Skip to contentSkip to conversation

drm: Allow modeset on unregisted connectors unconditionally

PATCH In review Imre Deak 3 participants 12 messages May 20, 2019 – May 21, 2019

Subsystems (2)
lore
Share & tools
lore

Imre Deak

author
#0 lore raw

PATCH 1/1 drm: Allow modeset on unregisted connectors unconditionally

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general. The check for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 6 lines
2.17.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 5 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.
-Daniel

Quoted 59 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 1 line
2.17.1
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author
#2 lore raw
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 16 lines
Quoted 6 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Quoted 53 lines

-Daniel

Quoted 51 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +0 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
Quoted 13 lines
Quoted 8 lines
drivers/gpu/drm/drm_atomic_helper.c +2 -0
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 1 line
2.17.1
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Quoted 20 lines
Quoted 15 lines
Quoted 5 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 67 lines
Quoted 67 lines

-Daniel

Quoted 59 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 1 line
2.17.1
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#4 lore raw
Quoted 23 lines · Daniel Vetter
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 7 lines

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

Right, pageflipping works.

Quoted 10 lines

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Yes, but what is the the actual description of the failing scenario? I can't see how anything can go wrong without this check. The resume time restoration modeset may have to act in the same way on an old connector.

I don't understand how userspace would not notice the connector change. It will get a hotplug uevent in response to which it would have to do a detect which returns to it the updated information about the new MST connector tree.

Quoted 90 lines
Quoted 20 lines
Quoted 15 lines
Quoted 5 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 58 lines
Quoted 58 lines

-Daniel

Quoted 55 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
Quoted 14 lines
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -0
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 1 line
2.17.1
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#5 lore raw
Quoted 45 lines · Imre Deak
Quoted 23 lines · Daniel Vetter
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 7 lines

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

Right, pageflipping works.

Quoted 10 lines

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Quoted 3 lines

Yes, but what is the the actual description of the failing scenario? I can't see how anything can go wrong without this check. The resume time restoration modeset may have to act in the same way on an old connector.

That's what the !state->duplicated thing is meant to check for btw.

Quoted 4 lines

I don't understand how userspace would not notice the connector change. It will get a hotplug uevent in response to which it would have to do a detect which returns to it the updated information about the new MST connector tree.

uevent handling can take positively forever, at which point the new connector could already be plugged in, and then userspace makes a mess aliasing the two since the path property matches. Just needs a wobbly cable. This is the issue with the "fixed" lifetime management, and I'm wary of breaking more stuff.

Other way round: What do we gain if userspace is allowed to turn on a connector again which doesn't exist and will not ever show any pixels? I'm not seeing a benefit here in allowing that. And history says we change something around mst handling, it'll break something somewhere. -Daniel

Quoted 105 lines
Quoted 105 lines
Quoted 21 lines
Quoted 16 lines
Quoted 6 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 58 lines
Quoted 58 lines

-Daniel

Quoted 55 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
                return 0;
        }
 
-       crtc_state = drm_atomic_get_new_crtc_state(state,
-                                                  new_connector_state->crtc);
-       /*
-        * For compatibility with legacy users, we want to make sure that
-        * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-        * which would result in anything else must be considered invalid, to
-        * avoid turning on new displays on dead connectors.
-        *
-        * Since the connector can be unregistered at any point during an
-        * atomic check or commit, this is racy. But that's OK: all we care
-        * about is ensuring that userspace can't do anything but shut off the
-        * display on a connector that was destroyed after it's been notified,
-        * not before.
-        *
-        * Additionally, we also want to ignore connector registration when
-        * we're trying to restore an atomic state during system resume since
-        * there's a chance the connector may have been destroyed during the
-        * process, but it's better to ignore that then cause
-        * drm_atomic_helper_resume() to fail.
-        */
-       if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-           crtc_state->active) {
-               DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-                                connector->base.id, connector->name);
-               return -EINVAL;
-       }
-
        funcs = connector->helper_private;
 
        if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
        set_best_encoder(state, new_connector_state, new_encoder);
 
+       crtc_state = drm_atomic_get_new_crtc_state(state,
+                                                  new_connector_state->crtc);
        crtc_state->connectors_changed = true;
 
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -1
        DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
--

2.17.1

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author
#6 lore raw
Quoted 57 lines · Daniel Vetter
Quoted 45 lines · Imre Deak
Quoted 23 lines · Daniel Vetter
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 7 lines

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

Right, pageflipping works.

Quoted 10 lines

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Quoted 3 lines

Yes, but what is the the actual description of the failing scenario? I can't see how anything can go wrong without this check. The resume time restoration modeset may have to act in the same way on an old connector.

That's what the !state->duplicated thing is meant to check for btw.

Quoted 4 lines

I don't understand how userspace would not notice the connector change. It will get a hotplug uevent in response to which it would have to do a detect which returns to it the updated information about the new MST connector tree.

Quoted 5 lines

uevent handling can take positively forever, at which point the new connector could already be plugged in, and then userspace makes a mess aliasing the two since the path property matches. Just needs a wobbly cable. This is the issue with the "fixed" lifetime management, and I'm wary of breaking more stuff.

Yea, processing can be delayed arbitrarily, the corresponding GETRESOURCES/GETCONNECTOR should still return to the userspace the correct information to do right thing, even if the path property
matches: the connector ID for the old and new connector will be different and the GETCONNECTOR for the old ID will return properly that this old connector is already disconnected (see intel_dp_mst_detect()). So I don't see how userspace could mess up things. Obviously if user space is buggy, then well, it is just buggy and then that user space bug would need to be fixed instead of papering over such problems in the kernel.

Quoted 4 lines

Other way round: What do we gain if userspace is allowed to turn on a connector again which doesn't exist and will not ever show any pixels? I'm not seeing a benefit here in allowing that. And history says we change something around mst handling, it'll break something somewhere.

Let's then describe the actual reason for this check, that description is missing.

We should remove this check to simplify things if there isn't an actual need for it. Userspace can do already a modeset on connectors in general that are not in a connected state. So this special casing - for MST connectors only - is a bad idea if there is no reason for special casing it. If it's there to paper over some user space bug that user space bug should be fixed instead of adding this special casing which adds complexity to the driver.

The only justification for a check in the driver if something could blow up in the kernel without that check. Nothing can blow up in the kernel without this check. Every other failure scenario should be handled/fixed in userspace.

Quoted 103 lines

-Daniel

Quoted 101 lines
Quoted 101 lines
Quoted 21 lines
Quoted 16 lines
Quoted 6 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 67 lines
Quoted 67 lines

-Daniel

Quoted 57 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -28
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
                return 0;
        }
 
-       crtc_state = drm_atomic_get_new_crtc_state(state,
-                                                  new_connector_state->crtc);
-       /*
-        * For compatibility with legacy users, we want to make sure that
-        * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-        * which would result in anything else must be considered invalid, to
-        * avoid turning on new displays on dead connectors.
-        *
-        * Since the connector can be unregistered at any point during an
-        * atomic check or commit, this is racy. But that's OK: all we care
-        * about is ensuring that userspace can't do anything but shut off the
-        * display on a connector that was destroyed after it's been notified,
-        * not before.
-        *
-        * Additionally, we also want to ignore connector registration when
-        * we're trying to restore an atomic state during system resume since
-        * there's a chance the connector may have been destroyed during the
-        * process, but it's better to ignore that then cause
-        * drm_atomic_helper_resume() to fail.
-        */
-       if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-           crtc_state->active) {
-               DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-                                connector->base.id, connector->name);
-               return -EINVAL;
-       }
-
        funcs = connector->helper_private;
 
        if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
        set_best_encoder(state, new_connector_state, new_encoder);
 
+       crtc_state = drm_atomic_get_new_crtc_state(state,
+                                                  new_connector_state->crtc);
        crtc_state->connectors_changed = true;
 
        DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
--
Quoted 2 lines

2.17.1

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Quoted 11 lines
Quoted 4 lines
Quoted 4 lines

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#7 lore raw
Quoted 65 lines · Imre Deak
Quoted 57 lines · Daniel Vetter
Quoted 45 lines · Imre Deak
Quoted 23 lines · Daniel Vetter
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 7 lines

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

Right, pageflipping works.

Quoted 10 lines

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Quoted 3 lines

Yes, but what is the the actual description of the failing scenario? I can't see how anything can go wrong without this check. The resume time restoration modeset may have to act in the same way on an old connector.

That's what the !state->duplicated thing is meant to check for btw.

Quoted 4 lines

I don't understand how userspace would not notice the connector change. It will get a hotplug uevent in response to which it would have to do a detect which returns to it the updated information about the new MST connector tree.

Quoted 5 lines

uevent handling can take positively forever, at which point the new connector could already be plugged in, and then userspace makes a mess aliasing the two since the path property matches. Just needs a wobbly cable. This is the issue with the "fixed" lifetime management, and I'm wary of breaking more stuff.

Quoted 10 lines

Yea, processing can be delayed arbitrarily, the corresponding GETRESOURCES/GETCONNECTOR should still return to the userspace the correct information to do right thing, even if the path property
matches: the connector ID for the old and new connector will be different and the GETCONNECTOR for the old ID will return properly that this old connector is already disconnected (see intel_dp_mst_detect()). So I don't see how userspace could mess up things. Obviously if user space is buggy, then well, it is just buggy and then that user space bug would need to be fixed instead of papering over such problems in the kernel.

Uh that's not how this "kernel change caused a regression" thing works. Dave just dropped a rant about this and i915 recently ...

Quoted 16 lines
Quoted 4 lines

Other way round: What do we gain if userspace is allowed to turn on a connector again which doesn't exist and will not ever show any pixels? I'm not seeing a benefit here in allowing that. And history says we change something around mst handling, it'll break something somewhere.

Let's then describe the actual reason for this check, that description is missing.

We should remove this check to simplify things if there isn't an actual need for it. Userspace can do already a modeset on connectors in general that are not in a connected state. So this special casing - for MST connectors only - is a bad idea if there is no reason for special casing it. If it's there to paper over some user space bug that user space bug should be fixed instead of adding this special casing which adds complexity to the driver.

Quoted 4 lines

The only justification for a check in the driver if something could blow up in the kernel without that check. Nothing can blow up in the kernel without this check. Every other failure scenario should be handled/fixed in userspace.

So I thought you need this because it's a pain on resume, but that should be handled with the state->duplicated check. I guess I still don't understand why you want to nuke this check, it seemed like it's getting in the way somewhere.

If it's not getting in the way then imo let's not touch stuff, I'm too much burned child in this area.
-Daniel

Quoted 114 lines
Quoted 114 lines

-Daniel

Quoted 101 lines
Quoted 101 lines
Quoted 21 lines
Quoted 16 lines
Quoted 6 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 67 lines
Quoted 67 lines

-Daniel

Quoted 57 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -28
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
                return 0;
        }
 
-       crtc_state = drm_atomic_get_new_crtc_state(state,
-                                                  new_connector_state->crtc);
-       /*
-        * For compatibility with legacy users, we want to make sure that
-        * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-        * which would result in anything else must be considered invalid, to
-        * avoid turning on new displays on dead connectors.
-        *
-        * Since the connector can be unregistered at any point during an
-        * atomic check or commit, this is racy. But that's OK: all we care
-        * about is ensuring that userspace can't do anything but shut off the
-        * display on a connector that was destroyed after it's been notified,
-        * not before.
-        *
-        * Additionally, we also want to ignore connector registration when
-        * we're trying to restore an atomic state during system resume since
-        * there's a chance the connector may have been destroyed during the
-        * process, but it's better to ignore that then cause
-        * drm_atomic_helper_resume() to fail.
-        */
-       if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-           crtc_state->active) {
-               DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-                                connector->base.id, connector->name);
-               return -EINVAL;
-       }
-
        funcs = connector->helper_private;
 
        if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
        set_best_encoder(state, new_connector_state, new_encoder);
 
+       crtc_state = drm_atomic_get_new_crtc_state(state,
+                                                  new_connector_state->crtc);
        crtc_state->connectors_changed = true;
 
        DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
--
Quoted 2 lines

2.17.1

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Quoted 4 lines
Quoted 4 lines

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author
#8 lore raw
Quoted 77 lines · Daniel Vetter
Quoted 65 lines · Imre Deak
Quoted 57 lines · Daniel Vetter
Quoted 45 lines · Imre Deak
Quoted 23 lines · Daniel Vetter
Quoted 13 lines · Imre Deak
Quoted 11 lines · Daniel Vetter
Quoted 6 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general.

I'm not parsing this at all ... maybe references to the commits that fix this? Or do you mean the refcounting work for all the things hanging of connectors, including the entire mst tree?

Quoted 8 lines

Yes the check was added to solve the issue related to the removal of MST connectors that could happen asynchronously wrt. a modeset referring to that MST connector. That could happen since the MST core doesn't hold any locks (for instance the connection_mutex) during removing an MST connector that would prevent doing a modeset at the same time.

Adding the refcounting for such MST connectors (via the drm_connector_get()/drm_connector_put()) got rid of the above problem.

Quoted 7 lines

We added the check way after that stuff landed. Before all the connector reworking connectors were forcefully disabled by the kernel. The idea behind this check is to make sure that that userspace notices a connector is gone (only thing that's not allowed is enabling it, you can keep pageflipping). I think we've always had behaviour like ever since mst (all userspace has some "oops mst connector probably gone" failure catching around modesets).

Right, pageflipping works.

Quoted 10 lines

So no idea what all blows up if we stop catching userspace this way.

Now very much possible I'm getting all this wrong again or missing something, this stuff is often way over my head. But I'm really vary of breaking userspace here. E.g. just the drm_connector_get/put lifetime changes results in some userspace breaking if you unplug/replug fast enough, because then it doesn't notice the connector change anymore. I couldn't figure out a way to paper over that regression without reintroduce the rather broken and oops-prone old connector lifetime management.

Quoted 3 lines

Yes, but what is the the actual description of the failing scenario? I can't see how anything can go wrong without this check. The resume time restoration modeset may have to act in the same way on an old connector.

That's what the !state->duplicated thing is meant to check for btw.

Quoted 4 lines

I don't understand how userspace would not notice the connector change. It will get a hotplug uevent in response to which it would have to do a detect which returns to it the updated information about the new MST connector tree.

Quoted 5 lines

uevent handling can take positively forever, at which point the new connector could already be plugged in, and then userspace makes a mess aliasing the two since the path property matches. Just needs a wobbly cable. This is the issue with the "fixed" lifetime management, and I'm wary of breaking more stuff.

Quoted 10 lines

Yea, processing can be delayed arbitrarily, the corresponding GETRESOURCES/GETCONNECTOR should still return to the userspace the correct information to do right thing, even if the path property
matches: the connector ID for the old and new connector will be different and the GETCONNECTOR for the old ID will return properly that this old connector is already disconnected (see intel_dp_mst_detect()). So I don't see how userspace could mess up things. Obviously if user space is buggy, then well, it is just buggy and then that user space bug would need to be fixed instead of papering over such problems in the kernel.

Quoted 2 lines

Uh that's not how this "kernel change caused a regression" thing works. Dave just dropped a rant about this and i915 recently ...

I still don't see what would depend on the current behavior instead of acting on the hotplug uevents properly.

Quoted 21 lines
Quoted 16 lines
Quoted 4 lines

Other way round: What do we gain if userspace is allowed to turn on a connector again which doesn't exist and will not ever show any pixels? I'm not seeing a benefit here in allowing that. And history says we change something around mst handling, it'll break something somewhere.

Let's then describe the actual reason for this check, that description is missing.

We should remove this check to simplify things if there isn't an actual need for it. Userspace can do already a modeset on connectors in general that are not in a connected state. So this special casing - for MST connectors only - is a bad idea if there is no reason for special casing it. If it's there to paper over some user space bug that user space bug should be fixed instead of adding this special casing which adds complexity to the driver.

Quoted 4 lines

The only justification for a check in the driver if something could blow up in the kernel without that check. Nothing can blow up in the kernel without this check. Every other failure scenario should be handled/fixed in userspace.

Quoted 7 lines

So I thought you need this because it's a pain on resume, but that should be handled with the state->duplicated check. I guess I still don't understand why you want to nuke this check, it seemed like it's getting in the way somewhere.

If it's not getting in the way then imo let's not touch stuff, I'm too much burned child in this area.

It's a pain to maintain such papering overs in the kernel, that nobody knows what purpose they serve. What I'm asking here is to preserve this check here only if there is a purpose for it along with a precise description of its purpose. If noone can give an explanation for it - or how it can lead to a regression - then yes, let's remove it to simplify things.

Quoted 112 lines

-Daniel

Quoted 110 lines
Quoted 110 lines

-Daniel

Quoted 105 lines
Quoted 105 lines
Quoted 21 lines
Quoted 16 lines
Quoted 6 lines

The check
for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Yeah this one actually makes sense to me. We could still keep this check here, but for the atomic ioctl only when called from userspace. But iirc Lyude also said she has some plans here, so no idea whether that all fits.

Quoted 4 lines

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines

Not saying we don't need this, but there's fairly enormous amounts of history behind all this stuff, and lots of discussions. Would be good to at least reference those, so we have a good story for when this then all goes wrong again.

Quoted 8 lines

I still don't see why this check is needed. There is no justification for it - besides the original reason for it as discussed above about the refcounting problem, which is solved now - so I think we should remove it, instead of just making it a special case for the user space modeset.

As I wrote a user space modeset can end up anyway doing a modeset on an unregistered connector when the unregistering - by MST core - happens just right after the check.

Yup. Always been like that.
-Daniel

Quoted 58 lines
Quoted 58 lines

-Daniel

Quoted 55 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
                return 0;
        }
 
-       crtc_state = drm_atomic_get_new_crtc_state(state,
-                                                  new_connector_state->crtc);
-       /*
-        * For compatibility with legacy users, we want to make sure that
-        * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-        * which would result in anything else must be considered invalid, to
-        * avoid turning on new displays on dead connectors.
-        *
-        * Since the connector can be unregistered at any point during an
-        * atomic check or commit, this is racy. But that's OK: all we care
-        * about is ensuring that userspace can't do anything but shut off the
-        * display on a connector that was destroyed after it's been notified,
-        * not before.
-        *
-        * Additionally, we also want to ignore connector registration when
-        * we're trying to restore an atomic state during system resume since
-        * there's a chance the connector may have been destroyed during the
-        * process, but it's better to ignore that then cause
-        * drm_atomic_helper_resume() to fail.
-        */
-       if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-           crtc_state->active) {
-               DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-                                connector->base.id, connector->name);
-               return -EINVAL;
-       }
-
        funcs = connector->helper_private;
 
        if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
        set_best_encoder(state, new_connector_state, new_encoder);
 
+       crtc_state = drm_atomic_get_new_crtc_state(state,
+                                                  new_connector_state->crtc);
        crtc_state->connectors_changed = true;
 
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -1
        DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
--

2.17.1

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Quoted 11 lines
Quoted 4 lines
Quoted 4 lines

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

author
#9 lore raw
In reply to Imre Deak
Quoted 16 lines · Imre Deak

We allowed modesetting an unregistered connector only in the case the mode is getting disabled on the connector.

The reason for this check was the lack of proper refcounting for the backing memory objects. That problem has been solved meanwhile so there is no reason any more to reject the modesetting in general. The check for that also makes driver internal modesets more cumbersome where we need to add exemptions for the cases where we do need to allow the modeset even for unregistered connectors. One such case is the restoration of the mode during resume.

Simplify things by removing the unneeded check. I can't see how modesetting an unregistered connector can cause any problem and the race (described in the code comment) can anyway result in such a modeset (if the connector is unregistered right after the check).

Quoted 4 lines
Cc Lyude Paul <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Ville Syrjälä <[email protected]>

Lyude, could you test this change against the DDX fail scenario that the check was added for originally (based on our IRC discussion)?

Thanks,
Imre

Quoted 58 lines
---
 drivers/gpu/drm/drm_atomic_helper.c | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -27
index 2e0cb4246cbd..e94e69483498 100644
@@ -319,33 +319,6 @@ update_connector_routing(struct drm_atomic_state *state,
 		return 0;
 	}
 
-	crtc_state = drm_atomic_get_new_crtc_state(state,
-						   new_connector_state->crtc);
-	/*
-	 * For compatibility with legacy users, we want to make sure that
-	 * we allow DPMS On->Off modesets on unregistered connectors. Modesets
-	 * which would result in anything else must be considered invalid, to
-	 * avoid turning on new displays on dead connectors.
-	 *
-	 * Since the connector can be unregistered at any point during an
-	 * atomic check or commit, this is racy. But that's OK: all we care
-	 * about is ensuring that userspace can't do anything but shut off the
-	 * display on a connector that was destroyed after it's been notified,
-	 * not before.
-	 *
-	 * Additionally, we also want to ignore connector registration when
-	 * we're trying to restore an atomic state during system resume since
-	 * there's a chance the connector may have been destroyed during the
-	 * process, but it's better to ignore that then cause
-	 * drm_atomic_helper_resume() to fail.
-	 */
-	if (!state->duplicated && drm_connector_is_unregistered(connector) &&
-	    crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
-				 connector->base.id, connector->name);
-		return -EINVAL;
-	}
-
 	funcs = connector->helper_private;
 
 	if (funcs->atomic_best_encoder)
@@ -390,6 +363,8 @@ update_connector_routing(struct drm_atomic_state *state,
 
 	set_best_encoder(state, new_connector_state, new_encoder);
 
+	crtc_state = drm_atomic_get_new_crtc_state(state,
+						   new_connector_state->crtc);
 	crtc_state->connectors_changed = true;
 
 	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] using [ENCODER:%d:%s] on [CRTC:%d:%s]\n",
Signature · 6 lines
2.17.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#10 lore raw

✓ Fi.CI.BAT: success for drm: Allow modeset on unregisted connectors unconditionally (rev2)

In reply to Imre Deak
== Series Details ==

== Series Details ==

Series: drm: Allow modeset on unregisted connectors unconditionally (rev2)
URL : https://patchwork.freedesktop.org/series/60871/
State : success

== Summary ==

CI Bug Log - changes from CI_DRM_6110 -> Patchwork_13054
====================================================

Summary
-------

**SUCCESS**

No regressions found.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/

Known issues
------------

Here are the changes found in Patchwork_13054 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@debugfs_test@read_all_entries:
- fi-ilk-650: [PASS][1] -> [DMESG-WARN][2] ([fdo#106387])
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/fi-ilk-650/igt@debugfs_test@read_all_entries.html
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/fi-ilk-650/igt@debugfs_test@read_all_entries.html

#### Possible fixes ####

* igt@kms_chamelium@hdmi-hpd-fast:
- fi-kbl-7500u: [FAIL][3] ([fdo#109485]) -> [PASS][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/fi-kbl-7500u/igt@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/fi-kbl-7500u/igt@[email protected]

#### Warnings ####

* igt@i915_selftest@live_hangcheck:
- fi-apl-guc: [DMESG-FAIL][5] ([fdo#110620]) -> [FAIL][6] ([fdo#110623])
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/fi-apl-guc/igt@i915_selftest@live_hangcheck.html
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/fi-apl-guc/igt@i915_selftest@live_hangcheck.html

[fdo#106387]: https://bugs.freedesktop.org/show_bug.cgi?id=106387
[fdo#109485]: https://bugs.freedesktop.org/show_bug.cgi?id=109485
[fdo#110620]: https://bugs.freedesktop.org/show_bug.cgi?id=110620
[fdo#110623]: https://bugs.freedesktop.org/show_bug.cgi?id=110623

Participating hosts (49 -> 43)
------------------------------

Additional (2): fi-icl-u2 fi-gdg-551
Missing (8): fi-kbl-soraka fi-ilk-m540 fi-hsw-4200u fi-byt-squawks fi-bsw-cyan fi-pnv-d510 fi-byt-clapper fi-bdw-samus

Build changes
-------------

* Linux: CI_DRM_6110 -> Patchwork_13054

CI_DRM_6110: ae35264eb9f700d5f9856552bf98e20ec2b74739 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4999: eee6e9ea7a7e22a64afc029a696b80da8053bc3c @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_13054: 949e61ed7490fb4da9dd29729084af60a1174d4d @ git://anongit.freedesktop.org/gfx-ci/linux

== Linux commits ==

949e61ed7490 drm: Allow modeset on unregisted connectors unconditionally

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#11 lore raw

✓ Fi.CI.IGT: success for drm: Allow modeset on unregisted connectors unconditionally (rev2)

In reply to Imre Deak
== Series Details ==

== Series Details ==

Series: drm: Allow modeset on unregisted connectors unconditionally (rev2)
URL : https://patchwork.freedesktop.org/series/60871/
State : success

== Summary ==

CI Bug Log - changes from CI_DRM_6110_full -> Patchwork_13054_full
====================================================

Summary
-------

**SUCCESS**

No regressions found.

Known issues
------------

Here are the changes found in Patchwork_13054_full that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@gem_create@create-clear:
- shard-hsw: [PASS][1] -> [INCOMPLETE][2] ([fdo#103540] / [fdo#110401])
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-hsw6/igt@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-hsw1/igt@[email protected]

* igt@gem_eio@in-flight-suspend:
- shard-glk: [PASS][3] -> [FAIL][4] ([fdo#110667])
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-glk5/igt@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-glk5/igt@[email protected]

* igt@gem_exec_suspend@basic-s3:
- shard-apl: [PASS][5] -> [DMESG-WARN][6] ([fdo#108566])
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-apl8/igt@[email protected]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-apl4/igt@[email protected]

* igt@gem_mmap_gtt@big-copy:
- shard-snb: [PASS][7] -> [INCOMPLETE][8] ([fdo#105411]) +4 similar issues
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-snb2/igt@[email protected]
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-snb7/igt@[email protected]

* igt@gem_pwrite@big-gtt-fbr:
- shard-kbl: [PASS][9] -> [INCOMPLETE][10] ([fdo#103665]) +5 similar issues
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-kbl7/igt@[email protected]
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-kbl2/igt@[email protected]

* igt@gem_pwrite@small-cpu-backwards:
- shard-glk: [PASS][11] -> [INCOMPLETE][12] ([fdo#103359] / [k.org#198133]) +1 similar issue
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-glk8/igt@[email protected]
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-glk4/igt@[email protected]

* igt@gem_pwrite@small-cpu-random:
- shard-hsw: [PASS][13] -> [INCOMPLETE][14] ([fdo#103540]) +2 similar issues
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-hsw7/igt@[email protected]
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-hsw5/igt@[email protected]

* igt@gem_tiled_swapping@non-threaded:
- shard-glk: [PASS][15] -> [DMESG-WARN][16] ([fdo#108686])
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-glk8/igt@[email protected]
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-glk1/igt@[email protected]

* igt@i915_pm_rpm@reg-read-ioctl:
- shard-skl: [PASS][17] -> [INCOMPLETE][18] ([fdo#107807])
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl9/igt@[email protected]
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl2/igt@[email protected]

* igt@kms_plane@plane-panning-bottom-right-suspend-pipe-c-planes:
- shard-skl: [PASS][19] -> [INCOMPLETE][20] ([fdo#104108])
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl7/igt@[email protected]
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl3/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-b-coverage-7efc:
- shard-skl: [PASS][21] -> [FAIL][22] ([fdo#108145] / [fdo#110403])
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl10/igt@[email protected]
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl1/igt@[email protected]

#### Possible fixes ####

* igt@gem_mmap_gtt@basic-small-bo:
- shard-hsw: [INCOMPLETE][23] ([fdo#103540]) -> [PASS][24] +1 similar issue
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-hsw8/igt@[email protected]
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-hsw6/igt@[email protected]

* igt@gem_mmap_gtt@forked-basic-small-copy-xy:
- shard-kbl: [INCOMPLETE][25] ([fdo#103665]) -> [PASS][26] +3 similar issues
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-kbl4/igt@[email protected]
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-kbl7/igt@[email protected]

* igt@gem_ppgtt@blt-vs-render-ctx0:
- shard-glk: [INCOMPLETE][27] ([fdo#103359] / [k.org#198133]) -> [PASS][28] +3 similar issues
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-glk4/igt@[email protected]
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-glk6/igt@[email protected]
- shard-snb: [INCOMPLETE][29] ([fdo#105411]) -> [PASS][30] +4 similar issues
[29]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-snb7/igt@[email protected]
[30]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-snb4/igt@[email protected]

* igt@i915_suspend@sysfs-reader:
- shard-apl: [DMESG-WARN][31] ([fdo#108566]) -> [PASS][32] +3 similar issues
[31]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-apl4/igt@[email protected]
[32]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-apl4/igt@[email protected]

* igt@kms_busy@basic-flip-a:
- shard-snb: [SKIP][33] ([fdo#109271] / [fdo#109278]) -> [PASS][34]
[33]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-snb2/igt@[email protected]
[34]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-snb7/igt@[email protected]

* {igt@kms_cursor_crc@pipe-a-cursor-suspend}:
- shard-skl: [INCOMPLETE][35] -> [PASS][36]
[35]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl8/igt@[email protected]
[36]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl7/igt@[email protected]

* igt@kms_draw_crc@draw-method-xrgb2101010-render-untiled:
- shard-skl: [FAIL][37] ([fdo#103184] / [fdo#103232]) -> [PASS][38]
[37]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl4/igt@[email protected]
[38]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl1/igt@[email protected]

* igt@kms_draw_crc@draw-method-xrgb8888-mmap-gtt-untiled:
- shard-snb: [SKIP][39] ([fdo#109271]) -> [PASS][40]
[39]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-snb2/igt@[email protected]
[40]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-snb7/igt@[email protected]

* igt@kms_pipe_crc_basic@read-crc-pipe-c:
- shard-skl: [FAIL][41] ([fdo#103191]) -> [PASS][42]
[41]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl2/igt@[email protected]
[42]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl3/igt@[email protected]

* igt@kms_plane_alpha_blend@pipe-c-coverage-7efc:
- shard-skl: [FAIL][43] ([fdo#108145] / [fdo#110403]) -> [PASS][44]
[43]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl6/igt@[email protected]
[44]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl9/igt@[email protected]

* igt@kms_vblank@pipe-c-ts-continuation-dpms-suspend:
- shard-skl: [INCOMPLETE][45] ([fdo#104108]) -> [PASS][46] +1 similar issue
[45]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-skl2/igt@[email protected]
[46]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-skl6/igt@[email protected]

#### Warnings ####

* igt@gem_pwrite@huge-cpu-forwards:
- shard-kbl: [INCOMPLETE][47] ([fdo#103665]) -> [SKIP][48] ([fdo#109271]) +1 similar issue
[47]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-kbl3/igt@[email protected]
[48]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-kbl1/igt@[email protected]

* igt@gem_userptr_blits@coherency-sync:
- shard-kbl: [SKIP][49] ([fdo#109271]) -> [INCOMPLETE][50] ([fdo#103665])
[49]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-kbl1/igt@[email protected]
[50]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-kbl3/igt@[email protected]
- shard-glk: [SKIP][51] ([fdo#109271]) -> [INCOMPLETE][52] ([fdo#103359] / [k.org#198133])
[51]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6110/shard-glk3/igt@[email protected]
[52]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/shard-glk4/igt@[email protected]

{name}: This element is suppressed. This means it is ignored when computing
the status of the difference (SUCCESS, WARNING, or FAILURE).

[fdo#103184]: https://bugs.freedesktop.org/show_bug.cgi?id=103184
[fdo#103191]: https://bugs.freedesktop.org/show_bug.cgi?id=103191
[fdo#103232]: https://bugs.freedesktop.org/show_bug.cgi?id=103232
[fdo#103359]: https://bugs.freedesktop.org/show_bug.cgi?id=103359
[fdo#103540]: https://bugs.freedesktop.org/show_bug.cgi?id=103540
[fdo#103665]: https://bugs.freedesktop.org/show_bug.cgi?id=103665
[fdo#104108]: https://bugs.freedesktop.org/show_bug.cgi?id=104108
[fdo#105411]: https://bugs.freedesktop.org/show_bug.cgi?id=105411
[fdo#107807]: https://bugs.freedesktop.org/show_bug.cgi?id=107807
[fdo#108145]: https://bugs.freedesktop.org/show_bug.cgi?id=108145
[fdo#108566]: https://bugs.freedesktop.org/show_bug.cgi?id=108566
[fdo#108686]: https://bugs.freedesktop.org/show_bug.cgi?id=108686
[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[fdo#109278]: https://bugs.freedesktop.org/show_bug.cgi?id=109278
[fdo#110401]: https://bugs.freedesktop.org/show_bug.cgi?id=110401
[fdo#110403]: https://bugs.freedesktop.org/show_bug.cgi?id=110403
[fdo#110667]: https://bugs.freedesktop.org/show_bug.cgi?id=110667
[k.org#198133]: https://bugzilla.kernel.org/show_bug.cgi?id=198133

Participating hosts (10 -> 9)
------------------------------

Missing (1): shard-iclb

Build changes
-------------

* Linux: CI_DRM_6110 -> Patchwork_13054

CI_DRM_6110: ae35264eb9f700d5f9856552bf98e20ec2b74739 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4999: eee6e9ea7a7e22a64afc029a696b80da8053bc3c @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_13054: 949e61ed7490fb4da9dd29729084af60a1174d4d @ git://anongit.freedesktop.org/gfx-ci/linux
piglit_4509: fdc5a4ca11124ab8413c7988896eec4c97336694 @ git://anongit.freedesktop.org/piglit

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13054/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx