drm/msm/mdp5: Fix global state lock backoff
PATCH 1/1 drm/msm/mdp5: Fix global state lock backoff
From: Rob Clark <[email protected]>
We need to grab the lock after the early return for !hwpipe case. Otherwise, we could have hit contention yet still returned 0.
Fixes an issue that the new CONFIG_DRM_DEBUG_MODESET_LOCK stuff flagged in CI:
WARNING: CPU: 0 PID: 282 at drivers/gpu/drm/drm_modeset_lock.c:296 drm_modeset_lock+0xf8/0x154
Modules linked in:
CPU: 0 PID: 282 Comm: kms_cursor_lega Tainted: G W 5.19.0-rc2-15930-g875cc8bc536a #1
Hardware name: Qualcomm Technologies, Inc. DB820c (DT)
pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : drm_modeset_lock+0xf8/0x154
lr : drm_atomic_get_private_obj_state+0x84/0x170
sp : ffff80000cfab6a0
x29: ffff80000cfab6a0 x28: 0000000000000000 x27: ffff000083bc4d00
x26: 0000000000000038 x25: 0000000000000000 x24: ffff80000957ca58
x23: 0000000000000000 x22: ffff000081ace080 x21: 0000000000000001
x20: ffff000081acec18 x19: ffff80000cfabb80 x18: 0000000000000038
x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffea0d0
x14: 0000000000000000 x13: 284e4f5f4e524157 x12: 5f534b434f4c5f47
x11: ffff80000a386aa8 x10: 0000000000000029 x9 : ffff80000cfab610
x8 : 0000000000000029 x7 : 0000000000000014 x6 : 0000000000000000
x5 : 0000000000000001 x4 : ffff8000081ad904 x3 : 0000000000000029
x2 : ffff0000801db4c0 x1 : ffff80000cfabb80 x0 : ffff000081aceb58
Call trace:
drm_modeset_lock+0xf8/0x154
drm_atomic_get_private_obj_state+0x84/0x170
mdp5_get_global_state+0x54/0x6c
mdp5_pipe_release+0x2c/0xd4
mdp5_plane_atomic_check+0x2ec/0x414
drm_atomic_helper_check_planes+0xd8/0x210
drm_atomic_helper_check+0x54/0xb0
...
---[ end trace 0000000000000000 ]---
drm_modeset_lock attempting to lock a contended lock without backoff:
drm_modeset_lock+0x148/0x154
mdp5_get_global_state+0x30/0x6c
mdp5_pipe_release+0x2c/0xd4
mdp5_plane_atomic_check+0x290/0x414
drm_atomic_helper_check_planes+0xd8/0x210
drm_atomic_helper_check+0x54/0xb0
drm_atomic_check_only+0x4b0/0x8f4
drm_atomic_commit+0x68/0xe0Fixes d59be579fa93 ("drm/msm/mdp5: Return error code in mdp5_pipe_release when deadlock is detected")
Signed-off-by Rob Clark <[email protected]>
--- drivers/gpu/drm/msm/disp/mdp5/mdp5_pipe.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/msm/disp/mdp5/mdp5_pipe.c +2 -1
@@ -123,12 +123,13 @@ int mdp5_pipe_release(struct drm_atomic_state *s, struct mdp5_hw_pipe *hwpipe) {struct msm_drm_private *priv = s->dev->dev_private;struct mdp5_kms *mdp5_kms =to_mdp5_kms (to_mdp_kms (priv->kms)); -struct mdp5_global_state *state =mdp5_get_global_state (s); +struct mdp5_global_state *state;struct mdp5_hw_pipe_state *new_state;if (!hwpipe)return 0 ; + state =mdp5_get_global_state (s);if (IS_ERR (state))return PTR_ERR (state);
Signature · 1 line
2.36.1
Quoted 46 lines · Rob Clark
From: Rob Clark <[email protected]>
We need to grab the lock after the early return for !hwpipe case. Otherwise, we could have hit contention yet still returned 0.
Fixes an issue that the new
CONFIG_DRM_DEBUG_MODESET_LOCKstuff flagged in CI:WARNING: CPU: 0 PID: 282 at drivers/gpu/drm/drm_modeset_lock.c:296 drm_modeset_lock+0xf8/0x154 Modules linked in: CPU: 0 PID: 282 Comm: kms_cursor_lega Tainted: G W 5.19.0-rc2-15930-g875cc8bc536a #1 Hardware name: Qualcomm Technologies, Inc. DB820c (DT) pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : drm_modeset_lock+0xf8/0x154 lr : drm_atomic_get_private_obj_state+0x84/0x170 sp : ffff80000cfab6a0 x29: ffff80000cfab6a0 x28: 0000000000000000 x27: ffff000083bc4d00 x26: 0000000000000038 x25: 0000000000000000 x24: ffff80000957ca58 x23: 0000000000000000 x22: ffff000081ace080 x21: 0000000000000001 x20: ffff000081acec18 x19: ffff80000cfabb80 x18: 0000000000000038 x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffea0d0 x14: 0000000000000000 x13: 284e4f5f4e524157 x12: 5f534b434f4c5f47 x11: ffff80000a386aa8 x10: 0000000000000029 x9 : ffff80000cfab610 x8 : 0000000000000029 x7 : 0000000000000014 x6 : 0000000000000000 x5 : 0000000000000001 x4 : ffff8000081ad904 x3 : 0000000000000029 x2 : ffff0000801db4c0 x1 : ffff80000cfabb80 x0 : ffff000081aceb58 Call trace: drm_modeset_lock+0xf8/0x154 drm_atomic_get_private_obj_state+0x84/0x170 mdp5_get_global_state+0x54/0x6c mdp5_pipe_release+0x2c/0xd4 mdp5_plane_atomic_check+0x2ec/0x414 drm_atomic_helper_check_planes+0xd8/0x210 drm_atomic_helper_check+0x54/0xb0 ... ---[ end trace 0000000000000000 ]--- drm_modeset_lock attempting to lock a contended lock without backoff: drm_modeset_lock+0x148/0x154 mdp5_get_global_state+0x30/0x6c mdp5_pipe_release+0x2c/0xd4 mdp5_plane_atomic_check+0x290/0x414 drm_atomic_helper_check_planes+0xd8/0x210 drm_atomic_helper_check+0x54/0xb0 drm_atomic_check_only+0x4b0/0x8f4 drm_atomic_commit+0x68/0xe0
Quoted 2 lines
Fixes d59be579fa93 ("drm/msm/mdp5: Return error code in mdp5_pipe_release when deadlock is detected")Signed-off-by Rob Clark <[email protected]>
Reviewed-by Abhinav Kumar <[email protected]>
Quoted 23 lines
--- drivers/gpu/drm/msm/disp/mdp5/mdp5_pipe.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)drivers/gpu/drm/msm/disp/mdp5/mdp5_pipe.c +2
-1@@ -123,12 +123,13 @@ int mdp5_pipe_release(struct drm_atomic_state *s, struct mdp5_hw_pipe *hwpipe) {struct msm_drm_private *priv = s->dev->dev_private;struct mdp5_kms *mdp5_kms =to_mdp5_kms (to_mdp_kms (priv->kms)); -struct mdp5_global_state *state =mdp5_get_global_state (s); +struct mdp5_global_state *state;struct mdp5_hw_pipe_state *new_state;if (!hwpipe)return 0 ; + state =mdp5_get_global_state (s);if (IS_ERR (state))return PTR_ERR (state);