On Tue, May 22, 2018 at 08:01:29PM +0200, Rafael J. Wysocki wrote: > On 5/22/2018 11:32 AM, Mark Brown wrote: > > That's still really confusing and surprising, it seems like if the > > driver wants to manually call the callbacks there should be a specific > > API that it can use to get that behaviour rather than requiring every > > user to know that this might happen and manually handle it. > That's a design choice made a long time a go and sorry for it turning out to > be inconvenient. It's not so much that it's inconvenient as that it's a really surprising and unusual which makes it error prone - both in terms of people not being aware of the behaviour and in terms of the expected handling looking wrong. > However, there are pieces of code in the kernel that don't check the return > value of pm_runtime_get_sync() and then call pm_runtime_put() (or > equivalent) after that unconditionally. They all would need to be changed to > do the latter only if the preceding pm_runtime_get_sync() was successful. This does mean that we're encouraging code that ignores errors which doesn't seem ideal. On the other hand trying to change anything at this point might be more trouble than it's worth if there's too many users, perhaps it's better to just try to document this more.
Attachment:
signature.asc
Description: PGP signature
