On Tue, Oct 18, 2016 at 03:30:45PM +0200, Daniel Wagner wrote:
> On 10/10/2016 10:37 PM, Luis R. Rodriguez wrote:
> > > fw_get_fileystem_firmware()
> > > fw_finish_direct_load()
> > > complete_all()
> > >
> > >
> > > 2nd request (waiter context)
> > >
> > > _request_firmware()
> > > _request_firmware_prepare()
> > > fw_lookup_allocate_buf() # finds previously allocated buf
> > > # returns 1 -> wait for loading
> > > sync_cached_firmware_buf()
> > > wait_for_completion_interruptible_timeout()
> > No, that's wait_for_completion_interruptible() not
> > wait_for_completion_interruptible_timeout()
> I confused that one from _request_firmware_load().
Right and wait_for_completion_interruptible() has no timeout.
> > Also note that we only call sync_cached_firmware_buf()
> > *iff* fw_lookup_and_allocate_buf() returned the 1 -- I mentioned
> > when this happens above. That happens only if we already had the entry on
> > the fw cache. As it stands -- concurrent calls against the same fw name
> > could cause a clash here, as such, the wait_for_completion_interruptible()
> > is indeed still needed.
> > Further optimizations can be considered later but for indeed, agreed
> > that completion is needed even for the direct fw load case. The timeout
> > though, I don't see a reason for it.
> So I think I found the source of the confusion about fw_umh_wait_timeout().
> When providing a timeout value of 0 you get the
> wait_for_completion_interruptible() version.
I fail to see that, how so? Note that 0 does is not allowed anyway:
static inline long firmware_loading_timeout(void)
return loading_timeout > 0 ? loading_timeout * HZ : MAX_JIFFY_OFFSET;