On Thu, 20 Sep 2012 09:49:19 +0200, Daniel Vetter <daniel at ffwll.ch> wrote:
> On Wed, Sep 19, 2012 at 04:08:51PM +0100, Chris Wilson wrote:
> > In commit 69c2fc891343cb5217c866d10709343cff190bdc
> > Author: Chris Wilson <chris at chris-wilson.co.uk>
> > Date: Fri Jul 20 12:41:03 2012 +0100
> >
> > drm/i915: Remove the per-ring write list
> >
> > the explicit flush was removed from i915_ring_idle(). However, we
> > continued to wait upon the next seqno which now did not correspond to
> > any request (except for the unusual condition of a failure to queue a
> > request after execbuffer) and so would wait indefinitely.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>
> Afaict gem_next_request_seqno sets ring->olr and i915_wait_seqno does
> check whether olr is set and then adds the request - which for ring_idle is
> pretty much guaranteed to be missing ;-)
Yeah, ok, I'll accept that this is not the root cause of the issue.
However, I'm convinced with the merits of the patch for not adding a
request to the ring everytime we idle.
> So tricky code, but I can't see the bug (and since both module unload and
> suspend works, it'd be surprised if there is one). What am I missing here?
Obviously you haven't encountered the impossible indefinite wait in
__wait_seqno(). It's fairly sporadic, but is an eater of machines... Ah,
what about a GPU reset+recovery leaving ring->outstanding_lazy_seqno set
but reseting dev_priv->next_seqno to 1.
Which is happily fixed up by this patch only looking at existing
requests!
diff --git a/drivers/gpu/drm/i915/i915_gem.c
b/drivers/gpu/drm/i915/i915_gem.c
index 39de523..2286e42 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2189,6 +2189,7 @@ static void i915_gem_reset_ring_lists(struct drm_i915_priv
i915_gem_request_remove_from_client(request);
kfree(request);
}
+ ring->outstanding_lazy_request = 0;
while (!list_empty(&ring->active_list)) {
struct drm_i915_gem_object *obj;
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
