| Message ID | 1342352064-6749-1-git-send-email-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Sun, Jul 15, 2012 at 12:34:22PM +0100, Chris Wilson wrote: > The issue is that we stale data in the CPU caches, when we come to > swap-out the object, the CPU may short-circuit the reads from those > cacheline and so corrupt the context object. > > Secondary, leaving the context object as being marked in the CPU write > domain whilst on the GPU active list is a bad idea and will throw > warnings later. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Reviewed-by: Ben Widawsky <ben@bwidawsk.net> > --- > drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c > index 9ae3f2c..fd978bb 100644 > --- a/drivers/gpu/drm/i915/i915_gem_context.c > +++ b/drivers/gpu/drm/i915/i915_gem_context.c > @@ -374,6 +374,13 @@ static int do_switch(struct drm_i915_gem_object *from_obj, > if (ret) > return ret; > > + /* Clear this page out of any CPU caches for coherent swap-in/out */ > + ret = i915_gem_object_set_to_gtt_domain(to->obj, false); > + if (ret) { > + i915_gem_object_unpin(to->obj); > + return ret; > + } > + Do I understand things correctly that thanks to clever use of set_to_gtt_domain with write = false and not setting a write_domain when moving the old context object to the active list we won't block? If so, I'll add a short note to that effect to the commit message when merging. And given that we have similar clever interface abuse in the pwrite/pread code, some code rework is in order I guess. We could replace all that domain tracking with bool gpu_coherent would be simpler (after the flushing_list's permanent demise, of course). -Daniel > if (!to->obj->has_global_gtt_mapping) > i915_gem_gtt_bind_object(to->obj, to->obj->cache_level); > > -- > 1.7.10.4 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
On Sun, 15 Jul 2012 17:16:34 +0200, Daniel Vetter <daniel@ffwll.ch> wrote: > On Sun, Jul 15, 2012 at 12:34:22PM +0100, Chris Wilson wrote: > > The issue is that we stale data in the CPU caches, when we come to > > swap-out the object, the CPU may short-circuit the reads from those > > cacheline and so corrupt the context object. > > > > Secondary, leaving the context object as being marked in the CPU write > > domain whilst on the GPU active list is a bad idea and will throw > > warnings later. > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Reviewed-by: Ben Widawsky <ben@bwidawsk.net> > > --- > > drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c > > index 9ae3f2c..fd978bb 100644 > > --- a/drivers/gpu/drm/i915/i915_gem_context.c > > +++ b/drivers/gpu/drm/i915/i915_gem_context.c > > @@ -374,6 +374,13 @@ static int do_switch(struct drm_i915_gem_object *from_obj, > > if (ret) > > return ret; > > > > + /* Clear this page out of any CPU caches for coherent swap-in/out */ > > + ret = i915_gem_object_set_to_gtt_domain(to->obj, false); > > + if (ret) { > > + i915_gem_object_unpin(to->obj); > > + return ret; > > + } > > + > > Do I understand things correctly that thanks to clever use of > set_to_gtt_domain with write = false and not setting a write_domain when > moving the old context object to the active list we won't block? If so, > I'll add a short note to that effect to the commit message when merging. Worse, it was accidentally very clever. :( Definitely needs a comment and in the future a new function to dtrt intentionally! -Chris
On Sun, Jul 15, 2012 at 08:09:36PM +0100, Chris Wilson wrote: > On Sun, 15 Jul 2012 17:16:34 +0200, Daniel Vetter <daniel@ffwll.ch> wrote: > > On Sun, Jul 15, 2012 at 12:34:22PM +0100, Chris Wilson wrote: > > > The issue is that we stale data in the CPU caches, when we come to > > > swap-out the object, the CPU may short-circuit the reads from those > > > cacheline and so corrupt the context object. > > > > > > Secondary, leaving the context object as being marked in the CPU write > > > domain whilst on the GPU active list is a bad idea and will throw > > > warnings later. > > > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > > Reviewed-by: Ben Widawsky <ben@bwidawsk.net> > > > --- > > > drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++++ > > > 1 file changed, 7 insertions(+) > > > > > > diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c > > > index 9ae3f2c..fd978bb 100644 > > > --- a/drivers/gpu/drm/i915/i915_gem_context.c > > > +++ b/drivers/gpu/drm/i915/i915_gem_context.c > > > @@ -374,6 +374,13 @@ static int do_switch(struct drm_i915_gem_object *from_obj, > > > if (ret) > > > return ret; > > > > > > + /* Clear this page out of any CPU caches for coherent swap-in/out */ > > > + ret = i915_gem_object_set_to_gtt_domain(to->obj, false); > > > + if (ret) { > > > + i915_gem_object_unpin(to->obj); > > > + return ret; > > > + } > > > + > > > > Do I understand things correctly that thanks to clever use of > > set_to_gtt_domain with write = false and not setting a write_domain when > > moving the old context object to the active list we won't block? If so, > > I'll add a short note to that effect to the commit message when merging. > > Worse, it was accidentally very clever. :( > > Definitely needs a comment and in the future a new function to dtrt > intentionally! Queued for -next (with comments to explain this trickery added as discussed on irc), thanks for the patch. I'll pick up the other two patches once Ben has smashed an r-b onto them. -Daniel
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index 9ae3f2c..fd978bb 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -374,6 +374,13 @@ static int do_switch(struct drm_i915_gem_object *from_obj, if (ret) return ret; + /* Clear this page out of any CPU caches for coherent swap-in/out */ + ret = i915_gem_object_set_to_gtt_domain(to->obj, false); + if (ret) { + i915_gem_object_unpin(to->obj); + return ret; + } + if (!to->obj->has_global_gtt_mapping) i915_gem_gtt_bind_object(to->obj, to->obj->cache_level);