Check the size of another pointer instead of NULL. This fixes this
smatch warning:

drivers/staging/media/atomisp/pci/sh_css.c:3609 ia_css_pipe_enqueue_buffer() warn: sizeof(NUMBER)?

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>

This fixes a sparse warning:

drivers/staging/media/atomisp/pci/atomisp_drvfs.c:40:14: warning: symbol 'dbgopt' was not declared. Should it be static?

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>

Remove the TODO items for using the main (drivers/media/i2c) ov2680 and
ov5693 drivers and removing the atomisp specific ones, this has been done.

Remove the TODO item for gracefully handling missing firmware the
"media: atomisp: Bind and do power-management without firmware" changes
have fixed this.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Remove unnecessary msleep(10) from atomisp_mrfld_power() error-exit path,
the success exit from atomisp_mrfld_power() happens if a test succeeds
inside the do { } while loop above the msleep().

The error-exit path with the removed msleep is only hit it the power-on is
not reflected in the iUNIT ISPSSPM0 status bits after a timeout of 50 ms.
Sleeping an extra 10 ms in the timeout path makes little sense.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

The ISP needs to be turned off in a special manner for the SoC
to be able to reach S0i3 during suspend.

When the firmware is missing still bind to the PCI device and
run in pm-only mode which takes care of turning the ISP off
(including turning it off again if the firmware has turned it
on during resume).

In this new pm-only mode the atomisp driver works exactly the same
as the non-staging, pm-only drivers/platform/x86/intel/atomisp2/pm.c
driver.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Move the power-management setup and cleanup code into
atomisp_pm_[un]init() helper functions.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

sysfs attributes preferably should not be manually be registered but
instead the driver.groups / driver.dev_groups driver struct members
should be used to have the driver core handle this in a race free
manner.

Using driver.groups would be the most direct replacement for
driver_[add|remove]_file, but some of the attributes actually need access
to the struct atomisp_device (*), so as part of modernizing this part of
the atomisp driver this change also makes the sysfs attribute device
attributes instead of driver attributes.

*) Before this change accessing these attributes without the driver having
bound would result in a NULL pointer deref, this commit fixes this.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Fix probe()/remove() power-management:

-Currently the driver uses pm_runtime_put_noidle() and relies on
 userspace to open + close the /dev/video# node at least once to
 actually turn the ISP off. Replace the pm_runtime_put_noidle()
 with pm_runtime_put_sync() to make sure that the device is turned
 off without relying on userspace for this.
 This also ensures that atomisp_css_init() is run (by atomisp_power_on())
 if the first userspace process opening /dev/video# wants to do more then
 just query the v4l2-caps.
 As part of this change move the pm setup code in probe() to just before
 calling v4l2_async_nf_register() which registers the /dev/* nodes, so
 that the device is left on for the entirety of the probe() function.

-Remove the turning off of the atomisp from the exit-error path,
 the PCI subsystem and subsequent probe() attempts expect the device
 to be in the on state when probe() fails.
 This also fixes the atomisp driver causing the system to hang / freeze
 when its firmware is missing. This freeze is caused by an unidentified
 bug in the power-off on exit-error code which is now removed.

-Make sure that remove() properly powers on the device by replacing
 pm_runtime_get_noresume() with pm_runtime_get_sync. The PCI subsystem
 and subsequent probe() attempts expect the device to be in the on state
 after unbinding the driver.

-Note this also swaps the order of put()/allow() and forbid()/get()
 so that the sync versions actually work by calling allow() before put()
 and forbid() after get()

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Group the cpu_latency_qos_add_request() call in probe() together with
the other PM calls in probe().

This is a preparation patch for futher PM fixes / work.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Fix atomisp_pci_remove():
-Remove uninformative "Removing atomisp driver" log message
-Add missing devm_free_irq(), atomisp_uninitialize_modules() and
 pci_free_irq_vectors() calls
-Move atomisp_msi_irq_uninit() down so that the remove() order is
 an exact mirror of the probe() order

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>