On Tue, Feb 19, 2019 at 1:10 PM Ulf Hansson <ulf.hans...@linaro.org> wrote:
>
> On Mon, 18 Feb 2019 at 23:09, Rafael J. Wysocki <r...@rjwysocki.net> wrote:
> >
> > From: Rafael J. Wysocki <rafael.j.wyso...@intel.com>
> >
> > Commit 4c06c4e6cf63 ("driver core: Fix possible supplier PM-usage
> > counter imbalance") introduced a regression that causes suppliers
> > to be suspended prematurely for device links added during consumer
> > driver probe if the initial PM-runtime status of the consumer is
> > "suspended" and the consumer is resumed after adding the link and
> > before pm_runtime_put_suppliers() is called. In that case,
> > pm_runtime_put_suppliers() will drop the rpm_active refcount for
> > the link by one and (since rpm_active is equal to two after the
> > preceding consumer resume) the supplier's PM-runtime usage counter
> > will be decremented, which may cause the supplier to suspend even
> > though the consumer's PM-runtime status is "active".
>
> Just to confirm, this is the behavior that I see as well.
>
> However, I am a bit a surprised on my side, that I did not run this
> particular test sequence earlier. Sorry about that!
>
> >
> > For this reason, partially revert commit 4c06c4e6cf63 as the problem
> > it tried to fix needs to be addressed somewhat differently, and
> > change pm_runtime_get_suppliers() and pm_runtime_put_suppliers() so
> > that the latter only drops rpm_active references acquired by the
> > former. [This requires adding a new field to struct device_link,
> > but I coulnd't find a cleaner way to address the issue that would
> > work in all cases.]
> >
> > Fixes: 4c06c4e6cf63 ("driver core: Fix possible supplier PM-usage counter
> > imbalance")
> > Reported-by: Jon Hunter <jonath...@nvidia.com>
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wyso...@intel.com>
>
> Besides validating the fix, I can confirm device link add/removal
> during probe works fine, as before.
>
> I also tested the runtime PM autosuspend (several seconds timeout)
> together with DL_FLAG_RPM_ACTIVE - it's working fine! During this
> test, the rpm_active count reaches 3 after probe and the supplier's
> usage count reaches 2. When the consumer becomes runtime suspended,
> after the timeout, these "counts" becomes decremented to 1 and 0, so
> the supplier also becomes runtime suspended.
>
> Tested-by: Ulf Hansson <ulf.hans...@linaro.org>
> Reviewed-by: Ulf Hansson <ulf.hans...@linaro.org>
Many thanks to you and Jon for verifying it!
