Re: [PATCH v3 1/6] PM / core: Add LEAVE_SUSPENDED driver flag

From: Rafael J. Wysocki
Date: Thu Nov 16 2017 - 18:08:31 EST


On Thursday, November 16, 2017 4:10:16 PM CET Ulf Hansson wrote:
> On 12 November 2017 at 01:37, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote:
> > From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> >
> > Define and document a new driver flag, DPM_FLAG_LEAVE_SUSPENDED, to
> > instruct the PM core and middle-layer (bus type, PM domain, etc.)
> > code that it is desirable to leave the device in runtime suspend
> > after system-wide transitions to the working state (for example,
> > the device may be slow to resume and it may be better to avoid
> > resuming it right away).
> >
> > Generally, the middle-layer code involved in the handling of the
> > device is expected to indicate to the PM core whether or not the
> > device may be left in suspend with the help of the device's
> > power.may_skip_resume status bit. That has to happen in the "noirq"
> > phase of the preceding system suspend (or analogous) transition.
> > The middle layer is then responsible for handling the device as
> > appropriate in its "noirq" resume callback which is executed
> > regardless of whether or not the device may be left suspended, but
> > the other resume callbacks (except for ->complete) will be skipped
> > automatically by the core if the device really can be left in
> > suspend.
> >
> > The additional power.must_resume status bit introduced for the
> > implementation of this mechanisn is used internally by the PM core
> > to track the requirement to resume the device (which may depend on
> > its children etc).
> >
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > Acked-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > ---
> >
> > v2 -> v3: Take dev->power.usage_count when updating power.must_resume in
> > __device_suspend_noirq().
> >
> > ---
> > Documentation/driver-api/pm/devices.rst | 24 ++++++++++-
> > drivers/base/power/main.c | 66 +++++++++++++++++++++++++++++---
> > drivers/base/power/runtime.c | 9 ++--
> > include/linux/pm.h | 14 +++++-
> > include/linux/pm_runtime.h | 9 ++--
> > 5 files changed, 104 insertions(+), 18 deletions(-)
> >
> > Index: linux-pm/include/linux/pm.h
> > ===================================================================
> > --- linux-pm.orig/include/linux/pm.h
> > +++ linux-pm/include/linux/pm.h
> > @@ -559,6 +559,7 @@ struct pm_subsys_data {
> > * NEVER_SKIP: Do not skip system suspend/resume callbacks for the device.
> > * SMART_PREPARE: Check the return value of the driver's ->prepare callback.
> > * SMART_SUSPEND: No need to resume the device from runtime suspend.
> > + * LEAVE_SUSPENDED: Avoid resuming the device during system resume if possible.
> > *
> > * Setting SMART_PREPARE instructs bus types and PM domains which may want
> > * system suspend/resume callbacks to be skipped for the device to return 0 from
> > @@ -572,10 +573,14 @@ struct pm_subsys_data {
> > * necessary from the driver's perspective. It also may cause them to skip
> > * invocations of the ->suspend_late and ->suspend_noirq callbacks provided by
> > * the driver if they decide to leave the device in runtime suspend.
> > + *
> > + * Setting LEAVE_SUSPENDED informs the PM core and middle-layer code that the
> > + * driver prefers the device to be left in runtime suspend after system resume.
> > */
>
> Question: Can LEAVE_SUSPENDED and NEVER_SKIP be valid combination? I
> guess not!? Should we validate for wrong combinations?

Why not? There's no real overlap between them.

>
> [...]
>
> > /**
> > * __device_suspend_noirq - Execute a "noirq suspend" callback for given device.
> > * @dev: Device to handle.
> > @@ -1127,10 +1161,28 @@ static int __device_suspend_noirq(struct
> > }
> >
> > error = dpm_run_callback(callback, dev, state, info);
> > - if (!error)
> > - dev->power.is_noirq_suspended = true;
> > - else
> > + if (error) {
> > async_error = error;
> > + goto Complete;
> > + }
> > +
> > + dev->power.is_noirq_suspended = true;
> > +
> > + if (dev_pm_test_driver_flags(dev, DPM_FLAG_LEAVE_SUSPENDED)) {
> > + /*
> > + * The only safe strategy here is to require that if the device
> > + * may not be left in suspend, resume callbacks must be invoked
> > + * for it.
> > + */
> > + dev->power.must_resume = dev->power.must_resume ||
> > + !dev->power.may_skip_resume ||
> > + atomic_read(&dev->power.usage_count);
>
> dev->power.usage_count is always > 0 at this point, meaning that
> dev->power.must_resume always becomes true. :-)
>
> You should rather use "atomic_read(&dev->power.usage_count) > 1".

Right, thanks. I tend to forget about that.

> > + } else {
> > + dev->power.must_resume = true;
> > + }
> > +
> > + if (dev->power.must_resume)
> > + dpm_superior_set_must_resume(dev);
> >
> > Complete:
> > complete_all(&dev->power.completion);
> > @@ -1487,6 +1539,9 @@ static int __device_suspend(struct devic
> > dev->power.direct_complete = false;
> > }
> >
> > + dev->power.may_skip_resume = false;
> > + dev->power.must_resume = false;
> > +
>
> First, these assignment could be bypassed if the direct_complete path
> is used. Perhaps it's more robust to reset these flags already in
> device_prepare().

In the direct-complete case may_skip_resume doesn't matter.

must_resume should be set to "false", however, so that parents of
direct-complete devices may be left in suspend (in case they don't
fall under direct-complete themselves), so good catch.

But it is sufficient to do that before the power.direct_complete check above. :-)

> Second, have you considered setting the default value of
> dev->power.may_skip_resume to true?

Yes.

> That would means the subsystem
> instead need to implement an opt-out method. I am thinking that it may
> not be an issue, since we anyway at this point, don't have drivers
> using the LEAVE_SUSPENDED flag.

Opt-out doesn't work because of the need to invoke the "noirq" callbacks.

> [...]
>
> > +However, it may be desirable to leave some devices in runtime suspend after
> > +system transitions to the working state and device drivers can use the
> > +``DPM_FLAG_LEAVE_SUSPENDED`` flag to indicate to the PM core (and middle-layer
> > +code) that this is the case. Whether or not the devices will actually be left
> > +in suspend may depend on their state before the given system suspend-resume
> > +cycle and on the type of the system transition under way. In particular,
> > +devices are not left suspended if that transition is a restore from hibernation,
> > +as device states are not guaranteed to be reflected by the information stored in
> > +the hibernation image in that case.
> > +
> > +The middle-layer code involved in the handling of the device has to indicate to
> > +the PM core if the device may be left in suspend with the help of its
> > +:c:member:`power.may_skip_resume` status bit. That has to happen in the "noirq"
> > +phase of the preceding system-wide suspend (or analogous) transition. The
>
> Does it have to be managed in the "noirq" phase? Wouldn't be perfectly
> okay do this in the suspend and suspend_late phases as well?

The wording is slightly misleading I think.

In fact technically may_skip_resume may be set earlier, but the core checks it
in the "noirq" phase only anyway.

Thanks,
Rafael