Re: [PATCH] PM: runtime: Allow rpm_resume() to succeed when runtime PM is disabled

[Rafael J. Wysocki]

On Wednesday, December 1, 2021 6:44:08 PM CET Rafael J. Wysocki wrote:
> On Wed, Dec 1, 2021 at 4:23 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> >
> > On Wed, 1 Dec 2021 at 14:49, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> > >
> > > On Wed, Dec 1, 2021 at 10:02 AM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> > > >
> > > > On Tue, 30 Nov 2021 at 18:26, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Tue, Nov 30, 2021 at 5:41 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Tue, 30 Nov 2021 at 14:02, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > On Tue, Nov 30, 2021 at 12:58 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > [...]
> > > > > > > >
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > Am I thinking correctly that this is mostly about working around the
> > > > > > > > > > > > > > limitations of pm_runtime_force_suspend()?
> > > > > > > > > > > > >
> > > > > > > > > > > > > No, this isn't related at all.
> > > > > > > > > > > > >
> > > > > > > > > > > > > The cpuidle-psci driver doesn't have PM callbacks, thus using
> > > > > > > > > > > > > pm_runtime_force_suspend() would not work here.
> > > > > > > > > > > >
> > > > > > > > > > > > Just wanted to send a ping on this to see if we can come to a
> > > > > > > > > > > > conclusion. Or maybe we did? :-)
> > > > > > > > > > > >
> > > > > > > > > > > > I think in the end, what slightly bothers me, is that the behavior is
> > > > > > > > > > > > a bit inconsistent. Although, maybe it's the best we can do.
> > > > > > > > > > >
> > > > > > > > > > > I've been thinking about this and it looks like we can do better, but
> > > > > > > > > > > instead of talking about this I'd rather send a patch.
> > > > > > > > > >
> > > > > > > > > > Alright.
> > > > > > > > > >
> > > > > > > > > > I was thinking along the lines of make similar changes for
> > > > > > > > > > rpm_idle|suspend(). That would make the behaviour even more
> > > > > > > > > > consistent, I think.
> > > > > > > > > >
> > > > > > > > > > Perhaps that's what you have in mind? :-)
> > > > > > > > >
> > > > > > > > > Well, not exactly.
> > > > > > > > >
> > > > > > > > > The idea is to add another counter (called restrain_depth in the patch)
> > > > > > > > > to prevent rpm_resume() from running the callback when that is potentially
> > > > > > > > > problematic.  With that, it is possible to actually distinguish devices
> > > > > > > > > with PM-runtime enabled and it allows the PM-runtime status to be checked
> > > > > > > > > when it is still known to be meaningful.
> > > > > > > >
> > > > > > > > Hmm, I don't quite understand the benefit of introducing a new flag
> > > > > > > > for this. rpm_resume() already checks the disable_depth to understand
> > > > > > > > when it's safe to invoke the callback. Maybe there is a reason why
> > > > > > > > that isn't sufficient?
> > > > > > >
> > > > > > > The problem is that disable_depth > 0 may very well mean that runtime
> > > > > > > PM has not been enabled at all for the given device which IMO is a
> > > > > > > problem.
> > > > > > >
> > > > > > > As it stands, it is necessary to make assumptions, like disable_depth
> > > > > > > == 1 meaning that runtime PM is really enabled, but the PM core has
> > > > > > > disabled it temporarily, which is somewhat questionable.
> > > > > > >
> > > > > > > Another problem with disabling is that it causes rpm_resume() to fail
> > > > > > > even if the status is RPM_ACTIVE and it has to do that exactly because
> > > > > > > it cannot know why runtime PM has been disabled.  If it has never been
> > > > > > > enabled, rpm_resume() must fail, but if it has been disabled
> > > > > > > temporarily, rpm_resume() may return 1 when the status is RPM_ACTIVE.
> > > > > > >
> > > > > > > The new count allows the "enabled in general, but temporarily disabled
> > > > > > > at the moment" to be handled cleanly.
> > > > > >
> > > > > > My overall comment is that I fail to understand why we need to
> > > > > > distinguish between these two cases. To me, it shouldn't really
> > > > > > matter, *why* runtime PM is (or have been) disabled for the device.
> > > > >
> > > > > It matters if you want to trust the status, because "disabled" means
> > > > > "the status doesn't matter".
> > > >
> > > > Well, that doesn't really match how the runtime PM interface is being
> > > > used today.
> > >
> > > Well, I clearly disagree.
> >
> > Alright, then we can agree to disagree. :-)
> >
> > >
> > > > For example, we have a whole bunch of helper functions, allowing us to
> > > > update and check the runtime PM state of the device, even when the
> > > > disable_depth > 0. Some functions, like pm_runtime_set_active() for
> > > > example, even take parents and device-links into account.
> > >
> > > That's true, but that's for a purpose.
> > >
> > > If runtime PM becomes enabled after using pm_runtime_set_active(), the
> > > status should better be consistent with the settings of the parent
> > > etc.
> > >
> > > > >
> > > > > If you want the status to stay meaningful, but prevent callbacks from
> > > > > running, you need something else.
> > > > >
> > > > > > The important point is that the default state for a device is
> > > > > > RPM_SUSPENDED and someone has moved into RPM_ACTIVE, for whatever
> > > > > > reason. That should be sufficient to allow rpm_resume() to return '1'
> > > > > > when disable_depth > 0, shouldn't it?
> > > > >
> > > > > No, because there is no rule by which the status of devices with
> > > > > PM-runtime disabled must be RPM_SUSPENDED.
> > > >
> > > > That's not what I was trying to say.
> > > >
> > > > The initial/default runtime PM state for a device is RPM_SUSPENDED,
> > > > which is being set in pm_runtime_init(). Although, I agree that it
> > > > can't be trusted that this state actually reflects the state of the
> > > > HW, it's still a valid state for the device from a runtime PM point of
> > > > view.
> > >
> > > No, it is not.  It's just the default.
> > >
> > > > However, and more importantly, if the state has moved to RPM_ACTIVE,
> > > > someone must have deliberately moved the device into that state.
> > >
> > > Sure, but it cannot be regarded as an indication on whether or not
> > > runtime PM is supported and has ever been enabled for the given
> > > device.
> > >
> > > Again, there is no rule regarding the status value for devices with
> > > runtime PM disabled, either way.
> >
> > If I understand correctly, that means you think the
> > pm_runtime_status_suspended() should really be converted to an
> > internal runtime PM interface, not being exported to users outside.
> > Right?
> 
> Not really.
> 
> I'm just saying that its usefulness is limited.
> 
> My basic concern is that system-wide PM transitions must always invoke
> callbacks for devices with PM-runtime disabled, because they may (or
> may not) be functional regardless of the PM-runtime status and if they
> are functional, they must be suspended.  And note that supporting
> system-wide PM is not optional and the only way to kind of disable it
> is to return an error from a device suspend callback (but that's nasty
> for some use cases).
> 
> So the "Has PM-runtime been enabled?" question is really fundamental
> for system-wide PM and it is not sufficient to look at the PM-runtime
> status to find out, but if the PM-core itself disables PM-runtime
> (which is has to do at one point to prevent PM-runtime from racing
> with system-wide PM), it is hard to answer definitely in general.
> 
> IMO the only way to make it possible to find that out in all cases is
> to make the PM core retain the power.disable_depth value and that can
> be done by making it use a different mechanism to prevent PM-runtime
> callbacks from being run.
> 
> Alternatively, the current PM-runtime status could be "latched" during
> the PM-runtime disable operation if power.disable_depth is 0 (and that
> "latched" value would be initialized to "invalid" in case PM-runtime
> is never enabled).

Which would be something like the patch below (which additionally cleans up
pm_runtime_enable() while at it).

The idea being that if the status was RPM_ACTIVE last time when
power.disable_depth was changing from 0 to 1 and it is still RPM_ACTIVE, it
can be assumed to reflect what happened to the device last time when it was
using PM-runtime.

---
 drivers/base/power/runtime.c |   38 ++++++++++++++++++++------------------
 include/linux/pm.h           |    2 ++
 2 files changed, 22 insertions(+), 18 deletions(-)

Index: linux-pm/drivers/base/power/runtime.c
===================================================================
--- linux-pm.orig/drivers/base/power/runtime.c
+++ linux-pm/drivers/base/power/runtime.c
@@ -744,11 +744,10 @@ static int rpm_resume(struct device *dev
  repeat:
 	if (dev->power.runtime_error)
 		retval = -EINVAL;
-	else if (dev->power.disable_depth == 1 && dev->power.is_suspended
-	    && dev->power.runtime_status == RPM_ACTIVE)
-		retval = 1;
 	else if (dev->power.disable_depth > 0)
-		retval = -EACCES;
+		retval = dev->power.runtime_status == RPM_ACTIVE &&
+			 dev->power.last_status == RPM_ACTIVE ? 1 : -EACCES;
+
 	if (retval)
 		goto out;
 
@@ -1410,8 +1409,10 @@ void __pm_runtime_disable(struct device
 	/* Update time accounting before disabling PM-runtime. */
 	update_pm_runtime_accounting(dev);
 
-	if (!dev->power.disable_depth++)
+	if (!dev->power.disable_depth++) {
 		__pm_runtime_barrier(dev);
+		dev->power.last_status = dev->power.runtime_status;
+	}
 
  out:
 	spin_unlock_irq(&dev->power.lock);
@@ -1428,23 +1429,23 @@ void pm_runtime_enable(struct device *de
 
 	spin_lock_irqsave(&dev->power.lock, flags);
 
-	if (dev->power.disable_depth > 0) {
-		dev->power.disable_depth--;
-
-		/* About to enable runtime pm, set accounting_timestamp to now */
-		if (!dev->power.disable_depth)
-			dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
-	} else {
+	if (!dev->power.disable_depth) {
 		dev_warn(dev, "Unbalanced %s!\n", __func__);
+		goto out;
 	}
 
-	WARN(!dev->power.disable_depth &&
-	     dev->power.runtime_status == RPM_SUSPENDED &&
-	     !dev->power.ignore_children &&
-	     atomic_read(&dev->power.child_count) > 0,
-	     "Enabling runtime PM for inactive device (%s) with active children\n",
-	     dev_name(dev));
+	if (--dev->power.disable_depth > 0)
+		goto out;
+
+	dev->power.last_status = RPM_INVALID;
+	dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
+
+	if (dev->power.runtime_status == RPM_SUSPENDED &&
+	    !dev->power.ignore_children &&
+	    atomic_read(&dev->power.child_count) > 0)
+		dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
 
+out:
 	spin_unlock_irqrestore(&dev->power.lock, flags);
 }
 EXPORT_SYMBOL_GPL(pm_runtime_enable);
@@ -1640,6 +1641,7 @@ EXPORT_SYMBOL_GPL(__pm_runtime_use_autos
 void pm_runtime_init(struct device *dev)
 {
 	dev->power.runtime_status = RPM_SUSPENDED;
+	dev->power.last_status = RPM_INVALID;
 	dev->power.idle_notification = false;
 
 	dev->power.disable_depth = 1;
Index: linux-pm/include/linux/pm.h
===================================================================
--- linux-pm.orig/include/linux/pm.h
+++ linux-pm/include/linux/pm.h
@@ -499,6 +499,7 @@ const struct dev_pm_ops __maybe_unused n
  */
 
 enum rpm_status {
+	RPM_INVALID = -1,
 	RPM_ACTIVE = 0,
 	RPM_RESUMING,
 	RPM_SUSPENDED,
@@ -612,6 +613,7 @@ struct dev_pm_info {
 	unsigned int		links_count;
 	enum rpm_request	request;
 	enum rpm_status		runtime_status;
+	enum rpm_status		last_status;
 	int			runtime_error;
 	int			autosuspend_delay;
 	u64			last_busy;