| Message ID | 1479276676-4060-1-git-send-email-min.he@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Nov 16, 2016 at 02:11:16PM +0800, Min He wrote: > For a singl_port_submission context, it can only be submitted to port 0, > and there shouldn't be any other context in port 1 at the same time. > This patch is to implement the correct logic in execlists_dequeue. There's a simpler fix. (Other than fixing the root cause of the bug in gvt.) -Chris
> -----Original Message----- > From: Chris Wilson [mailto:chris@chris-wilson.co.uk] > Sent: Wednesday, November 16, 2016 2:52 PM > To: He, Min <min.he@intel.com> > Cc: intel-gfx@lists.freedesktop.org > Subject: Re: [Intel-gfx] [PATCH] drm/i915: fix the dequeue logic for > single_port_submission context > > On Wed, Nov 16, 2016 at 02:11:16PM +0800, Min He wrote: > > For a singl_port_submission context, it can only be submitted to port 0, > > and there shouldn't be any other context in port 1 at the same time. > > This patch is to implement the correct logic in execlists_dequeue. > > There's a simpler fix. (Other than fixing the root cause of the bug in > gvt.) Can you describe this simpler fix? Thanks. > -Chris > > -- > Chris Wilson, Intel Open Source Technology Centre
On Wed, Nov 16, 2016 at 07:03:42AM +0000, He, Min wrote: > > > -----Original Message----- > > From: Chris Wilson [mailto:chris@chris-wilson.co.uk] > > Sent: Wednesday, November 16, 2016 2:52 PM > > To: He, Min <min.he@intel.com> > > Cc: intel-gfx@lists.freedesktop.org > > Subject: Re: [Intel-gfx] [PATCH] drm/i915: fix the dequeue logic for > > single_port_submission context > > > > On Wed, Nov 16, 2016 at 02:11:16PM +0800, Min He wrote: > > > For a singl_port_submission context, it can only be submitted to port 0, > > > and there shouldn't be any other context in port 1 at the same time. > > > This patch is to implement the correct logic in execlists_dequeue. > > > > There's a simpler fix. (Other than fixing the root cause of the bug in > > gvt.) > Can you describe this simpler fix? Thanks. Instead of changing the flow of the logic, add the break inside the cannot merge loop. Then you don't have multiple "cannot merge" questions. Also, my preferred solution would be to remove the gvt hacks that are impacting other users. -Chris
> -----Original Message----- > From: Chris Wilson [mailto:chris@chris-wilson.co.uk] > Sent: Wednesday, November 16, 2016 3:07 PM > To: He, Min <min.he@intel.com> > Cc: intel-gfx@lists.freedesktop.org > Subject: Re: [Intel-gfx] [PATCH] drm/i915: fix the dequeue logic for > single_port_submission context > > On Wed, Nov 16, 2016 at 07:03:42AM +0000, He, Min wrote: > > > > > -----Original Message----- > > > From: Chris Wilson [mailto:chris@chris-wilson.co.uk] > > > Sent: Wednesday, November 16, 2016 2:52 PM > > > To: He, Min <min.he@intel.com> > > > Cc: intel-gfx@lists.freedesktop.org > > > Subject: Re: [Intel-gfx] [PATCH] drm/i915: fix the dequeue logic for > > > single_port_submission context > > > > > > On Wed, Nov 16, 2016 at 02:11:16PM +0800, Min He wrote: > > > > For a singl_port_submission context, it can only be submitted to port 0, > > > > and there shouldn't be any other context in port 1 at the same time. > > > > This patch is to implement the correct logic in execlists_dequeue. > > > > > > There's a simpler fix. (Other than fixing the root cause of the bug in > > > gvt.) > > Can you describe this simpler fix? Thanks. > > Instead of changing the flow of the logic, add the break inside the > cannot merge loop. Then you don't have multiple "cannot merge" > questions. Thanks, will send out a v2 patch. > > Also, my preferred solution would be to remove the gvt hacks that are > impacting other users. > -Chris > > -- > Chris Wilson, Intel Open Source Technology Centre
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index f50feaa..be83e8c 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -424,9 +424,6 @@ static bool can_merge_ctx(const struct i915_gem_context *prev, if (prev != next) return false; - if (ctx_single_port_submission(prev)) - return false; - return true; } @@ -477,6 +474,13 @@ static void execlists_dequeue(struct intel_engine_cs *engine) struct drm_i915_gem_request *cursor = rb_entry(rb, typeof(*cursor), priotree.node); + /* If last ctx is single_submission, it means we can only + * submit this context in port 0, and cannot submit another + * context in port 1 at the same time. So we will break here + * in this situation. + */ + if (last && ctx_single_port_submission(last->ctx)) + break; /* Can we combine this request with the current port? It has to * be the same context/ringbuffer and not have any exceptions * (e.g. GVT saying never to combine contexts).