Hi Philip, > I think both cases are problematic. It doesn't matter which order you > do the writes, if you attempt to cache the value written to the > register, then you must do that atomically. Otherwise, the following > can potentially occur and lead to the wrong result: > > Thread A Thread B > Write first value > Write first value > Write second value > Write second value > > End result is second value is Thread A and first value is Thread B. > > Without locking, this sort of sequence is possible. As long as you have > to update two values in different locations, you've got to have some > kind of locking or synchronization. My original case was different as it had only one value being written, but you got point in that there *are* cases where we actually write to several registers. All in all, I think we made it clear that all sysfs callbacks which write to a register and update the cache need locking. If a few ones don't need it for obscure reasons, it's probably not a problem to still lock there, as this will prevent the introduction of race conditions in the future for a very small price. Thanks, -- Jean Delvare
