Message-ID: <CAPDyKFpmNPhyV3YoBFu7KnW04550DQgqzGHAbGLLqp7=TggVtw@mail.gmail.com>
Date: Thu, 30 Jan 2025 12:11:02 +0100
From: Ulf Hansson <ulf.hansson@...aro.org>
To: "Rafael J. Wysocki" <rafael@...nel.org>
Cc: "Rafael J. Wysocki" <rjw@...ysocki.net>, Linux PM <linux-pm@...r.kernel.org>, 
	LKML <linux-kernel@...r.kernel.org>, Alan Stern <stern@...land.harvard.edu>, 
	Bjorn Helgaas <helgaas@...nel.org>, Linux PCI <linux-pci@...r.kernel.org>, 
	Johan Hovold <johan@...nel.org>, Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>, 
	Kevin Xie <kevin.xie@...rfivetech.com>
Subject: Re: [PATCH v1] PM: sleep: core: Synchronize runtime PM status of
 parents and children

On Wed, 29 Jan 2025 at 17:58, Rafael J. Wysocki <rafael@...nel.org> wrote:
>
> On Wed, Jan 29, 2025 at 5:42 PM Ulf Hansson <ulf.hansson@...aro.org> wrote:
> >
> > On Wed, 29 Jan 2025 at 16:55, Rafael J. Wysocki <rafael@...nel.org> wrote:
> > >
> > > On Wed, Jan 29, 2025 at 12:53 PM Ulf Hansson <ulf.hansson@...aro.org> wrote:
> > > >
> > > > On Tue, 28 Jan 2025 at 20:24, Rafael J. Wysocki <rjw@...ysocki.net> wrote:
> > > > >
> > > > > From: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> > > > >
> > > > > Commit 6e176bf8d461 ("PM: sleep: core: Do not skip callbacks in the
> > > > > resume phase") overlooked the case in which the parent of a device with
> > > > > DPM_FLAG_SMART_SUSPEND set did not use that flag and could be runtime-
> > > > > suspended before a transition into a system-wide sleep state.  In that
> > > > > case, if the child is resumed during the subsequent transition from
> > > > > that state into the working state, its runtime PM status will be set to
> > > > > RPM_ACTIVE, but the runtime PM status of the parent will not be updated
> > > > > accordingly, even though the parent will be resumed too, because of the
> > > > > dev_pm_skip_suspend() check in device_resume_noirq().
> > > > >
> > > > > Address this problem by tracking the need to set the runtime PM status
> > > > > to RPM_ACTIVE during system-wide resume transitions for devices with
> > > > > DPM_FLAG_SMART_SUSPEND set and all of the devices depended on by them.
> > > > >
> > > > > Fixes: 6e176bf8d461 ("PM: sleep: core: Do not skip callbacks in the resume phase")
> > > > > Closes: https://lore.kernel.org/linux-pm/Z30p2Etwf3F2AUvD@hovoldconsulting.com/
> > > > > Reported-by: Johan Hovold <johan@...nel.org>
> > > > > Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>
> > > > > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> > > > > ---
> > > > >  drivers/base/power/main.c |   29 ++++++++++++++++++++---------
> > > > >  include/linux/pm.h        |    1 +
> > > > >  2 files changed, 21 insertions(+), 9 deletions(-)
> > > > >
> > > > > --- a/drivers/base/power/main.c
> > > > > +++ b/drivers/base/power/main.c
> > > > > @@ -656,13 +656,15 @@
> > > > >          * so change its status accordingly.
> > > > >          *
> > > > >          * Otherwise, the device is going to be resumed, so set its PM-runtime
> > > > > -        * status to "active", but do that only if DPM_FLAG_SMART_SUSPEND is set
> > > > > -        * to avoid confusing drivers that don't use it.
> > > > > +        * status to "active" unless its power.set_active flag is clear, in
> > > > > +        * which case it is not necessary to update its PM-runtime status.
> > > > >          */
> > > > > -       if (skip_resume)
> > > > > +       if (skip_resume) {
> > > > >                 pm_runtime_set_suspended(dev);
> > > > > -       else if (dev_pm_skip_suspend(dev))
> > > > > +       } else if (dev->power.set_active) {
> > > > >                 pm_runtime_set_active(dev);
> > > > > +               dev->power.set_active = false;
> > > > > +       }
> > > > >
> > > > >         if (dev->pm_domain) {
> > > > >                 info = "noirq power domain ";
> > > > > @@ -1189,18 +1191,24 @@
> > > > >         return PMSG_ON;
> > > > >  }
> > > > >
> > > > > -static void dpm_superior_set_must_resume(struct device *dev)
> > > > > +static void dpm_superior_set_must_resume(struct device *dev, bool set_active)
> > > > >  {
> > > > >         struct device_link *link;
> > > > >         int idx;
> > > > >
> > > > > -       if (dev->parent)
> > > > > +       if (dev->parent) {
> > > > >                 dev->parent->power.must_resume = true;
> > > > > +               if (set_active)
> > > > > +                       dev->parent->power.set_active = true;
> > > > > +       }
> > > > >
> > > > >         idx = device_links_read_lock();
> > > > >
> > > > > -       list_for_each_entry_rcu_locked(link, &dev->links.suppliers, c_node)
> > > > > +       list_for_each_entry_rcu_locked(link, &dev->links.suppliers, c_node) {
> > > > >                 link->supplier->power.must_resume = true;
> > > > > +               if (set_active)
> > > > > +                       link->supplier->power.set_active = true;
> > > >
> > > > If I understand correctly, the suppliers are already handled when the
> > > > pm_runtime_set_active() is called for consumers, so the above should
> > > > not be needed.
> > >
> > > It is needed because pm_runtime_set_active() doesn't cause the setting
> > > to propagate to the parent's/suppliers of the suppliers AFAICS.
> >
> > Hmm, even if that sounds reasonable, I don't think it's a good idea as
> > it may introduce interesting propagation problems between drivers.
> >
> > For example, consider that Saravana is trying to enable runtime PM for
> > fw_devlinks. It would mean synchronization issues for the runtime PM
> > status, all over the place.
>
> What synchronization issues?

Changing the runtime PM status for a parent/supplier that doesn't have
DPM_FLAG_SMART_SUSPEND, is likely to confuse their drivers.

You also removed that part of the comment a few lines above, in
device_resume_noirq(). I am not sure I understand why?

>
> > That said, is even consumer/suppliers part of the problem we are
> > trying to solve?
>
> They are in general.
>
> It's just that stuff that was runtime-suspended prior to a system-wide
> suspend may need to be resumed and marked as RPM_ACTIVE during
> system-wide resume because one of the devices wants/needs to be
> resumed then.
>
> If this turns out to be problematic, the problem will need to be
> addressed, but for now I'm not seeing why there would be a problem.
>
> > >
> > > > That said, maybe we instead allow parent/child to work in the similar
> > > > way as for consumer/suppliers, when pm_runtime_set_active() is called
> > > > for the child. In other words, when pm_runtime_set_active() is called
> > > > for a child and the parent is runtime PM enabled, let's runtime resume
> > > > it too, as we do for suppliers. Would that work, you think?
> > >
> > > The parent is not runtime-PM enabled when this happens.
> >
> > That sounds really weird to me.
> >
> > Does that mean that the parent has not been system resumed either?
>
> Yes.
>
> It hasn't been resumed yet, but it is known that it will be resumed.
>
> > If so, isn't that really the root cause for this problem,
>
> No, it is not.
>
> > or what am I missing?
>
> Essentially, what I said above.
>
> If a device that was suspended prior to a system-wide suspend
> wants/needs to be resumed during the subsequent system-wide resume,
> and it was runtime-PM-enabled before the suspend transition, it needs
> to (a) be runtime-PM-enabled during the subsequent system-wide resume
> transition and (b) it also needs to be marked as RPM_ACTIVE because in
> fact it is not suspended any more.  The existing code before the patch
> takes care of this for the device itself, but not for the devices it
> depends on which also need to be resumed (which happens) and marked as
> RPM_ACTIVE (which doesn't happen) and the patch just makes sure that
> the latter will happen.

Thanks for clarifying!

>
> Actually, what happens now is that the actual state of the parent
> during the system-wide resume, right before re-enabling runtime PM for
> it, does not match its runtime PM status which is still RPM_SUSPENDED.
> That's what is fixed here and it applies to the parent as well as to
> all of the other devices depended on by the child and the parent.

Well, unfortunately I don't think it will work to call
pm_runtime_set_active() for parents/suppliers like this.

I think we need the drivers for the parents/suppliers to be in
agreement with the behaviour of DPM_FLAG_SMART_SUSPEND to allow the
propagation. Not sure how to best achieve this though.

Kind regards
Uffe

Powered by blists - more mailing lists