drm/atomic-helpers: remove legacy_cursor_update hacks
Subsystems (5)
PATCH v4 1/1 drm/atomic-helpers: remove legacy_cursor_update hacks
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit 674e78acae0dfb4beb56132e41cbae5b60f7d662
Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500
drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The complete_all is the only thing missing afaict.
v4: Rebased on recent kernel, added extra link for vc4 bug.
--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +0 -13
@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;
@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );
drivers/gpu/drm/i915/display/intel_display.c +13 -0
@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ +if (state->base.legacy_cursor_update) { +struct intel_crtc_state *new_crtc_state; +struct intel_crtc *crtc; +int i; + +for_each_new_intel_crtc_in_state (state, crtc, new_crtc_state, i) +complete_all (&new_crtc_state->uapi.commit->flip_done); + } + ret =intel_atomic_prepare_commit (state);if (ret) { drm_dbg_atomic(&dev_priv->drm,
drivers/gpu/drm/msm/msm_atomic.c +2 -0
@@ -237,6 +237,8 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)/* async updates are limited to single-crtc updates: */ WARN_ON (crtc_mask !=drm_crtc_mask (async_crtc)); +complete_all (&async_crtc->state->commit->flip_done); +/* * Start timer if we don't already have an update pending * on this crtc:
Signature · 1 line
2.35.1
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0
drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/
Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8869
Quoted 8 lines
drivers/gpu/drm/i915/display/intel_display.c +8
-0+if (state->base.legacy_cursor_update) { +struct intel_crtc_state *new_crtc_state; +struct intel_crtc *crtc; +int i; + +for_each_new_intel_crtc_in_state (state, crtc, new_crtc_state, i) +complete_all (&new_crtc_state->uapi.commit->flip_done); + }
You can complete what doesn't yet exist. Missing cc: intel-gfx for fireworks.
Quoted 19 lines
drivers/gpu/drm/i915/display/intel_display.c +1
-0+ ret =intel_atomic_prepare_commit (state);if (ret) { drm_dbg_atomic(&dev_priv->drm,drivers/gpu/drm/msm/msm_atomic.c +2
-0@@ -237,6 +237,8 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)/* async updates are limited to single-crtc updates: */ WARN_ON (crtc_mask !=drm_crtc_mask (async_crtc)); +complete_all (&async_crtc->state->commit->flip_done); +/* * Start timer if we don't already have an update pending * on this crtc: Signature · 1 line
2.35.1
Signature · 2 lines
Ville Syrjälä Intel
Quoted 93 lines · Ville Syrjälä
Quoted 92 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; }
Quoted 8 lines
drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 4 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0+/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
Quoted 10 lines
drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8869
Quoted 8 lines
drivers/gpu/drm/i915/display/intel_display.c +8
-0+if (state->base.legacy_cursor_update) { +struct intel_crtc_state *new_crtc_state; +struct intel_crtc *crtc; +int i; + +for_each_new_intel_crtc_in_state (state, crtc, new_crtc_state, i) +complete_all (&new_crtc_state->uapi.commit->flip_done); + }You can complete what doesn't yet exist. Missing cc: intel-gfx for fireworks.
Yeah that's a rebase error, my patch has it at the right place further down.
-Daniel
Quoted 24 lines
Quoted 20 lines
drivers/gpu/drm/i915/display/intel_display.c +1
-0+ ret =intel_atomic_prepare_commit (state);if (ret) { drm_dbg_atomic(&dev_priv->drm,drivers/gpu/drm/msm/msm_atomic.c +2
-0@@ -237,6 +237,8 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)/* async updates are limited to single-crtc updates: */ WARN_ON (crtc_mask !=drm_crtc_mask (async_crtc)); +complete_all (&async_crtc->state->commit->flip_done); +/* * Start timer if we don't already have an update pending * on this crtc: Signature · 1 line
2.35.1Signature · 2 lines
Ville Syrjälä Intel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 103 lines · Daniel Vetter
Quoted 93 lines · Ville Syrjälä
Quoted 92 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; }Quoted 8 lines
drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 4 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0+/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Signature · 2 lines
Ville Syrjälä Intel
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use drm_atomic_commit and only overwrite atomic_commit_tail. But I'm not really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-P
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.
You mean drm_atomic_helper_commit() I suppose?
Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-P
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueues
Those are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Signature · 2 lines
Ville Syrjälä Intel
Quoted 121 lines · Ville Syrjälä
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.You mean
drm_atomic_helper_commit()I suppose?Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-P
Quoted 8 lines
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueuesThose are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Yeah and we could have fixed them in the shared code, and 5+ years ago I even had patches, but i915 gem team had the idea they know better. That part really needs to be fixed, and if that means redoing a bunch of display work because display team didn't push back on gem team reinventing all the wheels ... tough luck.
I suggest you get started.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 131 lines · Daniel Vetter
Quoted 121 lines · Ville Syrjälä
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.You mean
drm_atomic_helper_commit()I suppose?Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-PQuoted 8 lines
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueuesThose are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Quoted 7 lines
Yeah and we could have fixed them in the shared code, and 5+ years ago I even had patches, but i915 gem team had the idea they know better. That part really needs to be fixed, and if that means redoing a bunch of display work because display team didn't push back on gem team reinventing all the wheels ... tough luck.
I suggest you get started.
I offered to do the hw/uapi split in the core. You refused it.
I raised my objection to the introduction of the single lock scheme for private state objects. No one was interested.
I don't think this situation is on me.
Signature · 2 lines
Ville Syrjälä Intel
Quoted 138 lines · Ville Syrjälä
Quoted 137 lines · Daniel Vetter
Quoted 121 lines · Ville Syrjälä
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.You mean
drm_atomic_helper_commit()I suppose?Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-PQuoted 8 lines
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueuesThose are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Yeah and we could have fixed them in the shared code, and 5+ years ago I even had patches, but i915 gem team had the idea they know better. That part really needs to be fixed, and if that means redoing a bunch of display work because display team didn't push back on gem team reinventing all the wheels ... tough luck.
Quoted 3 lines
Quoted 1 line
I suggest you get started.
I offered to do the hw/uapi split in the core. You refused it.
That should work with with atomic_commit_tail too, or at least I'm not seeing why not.
Quoted 2 lines
I raised my objection to the introduction of the single lock scheme for private state objects. No one was interested.
Yeah add it to the list of things i915 reinvents I guess. And as long as you're out, you kinda don't get much of a vote. Which is why being out of this isn't a bright idea.
Quoted 1 line
I don't think this situation is on me.
These were your examples why you can't adopt the atomic commit helpers, and aside from the workqueue one (where i915 really should not be the only driver doing it differently, that makes no sense at all) they're not real reasons to not do this.
The real reasons are
- cursor code not yet adopted async plane commit
- i915_sw_fence
- ... including the boosting and everything else that i915 gem team hand rolled in there
And those need to be fixed eventually. And this time around not by doing more i915 hand rolling of stuff.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 139 lines · Daniel Vetter
Quoted 138 lines · Ville Syrjälä
Quoted 137 lines · Daniel Vetter
Quoted 121 lines · Ville Syrjälä
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.You mean
drm_atomic_helper_commit()I suppose?Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-PQuoted 8 lines
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueuesThose are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Yeah and we could have fixed them in the shared code, and 5+ years ago I even had patches, but i915 gem team had the idea they know better. That part really needs to be fixed, and if that means redoing a bunch of display work because display team didn't push back on gem team reinventing all the wheels ... tough luck.
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
I suggest you get started.
I offered to do the hw/uapi split in the core. You refused it.
That should work with with
atomic_commit_tailtoo, or at least I'm not seeing why not.
I haven't gone through all of it to see how much it's poking inside the plane/crtc states and assuming the uapi state matches the hardware state. Maybe there is not much.
Quoted 7 lines
Quoted 3 lines
I raised my objection to the introduction of the single lock scheme for private state objects. No one was interested.
Yeah add it to the list of things i915 reinvents I guess. And as long as you're out, you kinda don't get much of a vote. Which is why being out of this isn't a bright idea.
It was reinvented _after_ the locking change. We were using private objects before.
Quoted 9 lines
Quoted 1 line
I don't think this situation is on me.
These were your examples why you can't adopt the atomic commit helpers, and aside from the workqueue one (where i915 really should not be the only driver doing it differently, that makes no sense at all) they're not real reasons to not do this.
The real reasons are
- cursor code not yet adopted async plane commit
As long as people talk about async commits while apparently meaning something totally different it's hard to even be on the same page.
Quoted 11 lines
-
i915_sw_fence
- ... including the boosting and everything else that i915 gem team hand rolled in thereAnd those need to be fixed eventually. And this time around not by doing more i915 hand rolling of stuff.
-DanielSignature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Signature · 2 lines
Ville Syrjälä Intel
Quoted 156 lines · Ville Syrjälä
Quoted 139 lines · Daniel Vetter
Quoted 138 lines · Ville Syrjälä
Quoted 137 lines · Daniel Vetter
Quoted 121 lines · Ville Syrjälä
Quoted 105 lines · Daniel Vetter
Quoted 104 lines · Ville Syrjälä
Quoted 103 lines · Daniel Vetter
Quoted 101 lines · Ville Syrjälä
Quoted 89 lines · Maxime Ripard
From: Daniel Vetter <[email protected]>
The stuff never really worked, and leads to lots of fun because it out-of-order frees atomic states. Which upsets KASAN, among other things.
For async updates we now have a more solid solution with the ->atomic_async_check and ->atomic_async_commit hooks. Support for that for msm and vc4 landed. nouveau and i915 have their own commit routines, doing something similar.
For everyone else it's probably better to remove the use-after-free bug, and encourage folks to use the async support instead. The affected drivers which register a legacy cursor plane and don't either use the new async stuff or their own commit routine are: amdgpu, atmel, mediatek, qxl, rockchip, sti, sun4i, tegra, virtio, and vmwgfx.
Inspired by an amdgpu bug report.
v2: Drop RFC, I think with amdgpu converted over to use atomic_async_check/commit done in
commit
674e78acae0dfb4beb56132e41cbae5b60f7d662Author: Nicholas Kazlauskas <[email protected]> Date: Wed Dec 5 14:59:07 2018 -0500drm/amd/display: Add fast path for cursor plane updates
we don't have any driver anymore where we have userspace expecting solid legacy cursor support _and_ they are using the atomic helpers in their fully glory. So we can retire this.
v3: Paper over msm and i915 regression. The
complete_allis the only thing missing afaict.v4: Rebased on recent kernel, added extra link for vc4 bug.
Cc Michel Dänzer <[email protected]>Cc Rob Clark <[email protected]>Cc "Kazlauskas, Nicholas" <[email protected]>Tested-by Maxime Ripard <[email protected]>Signed-off-by Daniel Vetter <[email protected]>Signed-off-by Maxime Ripard <[email protected]>--- drivers/gpu/drm/drm_atomic_helper.c | 13 ------------- drivers/gpu/drm/i915/display/intel_display.c | 13 +++++++++++++ drivers/gpu/drm/msm/msm_atomic.c | 2 ++ 3 files changed, 15 insertions(+), 13 deletions(-)drivers/gpu/drm/drm_atomic_helper.c +0
-13@@ -1498,13 +1498,6 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,int i, ret;unsigned int crtc_mask =0 ; -/* -* Legacy cursor ioctls are completely unsynced, and userspace -* relies on that (by doing tons of cursor updates). -*/ -if (old_state->legacy_cursor_update) -return ; -for_each_oldnew_crtc_in_state (old_state, crtc, old_crtc_state, new_crtc_state, i) {if (!new_crtc_state->active)continue ;@@ -2135,12 +2128,6 @@ int drm_atomic_helper_setup_commit(struct drm_atomic_state *state,continue ; } -/* Legacy cursor updates are fully unsynced. */ -if (state->legacy_cursor_update) { -complete_all (&commit->flip_done); -continue ; - } -if (!new_crtc_state->event) { commit->event =kzalloc (sizeof (*commit->event),GFP_KERNEL );drivers/gpu/drm/i915/display/intel_display.c +0
-0drivers/gpu/drm/i915/display/intel_display.c static int intel_atomic_commit(struct drm_device *dev, · line 8861
Quoted 7 lines
drivers/gpu/drm/i915/display/intel_display.c +4
-0@@ -8855,6 +8855,19 @@ static int intel_atomic_commit(struct drm_device *dev, state->base.legacy_cursor_update = false; } +/* +* FIXME: Cut over to (async) commit helpers instead of hand-rolling +* everything. +*/ Intel cursors can't even do async updates so this is rather nonsensical. What we need is some kind of reasonable mailbox support.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is not the async plane update you're thinking of. i915 really should switch over more to atomic helpers.
The comment should be clarified then. As is I have no real idea what it's trying to say.
Use
drm_atomic_commitand only overwriteatomic_commit_tail.You mean
drm_atomic_helper_commit()I suppose?Quoted 4 lines
But I'm not
really clear why the comment isn't clear - i915 is the only driver not using them, maybe should start to take a look when they're so unfamiliar :-PQuoted 8 lines
There are a lot of problems with that:
- no uapi/hw state split so if there's anything that looks at the state it's very likely going to get things wrong
- it doesn't know anything about i915's own state objects
- uses wrong workqueuesThose are the ones that immediately came to mind when I looked at it. Probably I missed some others as well.
Yeah and we could have fixed them in the shared code, and 5+ years ago I even had patches, but i915 gem team had the idea they know better. That part really needs to be fixed, and if that means redoing a bunch of display work because display team didn't push back on gem team reinventing all the wheels ... tough luck.
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
I suggest you get started.
I offered to do the hw/uapi split in the core. You refused it.
That should work with with
atomic_commit_tailtoo, or at least I'm not seeing why not.I haven't gone through all of it to see how much it's poking inside the plane/crtc states and assuming the uapi state matches the hardware state. Maybe there is not much.
Quoted 4 lines
Quoted 3 lines
I raised my objection to the introduction of the single lock scheme for private state objects. No one was interested.
Quoted 6 lines
Quoted 3 lines
Yeah add it to the list of things i915 reinvents I guess. And as long as you're out, you kinda don't get much of a vote. Which is why being out of this isn't a bright idea.
It was reinvented _after_ the locking change. We were using private objects before.
So the atomic state subclassing was deprecated in 2017 in the kerneldoc and drivers started moving over to drm_private_state (including i915):
commit da6c05969785a0f4108a089ef33c55f46ae21775
Author: Daniel Vetter <[email protected]> Date: Thu Dec 14 21:30:53 2017 +0100
drm/atomic: document how to handle driver private objects
And I actually replied in that old thread from 2018 where the drm_private_state locking was added and explained how to do read-only state in drm_private_state, but you didn't follow up on that:
https://lore.kernel.org/dri-devel/[email protected]/
Other drivers have been using that approach without issue meanwhile.
There's also been a pile of discussions on #dri-devel irc since then with more involved ideas, but for that entire time I have not heard a single time again that this is causing issues for intel. Nor any kinds of proposals how to fix it.
Instead you figured that's not for you and started moving i915 away from drm_private_state. Without cc'ing dri-devel, without kicking off the discussion again and maybe clarifying what the problem is exactly, and without explaining in the commit message that you're moving away from drm_private_state because of some locking issue in that thing that you don't want to address on dri-devel.
At least if I got this all right and found the right commit:
commit fd1a9bba73fa10e1601a43264283fe4696d6f82c
Author: Ville Syrjälä <[email protected]> Date: Mon Jan 20 19:47:25 2020 +0200
drm/i915: Convert bandwidth state to global state
I'm probably way more pissed on this than what's appropriate, but I just spend the last 3 years fixing up a giantic dumpster fire in the i915-gem side, because a few people were way too keen on reinveinting wheels in their little driver corner instead of working with dri-devel to get it all sorted out.
Cheers, Daniel
Quoted 26 lines
Quoted 9 lines
Quoted 1 line
I don't think this situation is on me.
These were your examples why you can't adopt the atomic commit helpers, and aside from the workqueue one (where i915 really should not be the only driver doing it differently, that makes no sense at all) they're not real reasons to not do this.
The real reasons are
- cursor code not yet adopted async plane commitAs long as people talk about async commits while apparently meaning something totally different it's hard to even be on the same page.
Quoted 11 lines
-
i915_sw_fence
- ... including the boosting and everything else that i915 gem team hand rolled in thereAnd those need to be fixed eventually. And this time around not by doing more i915 hand rolling of stuff.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Quoted 3 lines
--
Ville Syrjälä
Intel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch