Re: [PATCH] ACPI / LPSS: Don't skip late system PM ops for hibernate on BYT/CHT

From: Rafael J. Wysocki
Date: Mon Jun 24 2019 - 19:14:42 EST


On Monday, June 24, 2019 12:51:33 PM CEST Hans de Goede wrote:
> Hi Rafael,
>
> <snip>
>
> > Sorry for the long delay.
> >
> > I haven't dropped this issue on the floor, I hope that you are still able to follow up here.
> >
> > Can you please test the appended patch instead of the previous one?
> >
> > I have found some inconsistencies in the handling of hibernation in the ACPI PM domain
> > and the LPSS driver that should be covered by this patch.
>
> I know this is just a testing patch for now, but still I've given it
> a quick look, some comments inline.
>
> > ---
> > drivers/acpi/acpi_lpss.c | 63 +++++++++++++++++++++++++++++++++++------------
> > drivers/acpi/device_pm.c | 30 ++++++++++++++++++++--
> > include/linux/acpi.h | 4 ++
> > 3 files changed, 79 insertions(+), 18 deletions(-)
> >
> > Index: linux-pm/drivers/acpi/device_pm.c
> > ===================================================================
> > --- linux-pm.orig/drivers/acpi/device_pm.c
> > +++ linux-pm/drivers/acpi/device_pm.c
> > @@ -1171,6 +1171,32 @@ int acpi_subsys_thaw_noirq(struct device
> > return pm_generic_thaw_noirq(dev);
> > }
> > EXPORT_SYMBOL_GPL(acpi_subsys_thaw_noirq);
> > +
> > +/**
> > + * acpi_subsys_restore_noirq - Run the device driver's "noirq" restore callback.
> > + * @dev: Device to handle.
> > + */
> > +int acpi_subsys_restore_noirq(struct device *dev)
> > +{
> > + /* This is analogous to what acpi_subsys_resune_noirq() does. */
> > + if (dev_pm_smart_suspend_and_suspended(dev))
> > + pm_runtime_set_active(dev);
> > +
> > + return pm_generic_restore_noirq(dev);
> > +}
> > +EXPORT_SYMBOL_GPL(acpi_subsys_restore_noirq);
> > +
> > +/**
> > + * acpi_subsys_restore_early - Restore device using ACPI.
> > + * @dev: Device to restore.
> > + */
> > +int acpi_subsys_restore_early(struct device *dev)
> > +{
> > + int ret = acpi_dev_resume(dev);
> > + return ret ? ret : pm_generic_restore_early(dev);
> > +}
> > +EXPORT_SYMBOL_GPL(acpi_subsys_restore_early);
> > +
> > #endif /* CONFIG_PM_SLEEP */
> >
> > static struct dev_pm_domain acpi_general_pm_domain = {
> > @@ -1192,8 +1218,8 @@ static struct dev_pm_domain acpi_general
> > .poweroff = acpi_subsys_suspend,
> > .poweroff_late = acpi_subsys_suspend_late,
> > .poweroff_noirq = acpi_subsys_suspend_noirq,
> > - .restore_noirq = acpi_subsys_resume_noirq,
> > - .restore_early = acpi_subsys_resume_early,
> > + .restore_noirq = acpi_subsys_restore_noirq,
> > + .restore_early = acpi_subsys_restore_early,
> > #endif
> > },
> > };
> > Index: linux-pm/drivers/acpi/acpi_lpss.c
> > ===================================================================
> > --- linux-pm.orig/drivers/acpi/acpi_lpss.c
> > +++ linux-pm/drivers/acpi/acpi_lpss.c
> > @@ -1069,36 +1069,67 @@ static int acpi_lpss_suspend_noirq(struc
> > return acpi_subsys_suspend_noirq(dev);
> > }
> >
> > -static int acpi_lpss_do_resume_early(struct device *dev)
> > +static int acpi_lpss_resume_noirq(struct device *dev)
> > {
> > - int ret = acpi_lpss_resume(dev);
> > + struct lpss_private_data *pdata = acpi_driver_data(ACPI_COMPANION(dev));
> > +
> > + /* Follow acpi_subsys_resune_noirq(). */
> > + if (dev_pm_may_skip_resume(dev))
> > + return 0;
> > +
> > + if (dev_pm_smart_suspend_and_suspended(dev))
> > + pm_runtime_set_active(dev);
> >
> > - return ret ? ret : pm_generic_resume_early(dev);
> > + if (pdata->dev_desc->resume_from_noirq) {
> > + int ret = acpi_lpss_resume(dev);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + return pm_generic_resume_noirq(dev);
> > }
>
> Hmm, normally acpi_lpss_resume runs at resume_early time, AFAIK
> the order of resume callbacks calling is: resume_noirq, resume_early, resume
>
> So normally our call order is:
>
> ---noirq-phase---
> pm_generic_resume_noirq()
> ---early-phase---
> acpi_lpss_resume()
> pm_generic_resume_early()
>
> My patch adding the resume_from_noirq flag, move the calling of
> acpi_lpss_resume() to the resume_noirq phase (if the flag is
> set) but kept the generic order, so the call order with the
> flag set currently is:
>
> ---noirq-phase---
> pm_generic_resume_noirq()
> acpi_lpss_resume()
> ---early-phase---
> pm_generic_resume_early()
>
> So the order of the 3 calls relative to each other did not change.
>
> You are changing this to:
>
> ---noirq-phase---
> acpi_lpss_resume()
> pm_generic_resume_noirq()
> ---early-phase---
> pm_generic_resume_early()
>
> So now when the flag is set acpi_lpss_resume() runs before
> pm_generic_resume_noirq(). Is this intentional ?

Kind of yes, but this is two patches in one. :-)

The ordering change should really be a separate patch IMO.