On Mon, Nov 30, 2020 at 7:50 PM Laurent Pinchart <laurent.pinch...@ideasonboard.com> wrote: > > Hi Rafael, > > On Mon, Nov 30, 2020 at 06:55:57PM +0100, Rafael J. Wysocki wrote: > > On Mon, Nov 30, 2020 at 6:35 PM Laurent Pinchart wrote: > > > On Mon, Nov 30, 2020 at 05:37:52PM +0100, Rafael J. Wysocki wrote: > > > > On Fri, Nov 27, 2020 at 11:16 AM Geert Uytterhoeven wrote: > > > > > On Tue, Nov 10, 2020 at 10:29 AM Zhang Qilong > > > > > <zhangqilo...@huawei.com> wrote: > > > > > > In many case, we need to check return value of pm_runtime_get_sync, > > > > > > but > > > > > > it brings a trouble to the usage counter processing. Many callers > > > > > > forget > > > > > > to decrease the usage counter when it failed, which could resulted > > > > > > in > > > > > > reference leak. It has been discussed a lot[0][1]. So we add a > > > > > > function > > > > > > to deal with the usage counter for better coding. > > > > > > > > > > > > [0]https://lkml.org/lkml/2020/6/14/88 > > > > > > [1]https://patchwork.ozlabs.org/project/linux-tegra/list/?series=178139 > > > > > > Signed-off-by: Zhang Qilong <zhangqilo...@huawei.com> > > > > > > > > > > Thanks for your patch, which is now commit dd8088d5a8969dc2 ("PM: > > > > > runtime: Add pm_runtime_resume_and_get to deal with usage counter") in > > > > > v5.10-rc5. > > > > > > > > > > > --- a/include/linux/pm_runtime.h > > > > > > +++ b/include/linux/pm_runtime.h > > > > > > @@ -386,6 +386,27 @@ static inline int pm_runtime_get_sync(struct > > > > > > device *dev) > > > > > > return __pm_runtime_resume(dev, RPM_GET_PUT); > > > > > > } > > > > > > > > > > > > +/** > > > > > > + * pm_runtime_resume_and_get - Bump up usage counter of a device > > > > > > and resume it. > > > > > > + * @dev: Target device. > > > > > > + * > > > > > > + * Resume @dev synchronously and if that is successful, increment > > > > > > its runtime > > > > > > + * PM usage counter. Return 0 if the runtime PM usage counter of > > > > > > @dev has been > > > > > > + * incremented or a negative error code otherwise. > > > > > > + */ > > > > > > +static inline int pm_runtime_resume_and_get(struct device *dev) > > > > > > > > > > Perhaps this function should be called > > > > > pm_runtime_resume_and_get_sync(), > > > > > > > > No, really. > > > > > > > > I might consider calling it pm_runtime_acquire(), and adding a > > > > matching _release() as a pm_runtime_get() synonym for that matter, but > > > > not the above. > > > > > > pm_runtime_acquire() seems better to me too. Would pm_runtime_release() > > > would be an alias for pm_runtime_put() ? > > > > Yes. This covers all of the use cases relevant for drivers AFAICS. > > > > > We would also likely need a pm_runtime_release_autosuspend() too then. > > > > Why would we? > > > > > But on that topic, I was wondering, is there a reason we can't select > > > autosuspend behaviour automatically when autosuspend is enabled ? > > > > That is the case already. > > > > pm_runtime_put() will autosuspend if enabled and the usage counter is > > 0, as long as ->runtime_idle() returns 0 (or is absent). > > > > pm_runtime_put_autosuspend() is an optimization allowing > > ->runtime_idle() to be skipped entirely, but I'm wondering how many > > users really need that. > > Ah, I didn't know that, that's good to know. We then don't need > pm_runtime_release_autosuspend() (unless the optimization really makes a > big difference). > > Should I write new drievr code with pm_runtime_put() instead of > pm_runtime_put_autosuspend() ?
If you don't have ->runtime_idle() in the driver (and in the bus type generally speaking, but none of them provide it IIRC), pm_runtime_put() is basically equivalent to pm_runtime_put_autosuspend() AFAICS, except for some extra checks done by the former. Otherwise it all depends on what the ->runtime_idle() callback does, but it is hard to imagine a practical use case when the difference would be really meaningful. > I haven't found clear guidelines on this in the documentation. Yes, that's one of the items I need to take care of.
