scsi: qla2xxx: Remove in_interrupt() from qla82xx-specific code

qla82xx_idc_lock() spins on a certain hardware state until it's updated. At
the end of each spin, if in_interrupt() is true, it does 20 loops of
cpu_relax(). Otherwise, it yields the CPU.

While in_interrupt() is ill-defined and does not provide what the name
suggests, it is not needed here: qla82xx_idc_lock() is always called from
process context. Below is an analysis of its callers, in order of
appearance:

  - qla_nx.c: qla82xx_device_bootstrap(), only called by
    qla82xx_device_state_handler(), has multiple msleep()s.

  - qla_nx.c: qla82xx_need_qsnt_handler(), has one second msleep()

  - qla_nx.c: qla82xx_wait_for_state_change(), one second msleep()

  - qla_nx.c: qla82xx_need_reset_handler(), can sleep up to 10 seconds

  - qla_nx.c: qla82xx_device_state_handler(), has multiple msleep()s

  - qla_nx.c: qla82xx_abort_isp(), if it's a qla82xx controller, calls
    qla82xx_device_state_handler(), which sleeps. It's also bound to
    isp_operations ->abort_isp() hook, where all the callers are in process
    context.

  - qla_nx.c: qla82xx_beacon_on(), bound to isp_operations ->beacon_on()
    hook.  That hook is only called once, in a mutex locked context, from
    qla2x00_beacon_store().

  - qla_nx.c: qla82xx_beacon_off(), bound to isp_operations ->beacon_off()
    hook.  Like ->beacon_on(), it's only called once, in a mutex locked
    context, from qla2x00_beacon_store().

  - qla_nx.c: qla82xx_fw_dump(), calls qla2x00_wait_for_chip_reset(), which
    has msleep() in a loop. It is bound to isp_operations ->fw_dump()
    hook. That hook *is* called from atomic context at qla_isr.c by
    multiple interrupt handlers. Nonetheless, it's other controllers
    interrupt handlers, and not the qla82xx.

  - qla82xx_msix_default() and qla82xx_msix_rsp_q() call
    qla24xx_process_response_queue() which doesn't implement the firmware
    dumping.

  - qla_attr.c: qla2x00_sysfs_write_fw_dump(), and
    qla2x00_sysfs_write_reset(), process-context sysfs ->write() hooks.

  - qla_os.c: qla2x00_probe_one(). PCI ->probe(), process context.

  - qla_os.c: qla2x00_clear_drv_active(), called solely from
    qla2x00_remove_one(), which is PCI ->remove() hook, process context.

  - qla_os.c: qla2x00_do_dpc(), kthread function, process context.

Remove the in_interrupt() check. Change qla82xx_idc_lock() specification to
a purely process-context function. Mark it with "Context: task, might
sleep".

Change qla82xx_idc_lock() implementation to sleep 100ms, instead of a
schedule(), for each spin. This is more deterministic, and it matches the
other qla models idc_lock() functions.

Link: https://lore.kernel.org/r/20201126132952.2287996-5-bigeasy@linutronix.de
Cc: Nilesh Javali <njavali@marvell.com>
Cc: <GR-QLogic-Storage-Upstream@marvell.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Daniel Wagner <dwagner@suse.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: qla4xxx: Remove in_interrupt()

qla4_82xx_crb_win_lock() spins on a certain hardware state until it's
updated. At the end of each spin, if in_interrupt() is true, it does 20
loops of cpu_relax(). Otherwise, it yields the CPU.

The in_interrupt() macro is ill-defined as it does not provide what the
name suggests, and it does not catch the intended use-case here.

qla4_82xx_crb_win_lock() is always invoked with scsi_qla_host::hw_lock
acquired, with disabled interrupts. If the caller is in process context, as
in qla4_82xx_need_reset_handler(), then in_interrupt() will return false
even though it is not allowed to call schedule().

Remove the in_interrupt() check.

Change qla4_82xx_crb_win_lock() specification to a purely atomic
function. Mark it as static, remove its forward declaration, and move it
above its callers. To avoid hammering the PCI bus while spinning, use a 10
micro-second delay instead of cpu_relax().

Link: https://lore.kernel.org/r/20201126132952.2287996-4-bigeasy@linutronix.de
Fixes: f4f5df23bf72 ("[SCSI] qla4xxx: Added support for ISP82XX")
Cc: Nilesh Javali <njavali@marvell.com>
Cc: Manish Rangankar <mrangankar@marvell.com>
Cc: <GR-QLogic-Storage-Upstream@marvell.com>
Reviewed-by: Daniel Wagner <dwagner@suse.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: hisi_sas: Remove preemptible()

hisi_sas_task_exec() uses preemptible() to see if it's safe to block.  This
does not work for CONFIG_PREEMPT_COUNT=n kernels in which preemptible()
always returns 0.

The problem is masked when enabling some of the common Kconfig.debug
options (like CONFIG_DEBUG_ATOMIC_SLEEP), as they implicitly enable the
preemption counter.

In general, driver leaf functions should not make logic decisions based on
the context they're called from. The caller should be the entity
responsible for explicitly indicating context.

Since hisi_sas_task_exec() already has a gfp_t flags parameter, use it as
the explicit context marker.

Link: https://lore.kernel.org/r/20201126132952.2287996-3-bigeasy@linutronix.de
Fixes: 214e702d4b70 ("scsi: hisi_sas: Adjust task reject period during host reset")
Fixes: 550c0d89d52d ("scsi: hisi_sas: Replace in_softirq() check in hisi_sas_task_exec()")
Cc: Xiaofei Tan <tanxiaofei@huawei.com>
Cc: Xiang Chen <chenxiang66@hisilicon.com>
Cc: John Garry <john.garry@huawei.com>
Acked-by: John Garry <john.garry@huawei.com>
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: pm80xx: Do not sleep in atomic context

hw_event_sas_phy_up() is used in hardirq/softirq context:

 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()
             => msleep(200)

Revert the msleep() back to an mdelay() to avoid sleeping in atomic
context.

Link: https://lore.kernel.org/r/20201126132952.2287996-2-bigeasy@linutronix.de
Fixes: 4daf1ef3c681 ("scsi: pm80xx: Convert 'long' mdelay to msleep")
Cc: Vikram Auradkar <auradkar@google.com>
Cc: Jack Wang <jinpu.wang@cloud.ionos.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
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: ufs: Remove unnecessary if condition in ufshcd_suspend()

In the case that auto_bkops_enable is false, which means auto bkops has
been disabled, there is no need to call ufshcd_disable_auto_bkops().

Link: https://lore.kernel.org/r/20201125185300.3394-1-huobean@gmail.com
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas: Move debugfs code to v3 hw driver

Relocate all the debugfs code for DFX to v3 hw since no other versions
support it.

Link: https://lore.kernel.org/r/1606207594-196362-4-git-send-email-john.garry@huawei.com
Signed-off-by: Luo Jiaxing <luojiaxing@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas: Fix up probe error handling for v3 hw

Fix some rollbacks in function hisi_sas_v3_probe() and
interrupt_init_v3_hw().

Link: https://lore.kernel.org/r/1606207594-196362-3-git-send-email-john.garry@huawei.com
Fixes: 8d98416a55eb ("scsi: hisi_sas: Switch v3 hw to MQ")
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas: Reduce some indirection in v3 hw driver

Sometimes local functions are called indirectly from the hw driver, which
only makes the code harder to follow. Remove these.

Method .hw_init is only called from platform driver probe, which is not
relevant, so don't set this either.

Link: https://lore.kernel.org/r/1606207594-196362-2-git-send-email-john.garry@huawei.com
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm8001: Remove space in a debug message

There are two words that need separating with a space in a pm8001_dbg()
message. Fix it.

Link: https://lore.kernel.org/r/20201124093828.307709-1-colin.king@canonical.com
Reviewed-by: Ewan D. Milne <emilne@redhat.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>

scsi: pm8001: Fix misindentation

kernel robot reported a misindentation of a goto.

Fix it.

At the same time, use a temporary for a repeated entry in the same block to
reduce visual noise.

Link: https://lore.kernel.org/r/9542a8be9954c1dca744f93f53bb1af6dd1436e8.1606192458.git.joe@perches.com
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm8001: Convert pm8001_printk() to pm8001_info()

Use the more common logging style.

[mkp: fixed a few conflicts]

Link: https://lore.kernel.org/r/69dc34ff63adfa60b3f203ed2d58143b5692af57.1606192458.git.joe@perches.com
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pmcraid: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-30-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pmcraid: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in pmcraid_resume(), and
there is no corresponding pci_enable_wake(...., true) in pmcraid_suspend().
Either it should do enable-wake the device in .suspend() or should not
invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
pmcraid_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-29-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mvumi: Update function description

There is no "device" parameter in mvumi_shutdown(). Instead there is "pdev"
which is not described.

Link: https://lore.kernel.org/r/20201102164730.324035-28-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mvumi: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-27-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mvumi: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in mvumi_resume(), and there
is no corresponding pci_enable_wake(...., true) in mvumi_suspend().  Either
it should do enable-wake the device in .suspend() or should not invoke
pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
mvumi_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-26-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-sas: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-25-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-sas: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in twl_resume(), and there is
no corresponding pci_enable_wake(...., true) in twl_suspend().  Either it
should do enable-wake the device in .suspend() or should not invoke
pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
twl_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-24-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-9xxx: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-23-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: 3w-9xxx: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in twa_resume(), and there is
no corresponding pci_enable_wake(...., true) in twa_suspend().  Either it
should do enable-wake the device in .suspend() or should not invoke
pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
twa_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-22-vaibhavgupta40@gmail.com
Acked-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hpsa: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-21-vaibhavgupta40@gmail.com
Acked-by: Don Brace <don.brace@microchip.com>
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm_8001: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-20-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm_8001: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in pm8001_pci_resume(), and
there is no corresponding pci_enable_wake(...., true) in
pm8001_pci_suspend(). Either it should do enable-wake the device in
.suspend() or should not invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
pm8001_pci__resume().

Link: https://lore.kernel.org/r/20201102164730.324035-19-vaibhavgupta40@gmail.com
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-18-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mpt3sas_scsih: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-17-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: mpt3sas_scsih: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in scsih_resume(), and there
is no corresponding pci_enable_wake(...., true) in scsih_suspend().  Either
it should do enable-wake the device in .suspend() or should not invoke
pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
scsih_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-16-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas_v3_hw: Remove extra function calls for runtime pm

Both runtime_suspend_v3_hw() and runtime_resume_v3_hw() do nothing else but
invoke suspend_v3_hw() and resume_v3_hw() respectively. This is the case of
unnecessary function calls. To use those functions for runtime pm as well,
simply use UNIVERSAL_DEV_PM_OPS.

make -j$(nproc) W=1, with CONFIG_PM disabled, throws '-Wunused-function'
warning for runtime_suspend_v3_hw() and runtime_resume_v3_hw(). After
dropping those function definitions, the warning was thrown for
suspend_v3_hw() and resume_v3_hw(). Hence, mark them as '__maybe_unused'.

Link: https://lore.kernel.org/r/20201102164730.324035-15-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas_v3_hw: Don't use PCI helper functions

Drivers using new-framework/generic-framework should not handle standard
power management operations. These operations were performed by legacy
framework through PCI helper functions like pci_save/restore_state(),
pci_set_power_state(), etc.

Drivers should not use them now.

Link: https://lore.kernel.org/r/20201102164730.324035-14-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: hisi_sas_v3_hw: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in hisi_sas_v3_resume(), and
there is no corresponding pci_enable_wake(...., true) in
hisi_sas_v3_suspend(). Either it should do enable-wake the device in
.suspend() or should not invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
hisi_sas_v3_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-13-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: esas2r: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-12-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: esas2r: Drop PCI Wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in esas2r_resume(), and there
is no corresponding pci_enable_wake(...., true) in esas2r_suspend().
Either it should do enable-wake the device in .suspend() or should not
invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
esas2r_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-11-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: arcmsr: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-10-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: arcmsr: Drop PCI wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in arcmsr_resume(), and there
is no corresponding pci_enable_wake(...., true) in arcmsr_suspend().
Either it should do enable-wake the device in .suspend() or should not
invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
arcmsr_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-9-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic79xx: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-8-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aic7xxx: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-7-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aacraid: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-6-vaibhavgupta40@gmail.com
Acked-by: Balsundar P <balsundar.p@microchip.com>
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: aacraid: Drop pci_enable_wake() from .resume

The driver calls pci_enable_wake(...., false) in aac_resume(), and there is
no corresponding pci_enable_wake(...., true) in aac_suspend(). Either it
should do enable-wake the device in .suspend() or should not invoke
pci_enable_wake() at all.

Concluding that this is a bug and PCI core calls pci_enable_wake(pci_dev,
PCI_D0, false) during resume, drop it from aac_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-5-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: megaraid_sas: Update function description

Function parameter 'pdev 'is described as Generic Device Structure. It is a
PCI device structure.

Link: https://lore.kernel.org/r/20201102164730.324035-4-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: megaraid_sas: Use generic power management

Drivers should do only device-specific jobs. But in general, drivers using
legacy PCI PM framework for .suspend()/.resume() have to manage many PCI
PM-related tasks themselves which can be done by PCI Core itself. This
brings extra load on the driver and it directly calls PCI helper functions
to handle them.

Switch to the new generic framework by updating function signatures and
define a "struct dev_pm_ops" variable to bind PM callbacks. Also, remove
unnecessary calls to the PCI Helper functions along with the legacy
.suspend & .resume bindings.

Link: https://lore.kernel.org/r/20201102164730.324035-3-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: megaraid_sas: Drop PCI wakeup calls from .resume

The driver calls pci_enable_wake(...., false) in megasas_resume(), and
there is no corresponding pci_enable_wake(...., true) in megasas_suspend().
Either it should do enable-wake the device in .suspend() or should not
invoke pci_enable_wake() at all.

Concluding that this driver doesn't support enable-wake and PCI core calls
pci_enable_wake(pci_dev, PCI_D0, false) during resume, drop it from
megasas_resume().

Link: https://lore.kernel.org/r/20201102164730.324035-2-vaibhavgupta40@gmail.com
Signed-off-by: Vaibhav Gupta <vaibhavgupta40@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: Validate io_req before others

We need to check for a valid io_req before we check other data. Also,
remove redundant checks.

Link: https://lore.kernel.org/r/20201121023337.19295-1-kartilak@cisco.com
Reviewed-by: Arulprabhu Ponnusamy <arulponn@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: Set scsi_set_resid() only for underflow

Set scsi_set_resid() only if FCPIO_ICMND_CMPL_RESID_UNDER is set.

Link: https://lore.kernel.org/r/20201121015134.18872-1-kartilak@cisco.com
Reviewed-by: Arulprabhu Ponnusamy <arulponn@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: Change shost_printk() to FNIC_MAIN_DBG()

Replace shost_printk() with FNIC_MAIN_DBG() so that these log messages are
controlled by fnic_log_level flag in fnic_handle_link.

Link: https://lore.kernel.org/r/20201121013739.18701-1-kartilak@cisco.com
Reviewed-by: Arulprabhu Ponnusamy <arulponn@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: Avoid looping in TRANS ETH on unload

Avoid looping in fnic_scsi_abort_io() before sending fw reset when fnic is
in TRANS ETH state and when we have not received any link events.

Link: https://lore.kernel.org/r/20201121012145.18522-1-kartilak@cisco.com
Reviewed-by: Arulprabhu Ponnusamy <arulponn@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: fnic: Change shost_printk() to FNIC_FCS_DBG()

Replacing shost_printk with FNIC_FCS_DBG() so that these log messages are
controlled by fnic_log_level flag in fnic_fip_handler_timer.

Bumping up version number from 47 to 49 to maintain same level as internal
version.

Link: https://lore.kernel.org/r/20201120220712.16708-1-kartilak@cisco.com
Reviewed-by: Arulprabhu Ponnusamy <arulponn@cisco.com>
Co-developed-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Signed-off-by: Karan Tilak Kumar <kartilak@cisco.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm8001: Make implicit use of pm8001_ha in pm8001_printk() explicit

Make the pm8001_printk() macro take an explicit HBA instead of assuming the
existence of an unspecified pm8001_ha argument.

Miscellanea:

 - Add pm8001_ha to the few uses of pm8001_printk()

 - Add HBA to the pm8001_dbg macro call to pm8001_printk()

Link: https://lore.kernel.org/r/0e17a4c845f15e18f98b346ffb9b039584d21cdd.1605914030.git.joe@perches.com
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm8001: Neaten debug logging macros and uses

Every PM8001_<FOO>_DBG macro uses an internal call to pm8001_printk.

Convert all uses of:

PM8001_<FOO>_DBG(hba, pm8001_printk(fmt, ...))
to
pm8001_dbg(hba, <FOO>, fmt, ...)

so the visual complexity of each macro is reduced.

The repetitive macro definitions are converted to a single pm8001_dbg and
the level is concatenated using PM8001_##level##_LOGGING for the specific
level test.

Done with coccinelle, checkpatch and a little typing of the new macro
definition.

Miscellanea:

 - Coalesce formats

 - Realign arguments

 - Add missing terminating newlines to formats

 - Remove trailing spaces from formats

 - Change defective loop with printk(KERN_INFO... to emit a 16 byte hex
   block to %p16h

Link: https://lore.kernel.org/r/49f36a93af7752b613d03c89a87078243567fd9a.1605914030.git.joe@perches.com
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: pm8001: Remove typecast for pointer returned by kcalloc()

Stop typecasting the value returned by kcalloc().

Link: https://lore.kernel.org/r/20201120083648.9319-1-vulab@iscas.ac.cn
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Xu Wang <vulab@iscas.ac.cn>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: NCR5380: Reduce NCR5380_maybe_release_dma_irq() call sites

Refactor to avoid needless calls to NCR5380_maybe_release_dma_irq().
This makes the machine code smaller and the source more readable.

Link: https://lore.kernel.org/r/c1317ae8fdcb498460de5d7ea0bd62a42f5eeca8.1605847196.git.fthain@telegraphics.com.au
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: atari_scsi: Fix race condition between .queuecommand and EH

It is possible that bus_reset_cleanup() or .eh_abort_handler could be
invoked during NCR5380_queuecommand(). If that takes place before the new
command is enqueued and after the ST-DMA "lock" has been acquired, the
ST-DMA "lock" will be released again. This will result in a lost DMA
interrupt and a command timeout. Fix this by excluding EH and interrupt
handlers while the new command is enqueued.

Link: https://lore.kernel.org/r/af25163257796b50bb99d4ede4025cea55787b8f.1605847196.git.fthain@telegraphics.com.au
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Adjust logic in common ADAPT helper

The introduction of ufshcd_dme_configure_adapt() refactored out duplication
from the Mediatek and Qualcomm drivers.

Both these implementations had the logic of:

    gear_tx == UFS_HS_G4 => PA_INITIAL_ADAPT
    gear_tx != UFS_HS_G4 => PA_NO_ADAPT

but now both implementations pass PA_INITIAL_ADAPT as "adapt_val" and if
gear_tx is not UFS_HS_G4 that is replaced with PA_INITIAL_ADAPT. In other
words, it's PA_INITIAL_ADAPT in both above cases.

The result is that e.g. Qualcomm SM8150 has no longer functional UFS, so
adjust the logic to match the previous implementation.

Link: https://lore.kernel.org/r/20201121044810.507288-1-bjorn.andersson@linaro.org
Fixes: fc85a74e28fe ("scsi: ufs: Refactor ADAPT configuration function")
Reviewed-by: Can Guo <cang@codeaurora.org>
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix variable 'vport' set but not used in lpfc_sli4_abts_err_handler()

Remove vport variable that is assigned but not used in
lpfc_sli4_abts_err_handler().

Link: https://lore.kernel.org/r/20201119203407.121913-1-james.smart@broadcom.com
Fixes: e7dab164a9aa ("scsi: lpfc: Fix scheduling call while in softirq context in lpfc_unreg_rpi")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix missing prototype for lpfc_nvmet_prep_abort_wqe()

lpfc_nvmet_prep_abort_wqe() needs to be declared static.

Link: https://lore.kernel.org/r/20201119203316.121725-1-james.smart@broadcom.com
Fixes: db7531d2b377 ("scsi: lpfc: Convert abort handling to SLI-3 and SLI-4 handlers")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix set but unused variables in lpfc_dev_loss_tmo_handler()

Remove set but not used variable shost in lpfc_dev_loss_tmo_handler().

Link: https://lore.kernel.org/r/20201119203353.121866-1-james.smart@broadcom.com
Fixes: 52edb2caf675 ("scsi: lpfc: Remove ndlp when a PLOGI/ADISC/PRLI/REG_RPI ultimately fails")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix set but not used warnings from Rework remote port lock handling

Remove local variables that are set but not used.

Link: https://lore.kernel.org/r/20201119203340.121819-1-james.smart@broadcom.com
Fixes: c6adba150191 ("scsi: lpfc: Rework remote port lock handling")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix missing prototype warning for lpfc_fdmi_vendor_attr_mi()

Function needs to be declared as static.

Link: https://lore.kernel.org/r/20201119203328.121772-1-james.smart@broadcom.com
Fixes: 8aaa7bcf07a2 ("scsi: lpfc: Add FDMI Vendor MIB support")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix memory leak on lcb_context

Currently there is an error return path that neglects to free the
allocation for lcb_context.  Fix this by adding a new error free exit path
that kfree's lcb_context before returning.  Use this new kfree exit path in
another exit error path that also kfree's the same object, allowing a line
of code to be removed.

Link: https://lore.kernel.org/r/20201118141314.462471-1-colin.king@canonical.com
Fixes: 4430f7fd09ec ("scsi: lpfc: Rework locations of ndlp reference taking")
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Addresses-Coverity: ("Resource leak")

scsi: lpfc: Remove dead code on second !ndlp check

Currently there is a null check on the pointer ndlp that exits via error
path issue_ct_rsp_exit followed by another null check on the same pointer
that is almost identical to the previous null check stanza and yet can
never can be reached because the previous check exited via
issue_ct_rsp_exit. This is deadcode and can be removed.

Link: https://lore.kernel.org/r/20201118133744.461385-1-colin.king@canonical.com
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Addresses-Coverity: ("Logically dead code")

scsi: lpfc: Fix pointer defereference before it is null checked issue

There is a null check on pointer lpfc_cmd after the pointer has been
dereferenced when pointers rdata and ndlp are initialized at the start of
the function. Fix this by only assigning rdata and ndlp after the pointer
lpfc_cmd has been null checked.

Link: https://lore.kernel.org/r/20201118131345.460631-1-colin.king@canonical.com
Fixes: 96e209be6ecb ("scsi: lpfc: Convert SCSI I/O completions to SLI-3 and SLI-4 handlers")
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Addresses-Coverity: ("Dereference before null check")

scsi: ibmvfc: Advertise client support for targetWWPN using v2 commands

The previous patch added support for the targetWWPN field in version 2 MADs
and vfcFrame structures.

Set the IBMVFC_CAN_SEND_VF_WWPN bit in our capabailites flag during NPIV
Login to inform the VIOS that this client supports the feature.

Link: https://lore.kernel.org/r/20201118011104.296999-7-tyreld@linux.ibm.com
Acked-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 support for target_wwpn field in v2 MADs and vfcFrame

Several version 2 MADs and the version 2 vfcFrame structures introduced a
new targetWWPN field for better identification of a target over the
scsi_id.

Set this field and MAD versioning fields when the VIOS advertises the
IBMVFC_HANDLE_VF_WWPN capability.

Link: https://lore.kernel.org/r/20201118011104.296999-6-tyreld@linux.ibm.com
Acked-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 FC payload retrieval routines for versioned vfcFrames

The FC iu and response payloads are located at different offsets depending
on the ibmvfc_cmd version. This is a result of the version 2 vfcFrame
definition adding an extra 64bytes of reserved space to the structure prior
to the payloads.

Add helper routines to determine the current vfcFrame version and return a
pointer to the proper iu or response structure within that ibmvfc_cmd.

Link: https://lore.kernel.org/r/20201118011104.296999-5-tyreld@linux.ibm.com
Acked-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 helper for testing capability flags

Testing the NPIV Login response capabilities is a long winded process of
dereferencing the vhost->login_buf->resp.capabilities field, then byte
swapping that value to host endian, and performing the bitwise test.
Currently we only ever check this in ibmvfc_cancel_all(), but follow-up
patches will need to regularly check for targetWWPN and channelization
support.

Add a helper to simplify checking various VIOS capabilities, namely
ibmvfc_check_caps().

Link: https://lore.kernel.org/r/20201118011104.296999-4-tyreld@linux.ibm.com
Acked-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 new fields for version 2 of several MADs

Introduce a target_wwpn field to several MADs. Its possible that a SCSI ID
of a target can change due to some fabric changes. The WWPN of the SCSI
target provides a better way to identify the target. Also, add flags for
receiving MAD versioning information and advertising client support for
targetWWPN with the VIOS. This latter capability flag will be required for
future clients capable of requesting multiple hardware queues from the host
adapter.

Link: https://lore.kernel.org/r/20201118011104.296999-3-tyreld@linux.ibm.com
Acked-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: Deduplicate common ibmvfc_cmd init code

The virtual FC frame command exchanged with the VIOS is used for device
reset and command abort TMF as well as normally queued commands. When
initializing the ibmvfc_cmd there are several elements of the command that
are set the same way regardless of the command type.

Deduplicate code by moving these commonally set fields into a
initialization helper routine, namely ibmvfc_init_vfc_cmd().

Link: https://lore.kernel.org/r/20201118011104.296999-2-tyreld@linux.ibm.com
Acked-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: Use correlation token to tag commands

The vfcFrame correlation field is a 64bit handle that is intended to trace
I/O operations through both the client stack and VIOS stack when the
underlying physical FC adapter supports tagging.

Tag vfcFrames with the associated ibmvfc_event pointer handle.

Link: https://lore.kernel.org/r/20201117185031.129939-3-tyreld@linux.ibm.com
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Remove trailing semicolon

Remove a superfluous semicolon following a closing function block bracket.

Link: https://lore.kernel.org/r/20201117185031.129939-2-tyreld@linux.ibm.com
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ibmvfc: Byte swap login_buf.resp values in attribute show functions

Both ibmvfc_show_host_(capabilities|npiv_version) functions retrieve values
from vhost->login_buf.resp buffer. This is the MAD response buffer from the
VIOS and as such any multi-byte non-string values are in big endian format.

Byte swap these values to host CPU endian format for better human
readability.

Link: https://lore.kernel.org/r/20201117185031.129939-1-tyreld@linux.ibm.com
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Show LBA and length for UNMAP commands

We have LBA and length for unmap commands.

Link: https://lore.kernel.org/r/20201117165839.1643377-8-jaegeuk@kernel.org
Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Leo Liou <leoliou@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Fix clkgating on/off

The following call stack prevents clk_gating at every I/O completion.  We
can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work
will check it again.

ufshcd_complete_requests(struct ufs_hba *hba)
  ufshcd_transfer_req_compl()
    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();

Note that this still requires work to deal with a potential race condition
when user sets clkgating.delay_ms to very small value. That can cause
preventing clkgating by the check of ufshcd_any_tag_in_use() in gate_work.

Link: https://lore.kernel.org/r/20201117165839.1643377-7-jaegeuk@kernel.org
Fixes: 7252a3603015 ("scsi: ufs: Avoid busy-waiting by eliminating tag conflicts")
Reviewed-by: Asutosh Das <asutoshd@codeaurora.org>
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Add more contexts in the ufs tracepoints

This adds user-friendly tracepoints with group id.

Link: https://lore.kernel.org/r/20201117165839.1643377-6-jaegeuk@kernel.org
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@google.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Use WQ_HIGHPRI for gating work

Must have WQ_MEM_RECLAIM

``WQ_MEM_RECLAIM``

  All workqueues which might be used in the memory reclaim paths **MUST**
  have this flag set.  The wq is guaranteed to have at least one execution
  context regardless of memory pressure.

Link: https://lore.kernel.org/r/20201117165839.1643377-5-jaegeuk@kernel.org
Reviewed-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@google.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Clear UAC for FFU and RPMB LUNs

In order to conduct FFU or RPMB operations, UFS needs to clear UNIT
ATTENTION condition. Clear it explicitly so that we get no failures during
initialization.

Link: https://lore.kernel.org/r/20201117165839.1643377-4-jaegeuk@kernel.org
Signed-off-by: Jaegeuk Kim <jaegeuk@google.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Atomic update for clkgating_enable

While running a stress test which enables/disables clkgating, we
occasionally hit device timeout. This patch avoids a subtle race condition
to address it.

Link: https://lore.kernel.org/r/20201117165839.1643377-3-jaegeuk@kernel.org
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@google.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Avoid to call REQ_CLKS_OFF to CLKS_OFF

Once UFS is gated with CLKS_OFF, it should not call REQ_CLKS_OFF
again. This can lead to hibern8_enter failure.

Link: https://lore.kernel.org/r/20201117165839.1643377-2-jaegeuk@kernel.org
Reviewed-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: qla4xxx: Remove redundant assignment to variable rval

The variable rval has been initialized with 'QLA_ERROR'. The assignment is
redundant in an error path. Remove it.

Link: https://lore.kernel.org/r/20201103120137.109717-1-jingxiangfeng@huawei.com
Signed-off-by: Jing Xiangfeng <jingxiangfeng@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-qcom: Use common ADAPT configuration function

Use common ADAPT configuration function to reduce duplicated code in UFS
drivers.

Link: https://lore.kernel.org/r/20201116065054.7658-10-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-mediatek: Use common ADAPT configuration function

Use common ADAPT configuration function to reduce duplicated code in UFS
drivers.

Link: https://lore.kernel.org/r/20201116065054.7658-9-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Refactor ADAPT configuration function

Several vendors are using same code to configure ADAPT settings for
HS-G4. Simply refactor it as common function.

Link: https://lore.kernel.org/r/20201116065054.7658-8-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-hisi: Use device parameter initialization function

Use common device parameter initialization function instead of initializing
those parameters by vendor driver itself.

Link: https://lore.kernel.org/r/20201116065054.7658-7-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-exynos: Use device parameter initialization function

Use common device parameter initialization function instead of initializing
those parameters by vendor driver itself.

Link: https://lore.kernel.org/r/20201116065054.7658-6-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-qcom: Use device parameter initialization function

Use common device parameter initialization function instead of initializing
those parameters by vendor driver itself.

Link: https://lore.kernel.org/r/20201116065054.7658-5-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-mediatek: Use device parameter initialization function

Use common device parameter initialization function instead of initializing
those parameters by vendor driver itself.

Link: https://lore.kernel.org/r/20201116065054.7658-4-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: Introduce device parameter initialization function

Nowadays many vendors initialize their device parameters in their own
vendor drivers. The initialization code is almost the same as well as the
pre-defined definitions. Introduce a common device parameter initialization
function which assign the most common initial values. With this function,
we could remove those duplicated codes in vendor drivers.

Link: https://lore.kernel.org/r/20201116065054.7658-3-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: ufs: ufs-mediatek: Refactor performance scaling functions

Refactor performance scaling related functions in MediaTek UFS driver.

Link: https://lore.kernel.org/r/20201116065054.7658-2-stanley.chu@mediatek.com
Reviewed-by: Bean Huo <beanhuo@micron.com>
Signed-off-by: Stanley Chu <stanley.chu@mediatek.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Update changed file copyrights for 2020

Update Copyright in files changed by the 12.8.0.6 patch set to 2020

Link: https://lore.kernel.org/r/20201115192646.12977-18-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Update lpfc version to 12.8.0.6

Update lpfc version to 12.8.0.6

Link: https://lore.kernel.org/r/20201115192646.12977-17-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Convert abort handling to SLI-3 and SLI-4 handlers

This patch reworks the abort interfaces such that SLI-3 retains the
iocb-based formatting and completions and SLI-4 now uses native WQEs and
completion routines.

The following changes are made:

 - The code is refactored from a confusing 2 routine sequence of
   xx_abort_iotag_issue(), which creates/formats and abort cmd, and
   xx_issue_abort_tag(), which then issues and handles the completion of
   the abort cmd - into a single interface of xx_issue_abort_iotag().  The
   new interface will determine whether SLI-3 or SLI-4 and then call the
   appropriate handler. A completion handler can now be specified to
   address the differences in completion handling.  Note: original code is
   all iocb based, with SLI-4 converting to SLI-3 for the SCSI/ELS path,
   and NVMe natively using wqes.

 - The SLI-3 side is refactored:

   The older iocb-base lpfc_sli_issue_abort_iotag() routine is combined
   with the logic of lpfc_sli_abort_iotag_issue() as well as the
   iocb-specific code in lpfc_abort_handler() and lpfc_sli_abort_iocb() to
   create the new single SLI-3 abort routine that formats and issues the
   iocb.

 - The SLI-4 side is refactored and added to:

   The native WQE abort code in NVMe is moved to the new SLI-4
   issue_abort_iotag() routine. Items in SCSI that set fields not set by
   NVMe is migrated into the new routine. Thus the routine supports NVMe
   and SCSI initiators. The nvmet block (target) formats the abort slightly
   different (like the old NVMe initiator) thus it has its own prep routine
   stolen from NVMe initiator and it retains the current code it has for
   issuing the WQE (does not use the commonized routine the initiators
   do). SLI-4 completion handlers were also added.

 - lpfc_abort_handler now becomes a wrapper that determines whether
   SLI-3 or SLI-4 and calls the proper abort handler.

Link: https://lore.kernel.org/r/20201115192646.12977-16-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Convert SCSI I/O completions to SLI-3 and SLI-4 handlers

The current driver implementation uses SLI-4 WQE to iocb conversion before
calling the cmpl callback function.

Rework the FCP I/O completion path to utilize the SLI-4 WQE.

This patch converts the SCSI I/O completion paths from the iocb-centric
interfaces to the routines are native for whether I/Os are iocb-based
(SLI-3) or WQE-based (SLI-4).

Most existing routines were iocb-based, so this creates a lot of SLI-4
specific routines to provide the functionality.

Link: https://lore.kernel.org/r/20201115192646.12977-15-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Convert SCSI path to use common I/O submission path

This patch converts the SCSI I/O path from the iocb-centric interfaces to
the common I/O submission path which supports native SLI-4 WQEs.

A wrapper routine is put in place to distinguish SLI-3 from SLI. If SLI-3,
the same iocb-centric paths are used, perhaps with refactored code that is
explicitly for SLI-3.  For SLI-4, any iocb-related formatting is replaced
by wqe-based formatting, although much of that is addressed by the common
wqe templates in the SLI-4 path.

Link: https://lore.kernel.org/r/20201115192646.12977-14-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Enable common send_io interface for SCSI and NVMe

To set up common use by the SCSI and NVMe I/O paths, create a new routine
that issues FCP I/O commands which can be used by either protocol.  The new
routine addresses SLI-3 vs SLI-4 differences within its implementation.

Replace the (SLI-3 centric) iocb routine in the SCSI path with this new
WQE-centric common routine.

Link: https://lore.kernel.org/r/20201115192646.12977-13-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Enable common wqe_template support for both SCSI and NVMe

The driver is currently using SLI-4 WQE templates only for NVMe.  Refactor
the template and the placement of the service routine so that it can be
used by both SCSI and NVMe.

Link: https://lore.kernel.org/r/20201115192646.12977-12-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Refactor WQE structure definitions for common use

In preparation of reworking the driver to use a native SLI-4 WQE interface
for the SCSI and NVMe I/O paths, start by commonizing the WQE exchange type
and command type attributes.

While adjusting these options also noted the variance in the pbde field.
Fix this by setting templates to 0 and in NVMe, which explicitly uses this
option, setting the value.

Link: https://lore.kernel.org/r/20201115192646.12977-11-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix NPIV Fabric Node reference counting

While testing initiator-side cable swaps with NPIV, oops occur.  The
reference counts for the Fabric nodes on the NPIV vports isn't balanced,
resulting in premature node removal.

The following fixes were made:

 - Removed the FC_LBIT check in lpfc_linkup_port. This removed the special
   case for vports that didn't have them clean up just like the physical
   port.

 - Removed the unreg_rpi call in lpfc_cleanup_node. In this section, the
   node is being removed in the context of a reference count release and a
   mailbox command can't be issued at this point.

 - Remove special case handling in the default mailbox completion handler
   that allowed the skipping of a node reference. Now, reference counting
   always requires the removal of the reference.

 - Move the location of the DEVICE_RM event is done during LOGO handling as
   the driver has additional work to do on the ndlp before puts/releases
   can be performed.

Link: https://lore.kernel.org/r/20201115192646.12977-10-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix NPIV discovery and Fabric Node detection

While testing NPIV and link bounces, the vport would not show a fabric node
for the F_Port, would not transition into NPR state during a link fault, or
leave the FDMI node untouched during error injection. Cause for this was
determined to be an inconsistent manner in which F_Port, Nameserver, and
FDMI controller nodes were created and linked. In some cases, the nodes
would never be unregistered from the transport, leaving references
active. In other cases, the fabric nodes may register with the transport
multiple times while still registered.

The following changes were made:

 - Fix the FDISC issue routine, which starts vport (re)creation, to mark
   the F_Port as a fabric node (NLP_FABRIC) and allow the F_Port node to
   fully be created and show up in the node list.

 - When remote ports are cleaned up on vport termination, cleanup the
   nameserver and FDMI controller nodes on the vport so they unregister
   from the transport.

 - On link bounces, don't exclude the NPIV Fabric remote ports from
   transitioning to the NPR state, allowing them to avoid re-registration
   if already registered.

Link: https://lore.kernel.org/r/20201115192646.12977-9-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Unsolicited ELS leaves node in incorrect state while dropping it

When a target swap happens, under certain conditions the node sends a
LOGO. The unsolicited ELS logic responds with a reject. The logic may
allocate a new node to handle this. Afterward, the new nodes are dropped
incorrectly leaving them in a mis-matched state and refcounting causes a
use-after-free situation leading to a crash.

It is also possible that the unsolicited els handling finds a node which is
in an UNUSED state. The handling moves these nodes to NPR state with a
refcount of 1. Although the end of the discovery logic assumes a final put
will free such a node, there are codes paths which could increment the
reference count, thus the node is in NPR state and not released.
Eventually this mismatch in state and refcount leads to premature release
of the node causing a crash.

Fix by always using the discovery engine DEVICE RM event to decrement and
release the nodes (rather than explicit code that tried to do it before).
This will take care of moving the node to the UNUSED state and then removes
the final ref count. If there is a trigger to reuse this node, the
transition from the UNUSED state clearly indicates that the initial
reference is then incremented and use can continue.

Link: https://lore.kernel.org/r/20201115192646.12977-8-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Remove ndlp when a PLOGI/ADISC/PRLI/REG_RPI ultimately fails

When a PLOGI/ADISC/PRLI/REG_RPI fails, the node remains in the nodelist in
that state.  Although the driver now frees a node when the ref count goes
to zero, in this case the ref cnt doesn't reach zero because there isn't a
mechanism to release the final reference.  Discovery just stops.

Fix by calling the node discovery state machine DEVICE_RM event whenever
one of these commands fail. This will remove the final reference count and
trigger node release.

Link: https://lore.kernel.org/r/20201115192646.12977-7-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Rework remote port lock handling

Currently the discovery layers within the driver use the SCSI midlayer
host_lock to access node-specific structures. This can contend with the I/O
path and is too coarse of a lock.

Rework the driver so that it uses a lock specific to the remote port node
structure when accessing the structure contents. A few of the changes
brought out spots were some slightly reorganized routines worked better.

Link: https://lore.kernel.org/r/20201115192646.12977-6-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix refcounting around SCSI and NVMe transport APIs

Due to bug history and code review, the node reference counting approach in
the driver isn't implemented consistently with how the scsi and nvme
transport perform registrations and unregistrations and their callbacks.
This resulted in many bad/stale node pointers.

Reword the driver so that reference handling is performed as follows:

 - The initial node reference is taken on structure allocation

 - Take a reference on any add/register call to the transport

 - Remove a reference on any delete/unregister call to the transport

 - After the node has fully removed from both the SCSI and NVMEe transports
   (dev_loss_callbacks have called back) call the discovery engine
   DEVICE_RM event which will remove the final reference and release the
   node structure.

 - Alter dev_loss handling when a vport or base port is unloading.

 - Remove the put_node handling - no longer needed.

 - Rewrite the vport_delete handling on reference counts.  Part of this
   effort was driven from the FDISC not registering with the transport and
   disrupting the model for node reference counting.

 - Deleted lpfc_nlp_remove.  Pushed it's remaining ops into
   lpfc_nlp_release.

 - Several other small code cleanups.

Link: https://lore.kernel.org/r/20201115192646.12977-5-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: lpfc: Fix removal of SCSI transport device get and put on dev structure

The lpfc driver is calling get_device and put_device on scsi_fc_transport
device structure. When this code was removed, the driver triggered an oops
in "scsi_is_host_dev" when the first SCSI target was unregistered from the
transport.

The reason the calls were necessary is that the driver is calling
scsi_remove_host too early, before the target rports are unregistered and
the scsi devices disconnected from the scsi_host.  The fc_host was torn
down during fc_remove_host.

Fix by moving the lpfc_pci_remove_one_s3/s4 calls to scsi_remove_host to
after the nodes are cleaned up.  Remove the get_device and put_device calls
and the supporting code.

Link: https://lore.kernel.org/r/20201115192646.12977-4-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>