LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Greg KH <gregkh@linuxfoundation.org>
To: psodagud@codeaurora.org
Cc: rjw@rjwysocki.net, len.brown@intel.com, pavel@ucw.cz,
	linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] PM: sleep: core: Avoid setting power.must_resume to false
Date: Sat, 7 Aug 2021 08:00:47 +0200	[thread overview]
Message-ID: <YQ4hj+blYxHdUl79@kroah.com> (raw)
In-Reply-To: <0fe9a873ca77293151a61f64d355895f@codeaurora.org>

On Fri, Aug 06, 2021 at 08:07:08AM -0700, psodagud@codeaurora.org wrote:
> On 2021-08-03 10:16, Greg KH wrote:
> > On Mon, Jul 26, 2021 at 08:24:34AM -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. This is leading to failure to call
> > > resume handler for some of the devices suspended in early suspend
> > > phase.
> > > So check device's DPM_FLAG_MAY_SKIP_RESUME flag before
> > > setting power.must_resume variable.
> > > 
> > > Signed-off-by: Prasad Sodagudi <psodagud@codeaurora.org>
> > > ---
> > >  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..8eebc4d 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)))
> > 
> > What is preventing that atomic value from changing _right_ after you
> > just read this?
> > 
> > and very odd indentation, checkpatch didn't complain about this?
> Sure. I will fix indentation problem once Rafael reviewed this patch.
> 
> > What commit does this fix?  Does it need to be backported to older
> > kernels?
> 
> No. LTS - 5.4 do not have this problem.
> 
> > Wait, how is your "noirq" device even getting called here?  Shouldn't
> > __device_suspend_noirq() be called instead?  Why isn't that the path for
> > your device?
> 
> Hi Gregh and Rafael,
> 
> This is regarding suspend/resume(s2idle) scenario of devices and differences
> between the LTS kernel 5.4 and 5.10 with respect to devices suspend and
> resume. Observing that devices suspended in suspend_late stage are not
> getting resumed in resume_early stage.
> 1)	LTS kernel 5.4 kernel do not have this problem but 5.10 kernel shows this
> problem.
> 2)	Commit - 6e176bf8d46194353163c2cb660808bc633b45d9 (PM: sleep: core: Do
> not skip callbacks in the resume phase) is skipping the driver early_resume
> callbacks.
> @@ -804,15 +793,25 @@ static int device_resume_early(struct device *dev,
> pm_message_t state, bool asyn
>         } else if (dev->bus && dev->bus->pm) {
>                 info = "early bus ";
>                 callback = pm_late_early_op(dev->bus->pm, state);
> -       } else if (dev->driver && dev->driver->pm) {
> +       }
> +       if (callback)
> +               goto Run;
> +
> +       if (dev_pm_may_skip_resume(dev))
> +               goto Skip;
> In device_resume_early function dev->power.must_resume is used to skip the
> resume call back. It looks this function is expecting that,
> __device_suspend_noirq() would set dev->power.must_resume = true for the
> devices which does not have DPM_FLAG_MAY_SKIP_RESUME flag set.
> 
> static int __device_suspend_noirq(struct device *dev, pm_message_t state,
> bool async)
> {
> …
> …
>         /*
>          * Skipping the resume of devices that were in use right before the
>          * system suspend (as indicated by their PM-runtime usage counters)
>          * would be suboptimal.  Also resume them if doing that is not
> allowed
>          * to be skipped.
>          */
>         if (atomic_read(&dev->power.usage_count) > 1 ||
>             !(dev_pm_test_driver_flags(dev, DPM_FLAG_MAY_SKIP_RESUME) &&
>               dev->power.may_skip_resume))
>                 dev->power.must_resume = true;
> 
> 
> 3)	Problematic scenario is as follows -  During the device suspend/resume
> scenario all the devices in  the suspend_late stage are successful and some
> device can fail to suspend in suspend_noirq(device_suspend_noirq->
> __device_suspend_noirq) phase.
> As a device failed in dpm_noirq_suspend_devices phase, dpm_resume_noirq is
> getting called to resume devices in dpm_late_early_list in the noirq phase.
> 4)	During the Devices_early_resume stage
> dpm_resume_early()-->device_resume_early() functions skipping the devices
> early resume callbacks.
>  799         if (dev_pm_skip_resume(dev))
> 800	              goto Skip;
> 
> 5)	Devices suspended in suspend_late stage are not getting resumed in
> Devices_early_resume stage because of Commit -
> 6e176bf8d46194353163c2cb660808bc633b45d9 (PM: sleep: core: Do not skip
> callbacks in the resume phase) is skipping the driver early_resume callbacks
> when dev->power.must_resume is false.
> 
> 6)	Below portion of the code in __device_suspend_noirq is not getting
> executed for some drivers successfully suspended in suspend_late stage and
> there is no chance to set must_resume to true.  So these devices are always
> having dev->power.must_resume=false.
> For example -
> i)	Devices A, B, C have suspend_late and resume_early handlers.
> ii)	Devices X, Y, Z  have suspend_noirq and resume_noirq handlers.
> Devices are getting suspended in this order – A, B, X , C , Y and Z and
> device X return failure for suspend_noirq callback. In this scenario, device
> C would never execute below portion of the code to set
> dev->power.must_resume = true and device – C would not get resumed in
> resume_early  stage.
> 
> 1192 static int __device_suspend_noirq(struct device *dev, pm_message_t
> state, bool async)
> 1193 {
> ….
> ….
> 1245         /*
> 1246          * Skipping the resume of devices that were in use right before
> the
> 1247          * system suspend (as indicated by their PM-runtime usage
> counters)
> 1248          * would be suboptimal.  Also resume them if doing that is not
> allowed
> 1249          * to be skipped.
> 1250          */
> 1251         if (atomic_read(&dev->power.usage_count) > 1 ||
> 1252             !(dev_pm_test_driver_flags(dev, DPM_FLAG_MAY_SKIP_RESUME)
> &&
> 1253               dev->power.may_skip_resume))
> 1254                 dev->power.must_resume = true;
> 1255
> 1256         if (dev->power.must_resume)
> 1257                 dpm_superior_set_must_resume(dev);
> 1258

Ok, that explains it a bit better, thank you.  Can you please try to
expand on your changelog text when you resubmit this to include this
information and properly identify what commit caused this problem to
happen by adding a Fixes: tag?

thanks,

greg k-h

  parent reply	other threads:[~2021-08-07  6:01 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-26 15:24 Prasad Sodagudi
2021-08-03 17:16 ` Greg KH
2021-08-06 15:07   ` psodagud
2021-08-06 15:16     ` Greg KH
2021-08-07  6:00     ` Greg KH [this message]
2021-08-08 16:10       ` [PATCH v2] " Prasad Sodagudi
2021-08-08 17:36         ` Greg KH

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=YQ4hj+blYxHdUl79@kroah.com \
    --to=gregkh@linuxfoundation.org \
    --cc=len.brown@intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=pavel@ucw.cz \
    --cc=psodagud@codeaurora.org \
    --cc=rjw@rjwysocki.net \
    --subject='Re: [PATCH] PM: sleep: core: Avoid setting power.must_resume to false' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).