[Intel-gfx] force wake reference counting (another try)

Chris Wilson chris at chris-wilson.co.uk
Tue Apr 12 10:02:21 CEST 2011


On Mon, 11 Apr 2011 18:01:15 -0700, Ben Widawsky <ben at bwidawsk.net> wrote:
> So once again, I expect this patch to potentially generate a lot of
> warnings, but I consider all of those warnings to be serious bugs for
> SNB.
> 
> If anyone has clever ideas on how to handle this outside of what I've
> already mentioned, please let me know.
> 
> I expect ongoing patches to fix these issues as they come up.

Continuing the general discussion from IRC, we really do need to clarify
which locks we expect to be holding when reading different sets of
registers.  Along with similar documentation over which parts of our
structures are covered by struct_mutex, mode_config.lock and the ensemble
of spinlocks.  This task is not limited to just our driver, but we need to
review the core as well as looking at how to improve the locking overall.
(The clear example that something sucks is that probing outputs causes a
rendering stall (fortunately less noticeable on recent hardware where
hotplug is prevalent and the polling itself is much quicker, but still
present.)

The only concern I have with the extra warnings are if we leave false
positives in the code we submit upstream. Those warnings will quickly
become regression reports and angry users. Alas, no clever ideas.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre



More information about the Intel-gfx mailing list