is a
SCSI status hence it can never occur here.

Link: https://lore.kernel.org/r/20210113090500.129644-11-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: bfa: Drop driver-defined SCSI status codes

Drop the driver-defined SCSI status codes and use the generic ones instead.

Link: https://lore.kernel.org/r/20210113090500.129644-10-hare@suse.de
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic7xxx: aic79xx: Remove driver-defined SAM status definitions

Replace the driver-defined SAM status definitions with the standard
mid-layer defined ones.

Link: https://lore.kernel.org/r/20210113090500.129644-9-hare@suse.de
Reviewed-by: Bart van Assche <bvanassche@acm.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic7xxx: aic79xx: Kill pointless forward declarations

Link: https://lore.kernel.org/r/20210113090500.129644-8-hare@suse.de
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic7xxx: aic79xx: Whitespace cleanup

Link: https://lore.kernel.org/r/20210113090500.129644-7-hare@suse.de
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: atp870u: Whitespace cleanup

Link: https://lore.kernel.org/r/20210113090500.129644-6-hare@suse.de
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-sas: Whitespace cleanup

Link: https://lore.kernel.org/r/20210113090500.129644-5-hare@suse.de
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-9xxx: Whitespace cleanup

Link: https://lore.kernel.org/r/20210113090500.129644-4-hare@suse.de
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-xxxx: Whitespace cleanup

Link: https://lore.kernel.org/r/20210113090500.129644-3-hare@suse.de
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: Drop gdth driver

The gdth driver refers to a SCSI parallel, PCI-only HBA RAID adapter which
was manufactured by the now-defunct ICP Vortex company, later acquired by
Adaptec and superseded by the aacraid series of controllers.  The driver
itself would require a major overhaul before any modifications can be
attempted, but seeing that it's unlikely to have any users left it should
rather be removed completely.

Link: https://lore.kernel.org/r/20210113090500.129644-2-hare@suse.de
Cautiously-Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: core: Remove in_interrupt() check in transport_handle_cdb_direct()

transport_handle_cdb_direct() uses in_interrupt() to detect if it is safe
to sleep. It produces a stack trace and returns with an error which is
clearly for debugging.

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

transport_handle_cdb_direct() has a comment saying that it may only be
invoked from process context. It invokes transport_generic_new_cmd() which
performs GFP_KERNEL memory allocations. in_interrupt() does not detect all
the contexts where it is invalid to sleep (for the blocking GFP_KERNEL
allocation) as it fails to detect sections with disabled preemption.

Replace the in_interrupt() based check with a might_sleep() annotation.

Link: https://lore.kernel.org/r/20201220203638.43615-7-bigeasy@linutronix.de
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: core: Replace in_interrupt() usage in target_submit_cmd_map_sgls()

target_submit_cmd_map_sgls() uses in_interrupt() to crash if it returns
true.

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

The usage of in_interrupt() is clearly for debugging. might_sleep() is
better at this because it also detects other contexts in which it is not
allowed to sleep, like preempt-disabled section.

Replace BUG_ON(in_interrupt) with might_sleep().

Link: https://lore.kernel.org/r/20201220203638.43615-6-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: alua: Remove in_interrupt() usage in core_alua_check_nonop_delay()

core_alua_check_nonop_delay() uses in_interrupt() to decide if it is safe
to sleep.

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

core_alua_check_nonop_delay() has two callers:

 - target_submit_cmd_map_sgls()
   Kernel doc says it that it must be called from process context. Also has
   a BUG_ON(in_interrupt()).

 - iscsit_setup_scsi_cmd()
   Invokes iscsit_add_reject_cmd() which does GFP_KERNEL allocation and
   target_cmd_init_cdb() which may do GFP_KERNEL allocations.

Remove the in_interrupt() check because all callers are from preemptible
context.

Link: https://lore.kernel.org/r/20201220203638.43615-5-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: iscsi: Redo iscsit_check_session_usage_count() return code

The return value of iscsit_check_session_usage_count() is only checked if
it was not allowed to sleep. If it returns `2' then a timer is prepared. If
it returns something else or if it was allowed to sleep then it is ignored.

Let iscsit_check_session_usage_count() return true if it needs to arm the
timer - otherwise false. This simplifies the code flow of the only caller.

Link: https://lore.kernel.org/r/20201220203638.43615-4-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: iscsi: Avoid in_interrupt() usage in iscsit_check_session_usage_count()

iscsit_check_session_usage_count() uses in_interrupt() to find out if it is
safe to invoke wait_for_completion().

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

There is only one caller of iscsit_check_session_usage_count() which
already has an argument indicating if it is safe to sleep.

Extend iscsit_check_session_usage_count() by an argument indicating if it
may sleep.

Link: https://lore.kernel.org/r/20201220203638.43615-3-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: target: iscsi: Avoid in_interrupt() usage in iscsit_close_session()

iscsit_close_session() uses in_interrupt() to decide if it needs to check
the return value of iscsit_check_session_usage_count() if it was not able
to sleep.

The usage of in_interrupt() in drivers is phased out and Linus clearly
requested that code which changes behaviour depending on context should
either be separated or the context be conveyed in an argument passed by the
caller, which usually knows the context.

iscsit_close_session() has two callers:

 - iscsit_handle_time2retain_timeout()
   A timer_list callback.

 - iscsit_close_connection()
   Runs in preemptible context, acquires a mutex.

Add an argument to iscsit_close_session() indicating if sleeping is
possible.

Link: https://lore.kernel.org/r/20201220203638.43615-2-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Clean up and refactor clk-scaling feature

Manipulate clock scaling related stuff only if the host capability supports
clock scaling feature to avoid redundant code execution.

Link: https://lore.kernel.org/r/20210120150142.5049-4-stanley.chu@mediatek.com
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Remove redundant null checking of devfreq instance

hba->devfreq is zero-initialized thus it is not required to check its
existence in ufshcd_add_lus() function which is invoked during
initialization only.

Link: https://lore.kernel.org/r/20210120150142.5049-3-stanley.chu@mediatek.com
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Refactor cancelling clkscaling works

Cancelling suspend_work and resume_work is only required while suspending
clk-scaling. Move these two invocations into ufshcd_suspend_clkscaling()
function.

Link: https://lore.kernel.org/r/20210120150142.5049-2-stanley.chu@mediatek.com
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

Revert "Make sure clk scaling happens only when HBA is runtime ACTIVE"

Commit 73cc291c2702 ("scsi: ufs: Make sure clk scaling happens only
when HBA is runtime ACTIVE") is no longer needed since commit
0e9d4ca43ba8 ("scsi: ufs: Protect some contexts from unexpected clock
scaling") is a more mature fix to protect UFS LLD stability from clock
scaling invoked through sysfs nodes by users.

Link: https://lore.kernel.org/r/1611137065-14266-4-git-send-email-cang@codeaurora.org
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Refactor ufshcd_init/exit_clk_scaling/gating()

ufshcd_hba_exit() is always called after ufshcd_exit_clk_scaling() and
ufshcd_exit_clk_gating(). Move ufshcd_exit_clk_scaling/gating() to
ufshcd_hba_exit(). Meanwhile, add dedicated functions to initialize
and remove sysfs nodes of clock scaling/gating to make the code more
readable. Overall functionality remains same.

Link: https://lore.kernel.org/r/1611137065-14266-3-git-send-email-cang@codeaurora.org
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Protect some contexts from unexpected clock scaling

In contexts like suspend, shutdown, and error handling we need to
suspend devfreq to make sure these contexts won't be disturbed by
clock scaling.  However, suspending devfreq is not enough since users
can still trigger a clock scaling by manipulating the devfreq sysfs
nodes like min/max_freq and governor even after devfreq is
suspended. Moreover, mere suspending devfreq cannot synchroinze a
clock scaling which has already been invoked through these sysfs
nodes. Add one more flag in struct clk_scaling and wrap the entire
func ufshcd_devfreq_scale() with the clk_scaling_lock, so that we can
use this flag and clk_scaling_lock to control and synchronize clock
scaling invoked through devfreq sysfs nodes.

Link: https://lore.kernel.org/r/1611137065-14266-2-git-send-email-cang@codeaurora.org
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Group UFS WB related flags in struct ufs_dev_info

UFS device-related flags should be grouped in ufs_dev_info. Move wb_enabled
and wb_buf_flush_enabled out from struct ufs_hba, group them in struct
ufs_dev_info, and align the names of the structure members vertically.

Link: https://lore.kernel.org/r/20210119163847.20165-6-huobean@gmail.com
Reviewed-by: Can Guo <cang@codeaurora.org>
Acked-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Remove two WB related fields from struct ufs_dev_info

d_wb_alloc_units and d_ext_ufs_feature_sup are only used during WB probe.
They are used to confirm the condition that "if bWriteBoosterBufferType
is set to 01h but dNumSharedWriteBoosterBufferAllocUnits is set to zero,
the WriteBooster feature is disabled", and if UFS device supports WB.

No need to keep them after probing is complete.

Link: https://lore.kernel.org/r/20210119163847.20165-5-huobean@gmail.com
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Update comment in the function ufshcd_wb_probe()

USFHCD supports both WriteBooster "LU dedicated buffer" mode and "shared
buffer" mode. Update the comment accordingly in the function
ufshcd_wb_probe().

Link: https://lore.kernel.org/r/20210119163847.20165-4-huobean@gmail.com
Reviewed-by: Can Guo <cang@codeaurora.org>
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: docs: ABI: Add wb_on documentation for new entry wb_on

Adds UFS sysfs documentation for new entry wb_on.

[mkp: fix doc formatting]

Link: https://lore.kernel.org/r/20210119163847.20165-3-huobean@gmail.com
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
fix format

scsi: ufs: Add "wb_on" sysfs node to control WB on/off

Currently UFS WriteBooster driver uses clock scaling up/down to set WB
on/off. For the platforms which don't support UFSHCD_CAP_CLK_SCALING, WB
will be always on. Provide a sysfs attribute to enable/disable WB during
runtime. Write 1/0 to "wb_on" sysfs node to enable/disable UFS WB.

Link: https://lore.kernel.org/r/20210119163847.20165-2-huobean@gmail.com
Reviewed-by: Avri Altman <avri.altman@wdc.com>
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-exynos: Use UFSHCD_QUIRK_ALIGN_SG_WITH_PAGE_SIZE

Exynos needs scatterlist entries aligned to page size because it isn't
capable of transferring data contained in one DATA IN operation to seversal
areas in memory.

Link: https://lore.kernel.org/r/80d7e27d6ec537e650a6bd74897b6c60618efcdc.1611026909.git.kwmad.kim@samsung.com
Signed-off-by: Kiwoong Kim <kwmad.kim@samsung.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Introduce a quirk to allow only page-aligned sg entries

Some SoCs require a single scatterlist entry for smaller than page size,
i.e. 4KB. When dispatching commands with more than one scatterlist entry
under 4KB in size the following behavior is observed:

A command to read a block range is dispatched with two scatterlist entries
that are named AAA and BBB. After dispatching, the host builds two PRDT
entries and during transmission, device sends just one DATA IN because
device doesn't care about host DMA. The host then transfers the combined
amount of data from start address of the area named AAA. As a consequence,
the area that follows AAA in memory would be corrupted.

    |<------------->|
    +-------+------------         +-------+
    +  AAA  + (corrupted)   ...   +  BBB  +
    +-------+------------         +-------+

To avoid this we need to enforce page size alignment for sg entries.

Link: https://lore.kernel.org/r/56dddef94f60bd9466fd77e69f64bbbd657ed2a1.1611026909.git.kwmad.kim@samsung.com
Signed-off-by: Kiwoong Kim <kwmad.kim@samsung.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Delete redundant if statement in ufshcd_intr()

Once going into while-do loop, intr_status is already true, this
if-statement is redundant, remove it.

Link: https://lore.kernel.org/r/20210118201233.3043-1-huobean@gmail.com
Reviewed-by: Avri Altman <avri.altman@wdc.com>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Fix spelling mistake "succeded" -> "succeeded"

There is a spelling mistake in a ibmvfc_dbg debug message. Fix it.

Link: https://lore.kernel.org/r/20210118111346.70798-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm80xx: Switch from 'pci_' to 'dma_' API

The wrappers in include/linux/pci-dma-compat.h should go away.

The patch has been generated with the coccinelle script below and has been
hand modified to replace GFP_ with a correct flag.  It has been compile
tested.

When memory is allocated in 'pm8001_init_ccb_tag()' GFP_KERNEL can be used
because this function already uses this flag a few lines above.

While at it, remove "pm80xx: " in a debug message. 'pm8001_dbg()' already
adds the driver name in the message.

@@
@@
-    PCI_DMA_BIDIRECTIONAL
+    DMA_BIDIRECTIONAL

@@
@@
-    PCI_DMA_TODEVICE
+    DMA_TO_DEVICE

@@
@@
-    PCI_DMA_FROMDEVICE
+    DMA_FROM_DEVICE

@@
@@
-    PCI_DMA_NONE
+    DMA_NONE

@@
expression e1, e2, e3;
@@
-    pci_alloc_consistent(e1, e2, e3)
+    dma_alloc_coherent(&e1->dev, e2, e3, GFP_)

@@
expression e1, e2, e3;
@@
-    pci_zalloc_consistent(e1, e2, e3)
+    dma_alloc_coherent(&e1->dev, e2, e3, GFP_)

@@
expression e1, e2, e3, e4;
@@
-    pci_free_consistent(e1, e2, e3, e4)
+    dma_free_coherent(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_map_single(e1, e2, e3, e4)
+    dma_map_single(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_unmap_single(e1, e2, e3, e4)
+    dma_unmap_single(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4, e5;
@@
-    pci_map_page(e1, e2, e3, e4, e5)
+    dma_map_page(&e1->dev, e2, e3, e4, e5)

@@
expression e1, e2, e3, e4;
@@
-    pci_unmap_page(e1, e2, e3, e4)
+    dma_unmap_page(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_map_sg(e1, e2, e3, e4)
+    dma_map_sg(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_unmap_sg(e1, e2, e3, e4)
+    dma_unmap_sg(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_dma_sync_single_for_cpu(e1, e2, e3, e4)
+    dma_sync_single_for_cpu(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_dma_sync_single_for_device(e1, e2, e3, e4)
+    dma_sync_single_for_device(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_dma_sync_sg_for_cpu(e1, e2, e3, e4)
+    dma_sync_sg_for_cpu(&e1->dev, e2, e3, e4)

@@
expression e1, e2, e3, e4;
@@
-    pci_dma_sync_sg_for_device(e1, e2, e3, e4)
+    dma_sync_sg_for_device(&e1->dev, e2, e3, e4)

@@
expression e1, e2;
@@
-    pci_dma_mapping_error(e1, e2)
+    dma_mapping_error(&e1->dev, e2)

@@
expression e1, e2;
@@
-    pci_set_dma_mask(e1, e2)
+    dma_set_mask(&e1->dev, e2)

@@
expression e1, e2;
@@
-    pci_set_consistent_dma_mask(e1, e2)
+    dma_set_coherent_mask(&e1->dev, e2)

Link: https://lore.kernel.org/r/20210117132445.562552-1-christophe.jaillet@wanadoo.fr
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm80xx: Clean up indentation of a code block

A block of code is indented one level too deeply, clean this up.

Link: https://lore.kernel.org/r/20210115095824.9170-1-colin.king@canonical.com
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Addresses-Coverity: ("Indentation does not match nesting level")

Merge branch '5.11/scsi-fixes' into 5.12/scsi-queue

Pull in the 5.11 SCSI fixes branch to provide an updated baseline for
megaraid and hisi_sas. Both drivers received core changes in
v5.11-rc3.

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

scsi: lpfc: Add support for eh_should_retry_cmd()

Add support for eh_should_retry_cmd callback in lpfc_template.

Link: https://lore.kernel.org/r/1609969748-17684-6-git-send-email-muneendra.kumar@broadcom.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: scsi_transport_fc: Add store capability to rport port_state in sysfs

Add store capability to the rport port_state using sysfs under
fc_remote_ports/rport-*/port_state.

With this the user can move the port_state from Marginal->Online and
Online->Marginal.

 - Marginal: This interface will set SCMD_NORETRIES_ABORT bit in
   scmd->state for all the pending I/Os on the SCSI device associated with
   target port.

 - Online: This interface will clear SCMD_NORETRIES_ABORT bit in
   scmd->state for all the pending I/Os on the SCSI device associated with
   target port.

The following interface is provided to set the port state to Marginal and
Online respectively:

echo "Marginal" >> /sys/class/fc_remote_ports/rport-X\:Y-Z/port_state
echo "Online" >> /sys/class/fc_remote_ports/rport-X\:Y-Z/port_state

Link: https://lore.kernel.org/r/1609969748-17684-5-git-send-email-muneendra.kumar@broadcom.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: scsi_transport_fc: Add a new rport state FC_PORTSTATE_MARGINAL

Add a new interface, fc_eh_should_retry_cmd(), which checks if the cmd
should be retried or not by checking the rport state. If the rport state is
marginal it returns false to make sure there won't be any retries on the
cmd.

Make the fc_remote_port_delete(), fc_user_scan_tgt(), and
fc_timeout_deleted_rport() functions handle the new rport state.

Link: https://lore.kernel.org/r/1609969748-17684-4-git-send-email-muneendra.kumar@broadcom.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: core: No retries on abort success

Add a new optional routine, eh_should_retry_cmd(), in scsi_host_template
that allows the transport to decide if a cmd is retryable. Return true if
the transport is in a state the cmd should be retried on.

Update scmd_eh_abort_handler() and scsi_eh_flush_done_q() to both call
scsi_eh_should_retry_cmd() to check whether the command needs to be
retried.

The above changes were based on a patch by Mike Christie.

Link: https://lore.kernel.org/r/1609969748-17684-3-git-send-email-muneendra.kumar@broadcom.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: core: Add a new error code DID_TRANSPORT_MARGINAL in scsi.h

Add code in scsi_result_to_blk_status to translate a new error
DID_TRANSPORT_MARGINAL to the corresponding blk_status_t i.e
BLK_STS_TRANSPORT.

Add DID_TRANSPORT_MARGINAL case to scsi_decide_disposition().

Link: https://lore.kernel.org/r/1609969748-17684-2-git-send-email-muneendra.kumar@broadcom.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Provide modules parameters for MQ settings

Add the various module parameter toggles for adjusting the MQ
characteristics at boot/load time as well as a device attribute for
changing the client scsi channel request amount.

Link: https://lore.kernel.org/r/20210114203148.246656-22-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Enable MQ and set reasonable defaults

Turn on MQ by default and set sane values for the upper limit on hw queues
for the SCSI host, and number of hw SCSI channels to request from the
partner VIOS.

Link: https://lore.kernel.org/r/20210114203148.246656-21-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Purge SCSI channels after transport loss/reset

Grab the queue and list lock for each Sub-CRQ and add any uncompleted
events to the host purge list.

Link: https://lore.kernel.org/r/20210114203148.246656-20-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Send Cancel MAD down each hw SCSI channel

In general the client needs to send Cancel MADs and task management
commands down the same channel as the command(s) intended to cancel or
abort. The client assigns cancel keys per LUN and thus must send a Cancel
down each channel commands were submitted for that LUN. Further, the client
then must wait for those cancel completions prior to submitting a LUN RESET
or ABORT TASK SET.

Add a cancel rsp iu syncronization field to the ibmvfc_queue struct such
that the cancel routine can sync the cancel response to each queue that
requires a cancel command. Build a list of each cancel event sent and wait
for the completion of each submitted cancel.

Link: https://lore.kernel.org/r/20210114203148.246656-19-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add cancel mad initialization helper

Add a helper routine for initializing a Cancel MAD. This will be useful for
a channelized client that needs to send Cancel commands down every channel
commands were sent for a particular LUN.

Link: https://lore.kernel.org/r/20210114203148.246656-18-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Register Sub-CRQ handles with VIOS during channel setup

If the ibmvfc client adapter requests channels it must submit a number of
Sub-CRQ handles matching the number of channels being requested. The VIOS
in its response will overwrite the actual number of channel resources
allocated which may be less than what was requested. The client then must
store the VIOS Sub-CRQ handle for each queue. This VIOS handle is needed as
a parameter with h_send_sub_crq().

Link: https://lore.kernel.org/r/20210114203148.246656-17-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Send commands down HW Sub-CRQ when channelized

When the client has negotiated the use of channels all vfcFrames are
required to go down a Sub-CRQ channel or it is a protocoal violation. If
the adapter state is channelized submit vfcFrames to the appropriate
Sub-CRQ via the h_send_sub_crq() helper.

Link: https://lore.kernel.org/r/20210114203148.246656-16-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Set and track hw queue in ibmvfc_event struct

Extract the hwq id from a SCSI command and store it in the ibmvfc_event
structure to identify which Sub-CRQ to send the command down when channels
are being utilized.

Link: https://lore.kernel.org/r/20210114203148.246656-15-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Advertise client support for using hardware channels

Previous patches have plumbed the necessary Sub-CRQ interface and channel
negotiation MADs to fully channelize via hardware backed queues.

Advertise client support via NPIV Login capability IBMVFC_CAN_USE_CHANNELS
when the client bits have MQ enabled via vhost->mq_enabled, or when
channels were already in use during a subsequent NPIV Login. The later is
required because channel support is only renegotiated after a CRQ pair is
broken. Simple NPIV Logout/Logins require the client to continue to
advertise the channel capability until the CRQ pair between the client is
broken.

Link: https://lore.kernel.org/r/20210114203148.246656-14-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Implement channel enquiry and setup commands

New NPIV_ENQUIRY_CHANNEL and NPIV_SETUP_CHANNEL management datagrams (MADs)
were defined in a previous patchset. If the client advertises a desire to
use channels and the partner VIOS is channel capable then the client must
proceed with channel enquiry to determine the maximum number of channels
the VIOS is capable of providing, and registering SubCRQs via channel setup
with the VIOS immediately following NPIV Login. This handshaking should not
be performed for subsequent NPIV Logins unless the CRQ connection has been
reset.

Implement these two new MADs and issue them following a successful NPIV
login where the VIOS has set the SUPPORT_CHANNELS capability bit in the
NPIV Login response.

Link: https://lore.kernel.org/r/20210114203148.246656-13-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Map/request irq and register Sub-CRQ interrupt handler

Create an irq mapping for the hw_irq number provided from phyp firmware.
Request an irq assigned our Sub-CRQ interrupt handler. Unmap these irqs at
Sub-CRQ teardown.

Link: https://lore.kernel.org/r/20210114203148.246656-12-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Define Sub-CRQ interrupt handler routine

Simple handler that calls Sub-CRQ drain routine directly.

Link: https://lore.kernel.org/r/20210114203148.246656-11-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add handlers to drain and complete Sub-CRQ responses

The logic for iterating over the Sub-CRQ responses is similiar to that of
the primary CRQ. Add the necessary handlers for processing those responses.

Link: https://lore.kernel.org/r/20210114203148.246656-10-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add Sub-CRQ IRQ enable/disable routine

Each Sub-CRQ has its own interrupt. A hypercall is required to toggle the
IRQ state. Provide the necessary mechanism via a helper function.

Link: https://lore.kernel.org/r/20210114203148.246656-9-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add alloc/dealloc routines for SCSI Sub-CRQ Channels

Allocate a set of Sub-CRQs in advance. During channel setup the client and
VIOS negotiate the number of queues the VIOS supports and the number that
the client desires to request. Its possible that the final channel
resources allocated is less than requested, but the client is still
responsible for sending handles for every queue it is hoping for.

Also, provide deallocation cleanup routines.

Link: https://lore.kernel.org/r/20210114203148.246656-8-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add Subordinate CRQ definitions

Subordinate Command Response Queues (Sub CRQ) are used in conjunction with
the primary CRQ when more than one queue is needed by the virtual I/O
adapter. Recent phyp firmware versions support Sub CRQ's with ibmvfc
adapters. This feature is a prerequisite for supporting multiple hardware
backed submission queues in the vfc adapter.

The Sub CRQ command element differs from the standard CRQ in that it is
32bytes long as opposed to 16bytes for the latter. Despite this extra
16bytes the ibmvfc protocol will use the original CRQ command element
mapped to the first 16bytes of the Sub CRQ element initially.

Add definitions for the Sub CRQ command element and queue.

Link: https://lore.kernel.org/r/20210114203148.246656-7-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Define hcall wrapper for registering a Sub-CRQ

Sub-CRQs are registred with firmware via a hypercall. Abstract that
interface into a simpler helper function.

Link: https://lore.kernel.org/r/20210114203148.246656-6-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Add size parameter to ibmvfc_init_event_pool()

With the upcoming addition of Sub-CRQs the event pool size may vary
per-queue.

Add a size parameter to ibmvfc_init_event_pool() such that different size
event pools can be requested by ibmvfc_alloc_queue().

Link: https://lore.kernel.org/r/20210114203148.246656-5-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Init/free event pool during queue allocation/free

The event pool and CRQ used to be separate entities of the adapter host
structure and as such were allocated and freed independently of each
other. Recent work as defined a generic queue structure with an event pool
specific to each queue. As such the event pool for each queue shouldn't be
allocated/freed independently, but instead performed as part of the queue
allocation/free routines.

Move the calls to ibmvfc_event_pool_{init|free} into
ibmvfc_{alloc|free}_queue respectively. The only functional change here is
that the CRQ cannot be released in ibmvfc_remove until after the event pool
has been successfully purged since releasing the queue will also free the
event pool.

Link: https://lore.kernel.org/r/20210114203148.246656-4-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Move event pool init/free routines

The next patch in this series reworks the event pool allocation calls to
happen within the individual queue allocation routines instead of as
independent calls.

Move the init/free routines earlier in ibmvfc.c to prevent undefined
reference errors when calling these functions from the queue allocation
code. No functional change.

Link: https://lore.kernel.org/r/20210114203148.246656-3-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>