[PATCH] drm/i915: clear fencing tracking state when retiring request

[PATCH] drm/i915: clear fencing tracking state when retiring request

[Daniel Vetter]

This fixes a regression introduce in

commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Mar 21 10:48:18 2012 +0000

    drm/i915: Mark untiled BLT commands as fenced on gen2/3

which fixed fencing tracking for untiled blt commands.

A side effect of that patch was that now also untiled objects have a
non-zero obj->last_fenced_seqno to track when a fence can be set up
after a pipelined tiling change. Unfortunately this was only cleared
by the fence setup and teardown code, resulting in tons of untiled but
inactive objects with non-zero last_fenced_seqno.

Now after resume we completely reset the seqno tracking, both on the
driver side (by setting dev_priv->next_seqno = 1) and on the hw side
(by allocating a new hws page, which contains the seqnos). Hilarity
and indefinite waits ensued from the stale seqnos in
obj->last_fenced_seqno from before the suspend.

The fix is to properly clear the fencing tracking state like we
already do for the normal gpu rendering while moving objects off the
active list.

Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jiri Slaby <jslaby@suse.cz>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/i915/i915_gem.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 71934dd..cf50fbc 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1415,6 +1415,7 @@ i915_gem_object_move_off_active(struct drm_i915_gem_object *obj)
 {
 	list_del_init(&obj->ring_list);
 	obj->last_rendering_seqno = 0;
+	obj->last_fenced_seqno = 0;
 }
 
 static void
@@ -1443,6 +1444,7 @@ i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
 	BUG_ON(!list_empty(&obj->gpu_write_list));
 	BUG_ON(!obj->active);
 	obj->ring = NULL;
+	obj->last_fenced_ring = NULL;
 
 	i915_gem_object_move_off_active(obj);
 	obj->fenced_gpu_access = false;
-- 
1.7.9.1

[PATCH] drm/i915: clear fencing tracking state when retiring requests

[Daniel Vetter]

This fixes a regression uncovered by

commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Mar 21 10:48:18 2012 +0000

    drm/i915: Mark untiled BLT commands as fenced on gen2/3

which fixed fencing tracking for untiled blt commands.

A side effect of that patch was that now also untiled objects have a
non-zero obj->last_fenced_seqno to track when a fence can be set up
after a pipelined tiling change. Unfortunately this was only cleared
by the fence setup and teardown code, resulting in tons of untiled but
inactive objects with non-zero last_fenced_seqno.

Now after resume we completely reset the seqno tracking, both on the
driver side (by setting dev_priv->next_seqno = 1) and on the hw side
(by allocating a new hws page, which contains the seqnos). Hilarity
and indefinite waits ensued from the stale seqnos in
obj->last_fenced_seqno from before the suspend.

The fix is to properly clear the fencing tracking state like we
already do for the normal gpu rendering while moving objects off the
active list.

Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jiri Slaby <jslaby@suse.cz>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/i915/i915_gem.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 71934dd..cf50fbc 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1415,6 +1415,7 @@ i915_gem_object_move_off_active(struct drm_i915_gem_object *obj)
 {
 	list_del_init(&obj->ring_list);
 	obj->last_rendering_seqno = 0;
+	obj->last_fenced_seqno = 0;
 }
 
 static void
@@ -1443,6 +1444,7 @@ i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
 	BUG_ON(!list_empty(&obj->gpu_write_list));
 	BUG_ON(!obj->active);
 	obj->ring = NULL;
+	obj->last_fenced_ring = NULL;
 
 	i915_gem_object_move_off_active(obj);
 	obj->fenced_gpu_access = false;
-- 
1.7.9.1

Re: [Intel-gfx] [PATCH] drm/i915: clear fencing tracking state when retiring request

[Chris Wilson]

On Thu, 12 Apr 2012 01:27:57 +0200, Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> This fixes a regression introduce in
s/introduce/introduced/
 
> commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
> Author: Chris Wilson <chris@chris-wilson.co.uk>
> Date:   Wed Mar 21 10:48:18 2012 +0000
> 
>     drm/i915: Mark untiled BLT commands as fenced on gen2/3
> 
> which fixed fencing tracking for untiled blt commands.
> 
> A side effect of that patch was that now also untiled objects have a
> non-zero obj->last_fenced_seqno to track when a fence can be set up
> after a pipelined tiling change. Unfortunately this was only cleared
> by the fence setup and teardown code, resulting in tons of untiled but
> inactive objects with non-zero last_fenced_seqno.
> 
> Now after resume we completely reset the seqno tracking, both on the
> driver side (by setting dev_priv->next_seqno = 1) and on the hw side
> (by allocating a new hws page, which contains the seqnos). Hilarity
> and indefinite waits ensued from the stale seqnos in
> obj->last_fenced_seqno from before the suspend.
> 
> The fix is to properly clear the fencing tracking state like we
> already do for the normal gpu rendering while moving objects off the
> active list.
> 
> Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
> Cc: Jiri Slaby <jslaby@suse.cz>
> Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

I spent sometime discussing whether or not we could hit a similar bug
with a well placed change of tiling after resume, and the outcome is
that as the fences are reset during freeze then all tiled objects that
had been used for rendering would have been flushed (and their
last_fenced_seqno set to 0).

So this is a new regression caused by the aforementioned patch and this
is the cleanest fix,
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [Intel-gfx] [PATCH] drm/i915: clear fencing tracking state when retiring request

[Daniel Vetter]

On Thu, Apr 12, 2012 at 12:38:09AM +0100, Chris Wilson wrote:
> On Thu, 12 Apr 2012 01:27:57 +0200, Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> > This fixes a regression introduce in
> s/introduce/introduced/
>  
> > commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
> > Author: Chris Wilson <chris@chris-wilson.co.uk>
> > Date:   Wed Mar 21 10:48:18 2012 +0000
> > 
> >     drm/i915: Mark untiled BLT commands as fenced on gen2/3
> > 
> > which fixed fencing tracking for untiled blt commands.
> > 
> > A side effect of that patch was that now also untiled objects have a
> > non-zero obj->last_fenced_seqno to track when a fence can be set up
> > after a pipelined tiling change. Unfortunately this was only cleared
> > by the fence setup and teardown code, resulting in tons of untiled but
> > inactive objects with non-zero last_fenced_seqno.
> > 
> > Now after resume we completely reset the seqno tracking, both on the
> > driver side (by setting dev_priv->next_seqno = 1) and on the hw side
> > (by allocating a new hws page, which contains the seqnos). Hilarity
> > and indefinite waits ensued from the stale seqnos in
> > obj->last_fenced_seqno from before the suspend.
> > 
> > The fix is to properly clear the fencing tracking state like we
> > already do for the normal gpu rendering while moving objects off the
> > active list.
> > 
> > Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
> > Cc: Jiri Slaby <jslaby@suse.cz>
> > Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> 
> I spent sometime discussing whether or not we could hit a similar bug
> with a well placed change of tiling after resume, and the outcome is
> that as the fences are reset during freeze then all tiled objects that
> had been used for rendering would have been flushed (and their
> last_fenced_seqno set to 0).
> 
> So this is a new regression caused by the aforementioned patch and this
> is the cleanest fix,
> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Picked up for -fixes, thanks for the review.
-Daniel
-- 
Daniel Vetter
Mail: daniel@ffwll.ch
Mobile: +41 (0)79 365 57 48