[PATCH] PM / Domains: restore calling of .suspend/resume_noirq() callbacks

[Grygorii Strashko]

On 11/13/2014 09:11 PM, Geert Uytterhoeven wrote:
> On Thu, Nov 13, 2014 at 7:43 PM, Grygorii Strashko
> <grygorii.strashko at ti.com> wrote:
>> On 11/13/2014 03:33 AM, Rafael J. Wysocki wrote:
>>> On Wednesday, November 12, 2014 06:00:47 PM Grygorii Strashko wrote:
>>>> Now .suspend/resume_noirq() callbacks will not be called during
>>>> system wide suspend/resume for devices which belongs to some GPD.
>>>> It seems, that this change was accidentally introduced by
>>>> commit d23b9b00cdde ("PM / Domains: Rework system suspend callback
>>>>    routines (v2)").
>>>
>>> I'm not sure if that was really accidentally.
>>>
>>> Can you describe the problem that the change below is attempting to
>>> address, without going to much into the history?  IOW, what's that
>>> doesn't work right now?
>>
>> There are no real issues - now in Kernel there are no users of GPD
>> which use "noirq" callbacks.
> 
> Indeed.
> 
> But as the .suspend_noirq() and .resume_noirq() callbacks are not called
> when using the generic PM domain, I had to manually handle interrupt
> disable/enable in commit a00d91ea264f974b ("fbdev: sh_mobile_hdmi:
> Re-init regs before irq re-enable on resume").

^ Honestly, this is very useful practice, because with SMP enabled
the IRQ can be triggered after .suspend() and before .suspend_noirq()
which, in turn, may schedule some kthread/work or even threaded_irq_handler
on secondary cpus (disable_nonboot_cpus() is called after suspend_noirq stage).

Funny things may happen after that :P like: some work/thread which is 
servicing request from I2C device (for example) will be frozen in the middle of
its execution and then it will be resumed right after enable_nonboot_cpus().

^Issue from real life.

regards,
-grygorii