Re: [PATCH v3] PM: sleep: core: Avoid setting power.must_resume to false

[psodagud]

On 2021-08-13 00:23, Greg KH wrote:
> On Tue, Aug 10, 2021 at 01:05:38PM -0700, Prasad Sodagudi wrote:
> > There are variables(power.may_skip_resume and dev->power.must_resume)
> > and DPM_FLAG_MAY_SKIP_RESUME flags to control the resume of devices 
> > after
> > a system wide suspend transition.
> >
> > Setting the DPM_FLAG_MAY_SKIP_RESUME flag means that the driver allows
> > its "noirq" and "early" resume callbacks to be skipped if the device
> > can be left in suspend after a system-wide transition into the working
> > state. PM core determines that the driver's "noirq" and "early" resume
> > callbacks should be skipped or not with dev_pm_skip_resume() function 
> > by
> > checking power.may_skip_resume variable.
> >
> > power.must_resume variable is getting set to false in 
> > __device_suspend()
> > function without checking device's DPM_FLAG_MAY_SKIP_RESUME and
> > dev->power.usage_count variables. In problematic scenario, where
> > all the devices in the suspend_late stage are successful and some
> > device can fail to suspend in suspend_noirq phase. So some devices
> > successfully suspended in suspend_late stage are not getting chance
> > to execute __device_suspend_noirq() to set dev->power.must_resume
> > variable to true and not getting resumed in early_resume phase.
> >
> > Add a check for device's DPM_FLAG_MAY_SKIP_RESUME flag before
> > setting power.must_resume variable in __device_suspend function.
> >
> > Fixes: 6e176bf8d461 ("PM: sleep: core: Do not skip callbacks in the 
> > resume phase")
> > Signed-off-by: Prasad Sodagudi <psodagud@xxxxxxxxxxxxxx>
> > ---
> >  V2 -> V3: Format issues patch posting
> >  V1 -> V2: Fixed indentation and commit text to include scenario
> >  drivers/base/power/main.c | 6 +++++-
> >  1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
> > index d568772..9ee6987 100644
> > --- a/drivers/base/power/main.c
> > +++ b/drivers/base/power/main.c
> > @@ -1642,7 +1642,11 @@ static int __device_suspend(struct device *dev, 
> > pm_message_t state, bool async)
> >  	}
> >
> >  	dev->power.may_skip_resume = true;
> > -	dev->power.must_resume = false;
> > +	if ((atomic_read(&dev->power.usage_count) <= 1) &&
> > +	     (dev_pm_test_driver_flags(dev, DPM_FLAG_MAY_SKIP_RESUME)))
> > +		dev->power.must_resume = false;
> > +	else
> > +		dev->power.must_resume = true;
>
> Again, what happens if the usage_count changes right after reading the
> value?  What protects that from happening?

Hi Gregh KH,
Yes. you are right.  I think, relying on  the usage_count at the  
__device_suspend stage may not be correct.
Devices IRQs are still enabled and usage_count can be changed even after 
reading.
I will send next patchset without power.usage_count check.

@@ -1649,7 +1651,10 @@ static int __device_suspend(struct device *dev, 
pm_message_t state, bool async)
        }

        dev->power.may_skip_resume = true;
-       dev->power.must_resume = false;
+       if (dev_pm_test_driver_flags(dev, DPM_FLAG_MAY_SKIP_RESUME))
+               dev->power.must_resume = false;
+       else
+               dev->power.must_resume = true;


> thanks,
>
> greg k-h