[PATCH 5/8] sync_file: add support for a semaphore object

[Chris Wilson]

On Thu, Apr 13, 2017 at 06:51:17AM +1000, Dave Airlie wrote:
> On 13 April 2017 at 06:39, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > On Wed, Apr 12, 2017 at 09:01:32PM +0100, Chris Wilson wrote:
> >> On Thu, Apr 13, 2017 at 05:05:27AM +1000, Dave Airlie wrote:
> >> > >>
> >> > >> Not sure what the best semantics are there, any opinions on barring
> >> > >> wakeups/polling on semaphore sync_files, and just punting this
> >> > >> until we need it.
> >> > >
> >> > > I think getting it right now will make writing sw_sync-esque (i.e. cpu)
> >> > > tests easier and more complete.
> >> >
> >> > I just don't have any use case for it, so we would be writing code to
> >> > write tests for it.
> >> >
> >> > That doesn't seem smart.
> >> >
> >> > If there is a future non-testing use case, the API is expressive
> >> > enough for someone
> >> > to add a flag or new sync obj to allow polling and to add support in a
> >> > nice easily
> >> > digestible patch.
> >>
> >> My first thought was to check the signaled status would be to use
> >> poll(0), but that can be retrieved from the sync_file_status ioctl. But
> >> to get that still needs for us to acquire an fd from the syncobj. And if
> >> we were to want check the flag on a driver syncobj, we would need to be
> >> able to export one. That doesn't look very promising...
> >
> > Hmm, you do export fd to pass syncobj between processes. Let's not start
> > with syncobj being a second class sync_file.
> 
> It's not the same semantics as a sync_file. Userspace should never be polling on
> semaphores.
> 
> Semaphores are one process signals, one process waits, no semantics for
> sniffing or use cases. Unless you have a practical use case for an new
> or proposed
> API I don't think we should start with adding functionality.

The problem, as I see it, is that you are taking functionality away from
sync_file. If you are wrapping them up inside a sync_file, we have a
fair expectation that our code to handle sync_files will continue to
work.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre