On Thu, May 28, 2020 at 04:55:19PM +1000, Herbert Xu wrote: > On Wed, May 20, 2020 at 12:45:56PM -0400, stern@xxxxxxxxxxxxxxxxxxx wrote: > > On Wed, May 20, 2020 at 03:42:17PM +0000, Sverdlin, Alexander (Nokia - DE/Ulm) wrote: > > > Hello Dinghao, > > > > > > On Wed, 2020-05-20 at 21:29 +0800, Dinghao Liu wrote: > > > > pm_runtime_get_sync() increments the runtime PM usage counter even > > > > the call returns an error code. Thus a pairing decrement is needed > > > > on the error handling path to keep the counter balanced. > > > > > > I believe, this is the wrong place for such kind of fix. > > > pm_runtime_get_sync() has obviously a broken semantics with regards to > > > your observation but no other driver does what you propose. > > > > Look again. For example, see what usb_autoresume_device() in > > drivers/usb/core/driver.c does. > > However, there seems to be some disagreement as to what to do > when pm_runtime_get_sync fails. Your driver chooses to call > put_sync while others prefer pm_runtime_put_noidle (e.g., see > drivers/base/power/runtime.c). It doesn't matter which function gets called; in these circumstances (device still suspended or in an error state because a resume attempt failed) they will do the same thing. > This API does seem to be in a bit of a mess. Suggestions and patches are welcome. Alan Stern
