PM: sleep: core: Clear power.must_resume in noirq suspend error path

[ Upstream commit be82483d1b60baf6747884bd74cb7de484deaf76 ]

If system suspend is aborted in the "noirq" phase (for instance, due to
an error returned by one of the device callbacks), power.is_noirq_suspended
will not be set for some devices and device_resume_noirq() will return
early for them.  Consequently, noirq resume callbacks will not run for
them at all because the noirq suspend callbacks have not run for them
yet.

If any of them has power.must_resume set and late suspend has been
skipped for it (due to power.smart_suspend), early resume should be
skipped for it either, or its state may become inconsistent (for
instance, if the early resume assumes that it will always follow
noirq resume).

Make that happen by clearing power.must_resume in device_resume_noirq()
for devices with power.is_noirq_suspended clear that have been left in
suspend by device_suspend_late(), which will subsequently cause
device_resume_early() to leave the device in suspend and avoid
changing its state.

Fixes: 0d4b54c6fee8 ("PM / core: Add LEAVE_SUSPENDED driver flag")
Link: https://lore.kernel.org/linux-pm/5d692b81-6f58-4e86-9cb0-ede69a09d799@rowland.harvard.edu/
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://patch.msgid.link/3381776.aeNJFYEL58@rafael.j.wysocki
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
index baa31194cf20d5ad9fe501ed5547fa68f89f0971..ef5157fc8dcc511dd2b491c00d6e43ecd55b91d9 100644 (file)
--- a/drivers/base/power/main.c
+++ b/drivers/base/power/main.c
@@ -600,8 +600,20 @@ static void __device_resume_noirq(struct device *dev, pm_message_t state, bool a
        if (dev->power.syscore || dev->power.direct_complete)
                goto Out;
 
-       if (!dev->power.is_noirq_suspended)
+       if (!dev->power.is_noirq_suspended) {
+               /*
+                * This means that system suspend has been aborted in the noirq
+                * phase before invoking the noirq suspend callback for the
+                * device, so if device_suspend_late() has left it in suspend,
+                * device_resume_early() should leave it in suspend either in
+                * case the early resume of it depends on the noirq resume that
+                * has not run.
+                */
+               if (dev_pm_skip_suspend(dev))
+                       dev->power.must_resume = false;
+
                goto Out;
+       }
 
        if (!dpm_wait_for_superior(dev, async))
                goto Out;