drm/i915: Avoid circular locking dependency when flush delayed work on gt reset
Subsystems (2)
PATCH v5 1/1 drm/i915: Avoid circular locking dependency when flush delayed work on gt reset
This attempts to avoid circular locking dependency between flush delayed work and intel_gt_reset.
When intel_gt_reset was called, task will hold a lock.
To cacel delayed work here, the _sync version will also acquire a lock, which might trigger the possible cirular locking dependency warning. When intel_gt_reset called, reset_in_progress flag will be set, add code to check the flag, call async verion if reset is in progress.
--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +10 -1
@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ +if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc)
Signature · 1 line
2.34.1
Patchwork
CI[Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
== Series Details ==
== Series Details ==
Series: drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
URL : https://patchwork.freedesktop.org/series/118898/
State : warning
== Summary ==
Error: dim checkpatch failed/home/kbuild/linux/maintainer-tools/dim: line 50: /home/kbuild/.dimrc: No such file or directory
Patchwork
CI[Intel-gfx] ✗ Fi.CI.SPARSE: warning for drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
In reply to Zhanjun Dong== Series Details ==
== Series Details ==
Series: drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
URL : https://patchwork.freedesktop.org/series/118898/
State : warning
== Summary ==
Error: dim sparse failed/home/kbuild/linux/maintainer-tools/dim: line 50: /home/kbuild/.dimrc: No such file or directory
Patchwork
CI[Intel-gfx] ✗ Fi.CI.BAT: failure for drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
In reply to Zhanjun Dong== Series Details ==
== Series Details ==
Series: drm/i915: Avoid circular locking dependency when flush delayed work on gt reset (rev5)
URL : https://patchwork.freedesktop.org/series/118898/
State : failure
== Summary ==
CI Bug Log - changes from CI_DRM_13510 -> Patchwork_118898v5
====================================================
Summary
-------
**FAILURE**
Serious unknown changes coming with Patchwork_118898v5 absolutely need to be verified manually.
If you think the reported changes have nothing to do with the changes introduced in Patchwork_118898v5, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/index.html
Participating hosts (43 -> 41)
------------------------------
Missing (2): bat-dg2-11 fi-snb-2520m
Possible new issues
-------------------
Here are the unknown changes that may have been introduced in Patchwork_118898v5:
### IGT changes ###
#### Possible regressions ####
* igt@gem_tiled_pread_basic:
- fi-kbl-soraka: [PASS][1] -> [INCOMPLETE][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/fi-kbl-soraka/igt@gem_tiled_pread_basic.html
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/fi-kbl-soraka/igt@gem_tiled_pread_basic.html
Known issues
------------
Here are the changes found in Patchwork_118898v5 that come from known issues:
### IGT changes ###
#### Issues hit ####
* igt@gem_lmem_swapping@parallel-random-engines:
- bat-adln-1: NOTRUN -> [SKIP][3] ([i915#4613]) +3 similar issues
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@[email protected]
- bat-rplp-1: NOTRUN -> [SKIP][4] ([i915#4613]) +3 similar issues
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-rplp-1/igt@[email protected]
* igt@gem_lmem_swapping@random-engines:
- bat-adlp-6: NOTRUN -> [SKIP][5] ([i915#4613]) +3 similar issues
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@[email protected]
* igt@i915_pm_rps@basic-api:
- bat-adlp-6: NOTRUN -> [SKIP][6] ([i915#6621])
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@[email protected]
- bat-rplp-1: NOTRUN -> [SKIP][7] ([i915#6621])
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-rplp-1/igt@[email protected]
- bat-adln-1: NOTRUN -> [SKIP][8] ([i915#6621])
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@[email protected]
* igt@i915_selftest@live@gt_lrc:
- bat-adlp-9: [PASS][9] -> [INCOMPLETE][10] ([i915#4983] / [i915#7913])
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-adlp-9/igt@i915_selftest@live@gt_lrc.html
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-9/igt@i915_selftest@live@gt_lrc.html
* igt@i915_selftest@live@gt_pm:
- bat-adln-1: NOTRUN -> [DMESG-FAIL][11] ([i915#4258])
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@i915_selftest@live@gt_pm.html
* igt@i915_selftest@live@slpc:
- bat-mtlp-6: [PASS][12] -> [DMESG-WARN][13] ([i915#6367])
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-mtlp-6/igt@i915_selftest@[email protected]
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-6/igt@i915_selftest@[email protected]
* igt@kms_chamelium_hpd@common-hpd-after-suspend:
- bat-rplp-1: NOTRUN -> [SKIP][14] ([i915#7828])
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-rplp-1/igt@[email protected]
- bat-adln-1: NOTRUN -> [SKIP][15] ([i915#7828])
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@[email protected]
- bat-adlp-6: NOTRUN -> [SKIP][16] ([i915#7828])
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@[email protected]
- fi-apl-guc: NOTRUN -> [SKIP][17] ([fdo#109271])
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/fi-apl-guc/igt@[email protected]
- bat-mtlp-8: NOTRUN -> [SKIP][18] ([i915#7828])
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-8/igt@[email protected]
* igt@kms_setmode@basic-clone-single-crtc:
- bat-adln-1: NOTRUN -> [SKIP][19] ([i915#3555])
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@[email protected]
- bat-adlp-6: NOTRUN -> [SKIP][20] ([i915#3555])
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@[email protected]
* igt@prime_vgem@basic-read:
- bat-rplp-1: NOTRUN -> [SKIP][21] ([i915#3708]) +2 similar issues
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-rplp-1/igt@[email protected]
* igt@prime_vgem@basic-write:
- bat-adln-1: NOTRUN -> [SKIP][22] ([fdo#109295] / [i915#3291]) +2 similar issues
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@[email protected]
- bat-adlp-6: NOTRUN -> [SKIP][23] ([fdo#109295] / [i915#3291] / [i915#3708]) +2 similar issues
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@[email protected]
#### Possible fixes ####
* igt@i915_pm_rpm@module-reload:
- fi-apl-guc: [DMESG-FAIL][24] ([i915#8585]) -> [PASS][25]
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/fi-apl-guc/igt@[email protected]
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/fi-apl-guc/igt@[email protected]
* igt@i915_selftest@live@gt_heartbeat:
- fi-kbl-soraka: [DMESG-FAIL][26] ([i915#5334] / [i915#7872]) -> [PASS][27]
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/fi-kbl-soraka/igt@i915_selftest@live@gt_heartbeat.html
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/fi-kbl-soraka/igt@i915_selftest@live@gt_heartbeat.html
- fi-glk-j4005: [DMESG-FAIL][28] ([i915#5334]) -> [PASS][29]
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/fi-glk-j4005/igt@i915_selftest@live@gt_heartbeat.html
[29]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/fi-glk-j4005/igt@i915_selftest@live@gt_heartbeat.html
* igt@i915_selftest@live@gt_mocs:
- bat-mtlp-8: [DMESG-FAIL][30] ([i915#7059]) -> [PASS][31]
[30]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-mtlp-8/igt@i915_selftest@live@gt_mocs.html
[31]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-8/igt@i915_selftest@live@gt_mocs.html
* igt@i915_selftest@live@requests:
- bat-mtlp-6: [DMESG-FAIL][32] ([i915#8497]) -> [PASS][33]
[32]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-mtlp-6/igt@i915_selftest@[email protected]
[33]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-6/igt@i915_selftest@[email protected]
* igt@i915_selftest@live@slpc:
- bat-mtlp-8: [DMESG-WARN][34] ([i915#6367]) -> [PASS][35]
[34]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-mtlp-8/igt@i915_selftest@[email protected]
[35]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-8/igt@i915_selftest@[email protected]
* igt@kms_pipe_crc_basic@read-crc-frame-sequence@pipe-a-edp-1:
- bat-adlp-6: [ABORT][36] ([i915#7977] / [i915#8469] / [i915#8668]) -> [PASS][37]
[36]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-adlp-6/igt@kms_pipe_crc_basic@[email protected]
[37]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adlp-6/igt@kms_pipe_crc_basic@[email protected]
* igt@kms_psr@sprite_plane_onoff:
- bat-adln-1: [ABORT][38] ([i915#8442] / [i915#8668]) -> [PASS][39]
[38]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-adln-1/igt@kms_psr@sprite_plane_onoff.html
[39]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-adln-1/igt@kms_psr@sprite_plane_onoff.html
#### Warnings ####
* igt@i915_suspend@basic-s3-without-i915:
- bat-mtlp-8: [ABORT][40] ([i915#8802]) -> [SKIP][41] ([i915#6645])
[40]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-mtlp-8/igt@[email protected]
[41]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-mtlp-8/igt@[email protected]
* igt@kms_setmode@basic-clone-single-crtc:
- bat-rplp-1: [ABORT][42] ([i915#8260] / [i915#8668]) -> [SKIP][43] ([i915#3555])
[42]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_13510/bat-rplp-1/igt@[email protected]
[43]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/bat-rplp-1/igt@[email protected]
[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[fdo#109295]: https://bugs.freedesktop.org/show_bug.cgi?id=109295
[i915#3291]: https://gitlab.freedesktop.org/drm/intel/issues/3291
[i915#3555]: https://gitlab.freedesktop.org/drm/intel/issues/3555
[i915#3708]: https://gitlab.freedesktop.org/drm/intel/issues/3708
[i915#4258]: https://gitlab.freedesktop.org/drm/intel/issues/4258
[i915#4613]: https://gitlab.freedesktop.org/drm/intel/issues/4613
[i915#4983]: https://gitlab.freedesktop.org/drm/intel/issues/4983
[i915#5334]: https://gitlab.freedesktop.org/drm/intel/issues/5334
[i915#6367]: https://gitlab.freedesktop.org/drm/intel/issues/6367
[i915#6621]: https://gitlab.freedesktop.org/drm/intel/issues/6621
[i915#6645]: https://gitlab.freedesktop.org/drm/intel/issues/6645
[i915#7059]: https://gitlab.freedesktop.org/drm/intel/issues/7059
[i915#7828]: https://gitlab.freedesktop.org/drm/intel/issues/7828
[i915#7872]: https://gitlab.freedesktop.org/drm/intel/issues/7872
[i915#7913]: https://gitlab.freedesktop.org/drm/intel/issues/7913
[i915#7977]: https://gitlab.freedesktop.org/drm/intel/issues/7977
[i915#8260]: https://gitlab.freedesktop.org/drm/intel/issues/8260
[i915#8442]: https://gitlab.freedesktop.org/drm/intel/issues/8442
[i915#8469]: https://gitlab.freedesktop.org/drm/intel/issues/8469
[i915#8497]: https://gitlab.freedesktop.org/drm/intel/issues/8497
[i915#8585]: https://gitlab.freedesktop.org/drm/intel/issues/8585
[i915#8668]: https://gitlab.freedesktop.org/drm/intel/issues/8668
[i915#8802]: https://gitlab.freedesktop.org/drm/intel/issues/8802
Build changes
-------------
* Linux: CI_DRM_13510 -> Patchwork_118898v5
CI-20190529: 20190529
CI_DRM_13510: 3e03d3e65440f4d60f19ff7b0313f3faf70de813 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_7432: 1b874bbe5e3576f89f5b6abf15a56917a96ec1ad @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
Patchwork_118898v5: 3e03d3e65440f4d60f19ff7b0313f3faf70de813 @ git://anongit.freedesktop.org/gfx-ci/linux
### Linux commits
6e6acbea108a drm/i915: Avoid circular locking dependency when flush delayed work on gt reset
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_118898v5/index.html
Hi Zhanjun,
Quoted 4 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock,
/cacel/cancel
Quoted 3 lines
which might trigger the possible cirular locking dependency warning. When
intel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.
/verion/version/
Quoted 5 lines
Signed-off-by Zhanjun Dong <[email protected]>Cc John Harrison <[email protected]>Cc Andi Shyti <[email protected]>Cc Daniel Vetter <[email protected]>---
There is no changelog here :/
Can you please add the changelog after the '---' section?
The commit log has changed and...
Quoted 12 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work);
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1371
Quoted 10 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +10
-0+/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ +if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work);
... now you are checking out of reset_in_progress.
Normally the convention here is to have the *_locked() version of the function. But I'm OK with this, as well... John, any opinion?
Anyway, comparing with your previous patch the decision is made out of different elements and only __reset_guc_busyness_stats() needed this change.
Andi
Quoted 5 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0}static void __reset_guc_busyness_stats (struct intel_guc *guc)Signature · 1 line
2.34.1
Quoted 21 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong <[email protected]>Cc John Harrison <[email protected]>Cc Andi Shyti <[email protected]>Cc Daniel Vetter <[email protected]>--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1365
Quoted 8 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +4
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning.
This is not even close to a locking bugfix. Consider this a formal nack, because the issue here is not even close to "needs more comments to explain what's going on".
-Daniel
Quoted 12 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-0+* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ +if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc)Signature · 1 line
2.34.1
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 44 lines
-----Original Message-----
From: Daniel Vetter <[email protected]>
Sent: August 22, 2023 9:51 AM
To: Dong, Zhanjun <[email protected]>
Cc: [email protected]; [email protected]; Harrison, John C <[email protected]>; Andi Shyti <[email protected]>; Daniel Vetter <[email protected]>
Subject: Re: [PATCH v5] drm/i915: Avoid circular locking dependency when flush delayed work on gt resetQuoted 12 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong <[email protected]>Cc John Harrison <[email protected]>Cc Andi Shyti <[email protected]>Cc Daniel Vetter <[email protected]>Review
Quoted 5 lines
--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cdrivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct
Quoted 4 lines
index
a0e3ef1c65d2..600388c849f7100644drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct
intel_guc*guc)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc)
Quoted 7 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +3
-1static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might
Quoted 6 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1365
Quoted 1 line
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +1
-0+* trigger the possible cirular locking dependency warning. This is not even close to a locking bugfix. Consider this a formal nack, because the issue here is not even close to "needs more comments to explain what's going on".
-Daniel
The purpose of the comment here it is to explain locking issue condition
Quoted 3 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct · line 1365
Quoted 2 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +1
-0+* Check the reset_in_progress flag, call async verion if reset is in progress.
The comment here explains check with the flag to avoid locking condition.
The reset process is not considered to be complete in short time, other than that, do we missed anything?
Quoted 16 lines
Quoted 11 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +5
-1+*/ +if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc) --2.34.1
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Quoted 51 lines · Zhanjun
Quoted 44 lines
-----Original Message-----
From: Daniel Vetter <[email protected]>
Sent: August 22, 2023 9:51 AM
To: Dong, Zhanjun <[email protected]>
Cc: [email protected]; [email protected]; Harrison, John C <[email protected]>; Andi Shyti <[email protected]>; Daniel Vetter <[email protected]>
Subject: Re: [PATCH v5] drm/i915: Avoid circular locking dependency when flush delayed work on gt resetQuoted 12 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong <[email protected]>Cc John Harrison <[email protected]>Cc Andi Shyti <[email protected]>Cc Daniel Vetter <[email protected]>Review
Quoted 5 lines
--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)(quoted diff) +0
-0
b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cdrivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct
Quoted 4 lines
index
a0e3ef1c65d2..600388c849f7100644drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct
intel_guc*guc)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc)
Quoted 7 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +3
-1static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might
Quoted 6 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1365
Quoted 1 line
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +1
-0+* trigger the possible cirular locking dependency warning. This is not even close to a locking bugfix. Consider this a formal nack, because the issue here is not even close to "needs more comments to explain what's going on".
-Daniel
Quoted 8 lines
The purpose of the comment here it is to explain locking issue condition
Quoted 3 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct · line 1365
Quoted 2 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +1
-0+* Check the reset_in_progress flag, call async verion if reset is in progress.
The comment here explains check with the flag to avoid locking condition.
The reset process is not considered to be complete in short time, other than that, do we missed anything?
Either the _sync is not needed at all, in case you need to explain why. Which this patch doesn't. And if the _sync isn't needed, then it's probably not needed in all/most cases?
Or the _sync is needed, and in that case you just replace a potential deadlock scenario with a potential race condition.
In neither case should this patch here be merged.
-Daniel
Quoted 17 lines
Quoted 17 lines
Quoted 10 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +5
-1+*/ +if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc) --Quoted 2 lines
2.34.1
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 21 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1367
Quoted 10 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/
This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the cancel_sync version is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
John.
Quoted 7 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +4
-0+if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc)
Quoted 51 lines · John Harrison
Quoted 21 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1366
Quoted 9 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
Just remove the cancel from the reset path as uneeded instead, and explain why that's ok? Because that's defacto what the cancel_work with a potential deadlock scenario for cancel_work_sync does, you either don't need it at all, or the replacement creates a bug.
-Daniel
Quoted 11 lines
John.
Quoted 6 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +4
-0+if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc)
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 52 lines · Daniel Vetter
Quoted 51 lines · John Harrison
Quoted 21 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1366
Quoted 9 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.
Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
John.
Quoted 17 lines
Just remove the cancel from the reset path as uneeded instead, and explain why that's ok? Because that's defacto what the
cancel_workwith a potential deadlock scenario forcancel_work_syncdoes, you either don't need it at all, or the replacement creates a bug.
-DanielQuoted 10 lines
John.
Quoted 6 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +4
-0+if (guc_to_gt (guc)->uc.reset_in_progress) +cancel_delayed_work (&guc->timestamp.work); +else +cancel_delayed_work_sync (&guc->timestamp.work); }static void __reset_guc_busyness_stats (struct intel_guc *guc)
Quoted 56 lines · John Harrison
Quoted 55 lines · Daniel Vetter
Quoted 54 lines · John Harrison
Quoted 28 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>---
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c| 11 ++++++++++-1 file changed, 10 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +3
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc) static void guc_cancel_busyness_worker(struct intel_guc *guc) { - cancel_delayed_work_sync(&guc->timestamp.work); + /* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also Quoted 5 lines
acquire a lock, which might
+ * trigger the possible cirular locking dependency warning. + * Check thereset_in_progressflag, call async verion if reset is in progress.
+ */This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
John.
@Daniel Vetter - ping? Is this explanation sufficient? Are you okay with this change now?
John.
Quoted 19 lines
Quoted 18 lines
Just remove the cancel from the reset path as uneeded instead, and explain why that's ok? Because that's defacto what the
cancel_workwith a potential deadlock scenario forcancel_work_syncdoes, you either don't need it at all, or the replacement creates a bug.
-DanielQuoted 10 lines
John.
Quoted 6 lines
+ if (guc_to_gt(guc)->uc.reset_in_progress)
+cancel_delayed_work(&guc->timestamp.work);
+ else
+cancel_delayed_work_sync(&guc->timestamp.work);
}
static void__reset_guc_busyness_stats(struct intel_guc *guc)
Quoted 77 lines · John Harrison
Quoted 76 lines · John Harrison
Quoted 62 lines · Daniel Vetter
Quoted 61 lines · John Harrison
Quoted 33 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed
work andintel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock,
which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code
to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>---
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c| 11 ++++++++++-1 file changed, 10 insertions(+), 1 deletion(-)diff --git
a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cb/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cindexa0e3ef1c65d2..600388c849f7100644drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +3
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc) static void guc_cancel_busyness_worker(struct intel_guc *guc) { - cancel_delayed_work_sync(&guc->timestamp.work); + /* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will Quoted 5 lines
also acquire a lock, which might
+ * trigger the possible cirular locking dependency warning. + * Check thereset_in_progressflag, call async verion if reset is in progress.
+ */This needs to explain in much more detail what is going on and why it is not
a problem. E.g.:The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the
reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if thecancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 5 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while
the reset is in progress?It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
Quoted 3 lines
Quoted 1 line
John.
@Daniel Vetter - ping? Is this explanation sufficient? Are you okay with this change now?
Sorry for the late reply, I'm constantly behind on mails :-/ Ping me on irc next time around if I don't reply, that's quicker.
"presumably" isn't good enough for locking design. Either you know, and can prove it all, or you shouldn't touch the code and its locking design before you've figured this out.
Again, either this is a deadlock, race condition, or the cancel isn't necessary. And this argument works in full generality. All this patch does it replace the dealock with one of the other two, and that's not good enough if you don't even know which one it is.
- if you need the cancel, you have a race condition
- if you don't have a race condition, you don't need the cancel
- currently you have the deadlock
"presumably" and "maybe" aint enoug for locking design.
Cheers, Daniel
Quoted 24 lines
John.
Quoted 20 lines
Quoted 19 lines
Just remove the cancel from the reset path as uneeded instead, and explain
why that's ok? Because that's defacto what thecancel_workwith a potential deadlock scenario forcancel_work_syncdoes, you either don't need it at all, or the replacement creates a bug.
-DanielQuoted 10 lines
John.
Quoted 6 lines
+ if (guc_to_gt(guc)->uc.reset_in_progress)
+cancel_delayed_work(&guc->timestamp.work);
+ else
+cancel_delayed_work_sync(&guc->timestamp.work);
}
static void__reset_guc_busyness_stats(struct intel_guc *guc)
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 88 lines · Daniel Vetter
Quoted 77 lines · John Harrison
Quoted 76 lines · John Harrison
Quoted 62 lines · Daniel Vetter
Quoted 61 lines · John Harrison
Quoted 33 lines · Zhanjun Dong
This attempts to avoid circular locking dependency between flush delayed
work andintel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock,
which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code
to check the flag, call async verion if reset is in progress.Signed-off-by Zhanjun Dong<[email protected]>Cc John Harrison<[email protected]>Cc Andi Shyti<[email protected]>Cc Daniel Vetter<[email protected]>--- drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)diff --git
a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cb/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cindexa0e3ef1c65d2..600388c849f7100644drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +3
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc) static void guc_cancel_busyness_worker(struct intel_guc *guc) { - cancel_delayed_work_sync(&guc->timestamp.work); + /* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will Quoted 5 lines
also acquire a lock, which might
+ * trigger the possible cirular locking dependency warning. + * Check thereset_in_progressflag, call async verion if reset is in progress.
+ */This needs to explain in much more detail what is going on and why it is not
a problem. E.g.:The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails
out if the
reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if thecancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 5 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while
the reset is in progress?It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
Quoted 3 lines
Quoted 1 line
John.
@Daniel Vetter - ping? Is this explanation sufficient? Are you okay with this change now?
Sorry for the late reply, I'm constantly behind on mails :-/ Ping me on irc next time around if I don't reply, that's quicker.
"presumably" isn't good enough for locking design. Either you know, and can prove it all, or you shouldn't touch the code and its locking design before you've figured this out.
Quoted 8 lines
Again, either this is a deadlock, race condition, or the cancel isn't necessary. And this argument works in full generality. All this patch does it replace the dealock with one of the other two, and that's not good enough if you don't even know which one it is.
- if you need the cancel, you have a race condition
- if you don't have a race condition, you don't need the cancel
In the case of a reset in progress, we do not strictly need the cancel. The worker thread will take care of avoiding a deadlock by itself. But it is more efficient to do the cancel and avoid unnecessary code execution if possible. It is also more logically correct - the worker is being stopped, therefore we should cancel any pending execution of the worker.
In the case of a reset not being in progress, we absolutely do need the cancel as there are multiple race conditions.
Quoted 2 lines
- currently you have the deadlock
No, we do not. There is no deadlock.
The worker thread explicitly does a trylock and reschedules itself for later if it could not get the lock. Lockdep does not understand the back off semantics of the trylock and reports a false failure.
As explained in the above code comment, if a reset is in progress then the synchronous cancel is not required because the trylock will take care of it. An asynchronous cancel is still better than no cancel because it at least tries to stop the worker from running. It's not a problem if the worker does run, but there is no advantage to running it so why not attempt to cancel it and prevent unnecessary code execution? If a reset is not in progress then we do want the synchronous cancel because the disable is potentially part of a driver unload or similar operation that requires the worker to be stopped first. Otherwise, the worker will attempt to reference pointers that might no longer exist at the time it runs. Clearly, that is not the situation if a reset is in progress. The driver cannot be unloading if it is in the middle of a reset operation.
Quoted 2 lines
"presumably" and "maybe" aint enoug for locking design.
Bad choice of words. I simply meant that the low level helper does not know the exact call stack it was called with. It can trivially determine if a reset is in progress or not, and that is all it really needs to know about. There are multiple other paths to this helper, none of which involve resets and some of which require a synchronous cancel.
A previous version of this patch attempted to have two separate helpers - one synchronous and one asynchronous. The intention being that any call stack involving reset would call the async version and any stack requiring the sync version would call that. However, the i915 reset design is hideously complex and those changes were much more intrusive and fragile. Simply testing the reset flag at the lowest level is significantly simpler and safer.
John.
Quoted 21 lines
Cheers, Daniel
Quoted 18 lines
John.
Quoted 16 lines
Quoted 16 lines
Just remove the cancel from the reset path as uneeded instead, and explain
why that's ok? Because that's defacto what thecancel_workwith a potential deadlock scenario forcancel_work_syncdoes, you either don't need it at all, or the replacement creates a bug.
-DanielQuoted 9 lines
John.
Quoted 6 lines
+ if (guc_to_gt(guc)->uc.reset_in_progress)
+cancel_delayed_work(&guc->timestamp.work);
+ else
+cancel_delayed_work_sync(&guc->timestamp.work);
}
static void__reset_guc_busyness_stats(struct intel_guc *guc)
Hi,
Quoted 38 lines
Quoted 34 lines
Quoted 34 lines
Quoted 5 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1366
Quoted 9 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
Quoted 7 lines
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
I was actually thinking why not leave things as they are and just disable lockdep from CI. This doesn't look like a relevant report to me.
Andi
Quoted 47 lines · Andi Shyti
Hi,
Quoted 38 lines
Quoted 34 lines
Quoted 34 lines
Quoted 5 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +0
-0drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1366
Quoted 9 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1@@ -1359,7 +1359,16 @@ static void guc_enable_busyness_worker(struct intel_guc *guc)static void guc_cancel_busyness_worker (struct intel_guc *guc) { -cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
Quoted 7 lines
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
Quoted 5 lines
I was actually thinking why not leave things as they are and just disable lockdep from CI. This doesn't look like a relevant report to me.
Andi
Disable lockdep? The whole of lockdep? We absolutely do not want to disable an extremely important deadlock testing infrastructure in our test framework. That would be defeating the whole point of CI.
Potentially we could annotate this one particular scenario to suppress this one particular error. But it seems simpler and safer to just update the code to not hit that scenario in the first place.
John.
Hi John,
Quoted 44 lines
Quoted 44 lines
Quoted 33 lines
Quoted 29 lines
Quoted 29 lines
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c static void guc_enable_busyness_worker(struct intel_guc *guc) · line 1367
Quoted 9 lines
static void
guc_cancel_busyness_worker(struct intel_guc *guc){drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +6
-1-cancel_delayed_work_sync (&guc->timestamp.work); +/* +* When intel_gt_reset was called, task will hold a lock. +* To cacel delayed work here, the _sync version will also acquire a lock, which might +* trigger the possible cirular locking dependency warning. +* Check the reset_in_progress flag, call async verion if reset is in progress. +*/ This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
Quoted 7 lines
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
I was actually thinking why not leave things as they are and just disable lockdep from CI. This doesn't look like a relevant report to me.
Quoted 8 lines
Quoted 1 line
Andi
Disable lockdep? The whole of lockdep? We absolutely do not want to disable an extremely important deadlock testing infrastructure in our test framework. That would be defeating the whole point of CI.
Potentially we could annotate this one particular scenario to suppress this one particular error. But it seems simpler and safer to just update the code to not hit that scenario in the first place.
yes... lockdep is a debug tool and might provide false reports... We need to have a great willingness to start fixing and hunting debug lockdep's false positives (like this one, for instance).
It's even more annoying to reduce our CI pass rates, especially when in BAT tests, with such false deadlocks.
It's the developer's responsibility to test its code with debug_lockdep and fix all the potential deadlocks and ignore the false ones.
I sent a patch for this[*] already.
Andi
[*] https://gitlab.freedesktop.org/gfx-ci/i915-infra/-/merge_requests/128
Jani Nikula
maintainerQuoted 13 lines · Andi Shyti
Quoted 4 lines
Quoted 4 lines
I was actually thinking why not leave things as they are and just disable lockdep from CI. This doesn't look like a relevant report to me.
Quoted 8 lines
Quoted 1 line
Andi
Disable lockdep? The whole of lockdep? We absolutely do not want to disable an extremely important deadlock testing infrastructure in our test framework. That would be defeating the whole point of CI.
Potentially we could annotate this one particular scenario to suppress this one particular error. But it seems simpler and safer to just update the code to not hit that scenario in the first place.
Quoted 6 lines
yes... lockdep is a debug tool and might provide false reports... We need to have a great willingness to start fixing and hunting debug lockdep's false positives (like this one, for instance).
It's even more annoying to reduce our CI pass rates, especially when in BAT tests, with such false deadlocks.
Make lockdep understand what you're doing, and there are no false positives. That's all there is to it.
Quoted 3 lines
It's the developer's responsibility to test its code with
debug_lockdepand fix all the potential deadlocks and ignore the false ones.
No. Manual validation of lockdep reports is not feasible. Lockdep is the tool to validate locking. It's the developer's responsibility to make lockdep understand the design.
Besides, locking is often subtle. Stuff can change as a side effect even when you're not intentionally changing locking, e.g. during refactoring. What you're suggesting effectively means all developers should run all of igt on a bunch of different generations of machines with lockdep enabled. Realistically, not going to happen, and we have CI because of this.
Quoted 5 lines
I sent a patch for this[*] already.
Andi
[*] https://gitlab.freedesktop.org/gfx-ci/i915-infra/-/merge_requests/128
Yeah, no.
BR,
Jani.
Signature · 1 line
Jani Nikula, Intel Open Source Graphics Center
Quoted 8 lines · Jani Nikula
Quoted 3 lines · Andi Shyti
It's the developer's responsibility to test its code with
debug_lockdepand fix all the potential deadlocks and ignore the false ones.No. Manual validation of lockdep reports is not feasible. Lockdep is the tool to validate locking. It's the developer's responsibility to make lockdep understand the design.
Yeah I guess I need to drop my locking design principle once more:
If lockdep doesn't understand your locking design, your design is shit.
You need to fix the design, not play whack-a-mole with lockdep. Or worse, pretend there's no problem and just disable lockdep outright.
If you don't understand your design, and can't succinctly explain it (or demonstrate the full hierarchy with lockdep priming, testing in CI isn't good enough for anything remotely complex), then you have a _really_ big problem. Yes CI is good at catching accidental changes in locking design, but if you use it for anything more than that you're in deep trouble.
Cheers, Sima
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Quoted 51 lines · Andi Shyti
Hi John,
Quoted 49 lines
Quoted 40 lines
Quoted 33 lines
Quoted 29 lines
Quoted 29 lines
Review
Quoted 9 lines
static void
guc_cancel_busyness_worker(struct intel_guc *guc){(quoted diff) +6
-1- cancel_delayed_work_sync(&guc->timestamp.work); + /* + * When intel_gt_reset was called, task will hold a lock. + * To cacel delayed work here, the _sync version will also acquire a lock, which might + * trigger the possible cirular locking dependency warning. + * Check the reset_in_progress flag, call async verion if reset is in progress. + */This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.Quoted 4 lines
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
So why do we even need to cancel anything if it doesn't do anything while the reset is in progress?
Quoted 7 lines
It still needs to be cancelled. The worker only aborts if it is actively executing concurrently with the reset. It might not start to execute until after the reset has completed. And there is presumably a reason why the cancel is being called, a reason not necessarily related to resets at all. Leaving the worker to run arbitrarily after the driver is expecting it to be stopped will lead to much worse things than a fake lockdep splat, e.g. a use after free pointer deref.
Quoted 5 lines
I was actually thinking why not leave things as they are and just disable lockdep from CI. This doesn't look like a relevant report to me.
Andi
Disable lockdep? The whole of lockdep? We absolutely do not want to disable an extremely important deadlock testing infrastructure in our test framework. That would be defeating the whole point of CI.
Quoted 6 lines
Quoted 3 lines
Potentially we could annotate this one particular scenario to suppress this one particular error. But it seems simpler and safer to just update the code to not hit that scenario in the first place.
yes... lockdep is a debug tool and might provide false reports... We need to have a great willingness to start fixing and hunting debug lockdep's false positives (like this one, for instance).
That is how lockdep works. It's like a compiler warning. You have to fix them even if you think they don't matter. Because otherwise, when someone tries to turn warnings on, they drown in a sea of other people's unrelated garbage that they did not bother to fix. If lockdep is to be of any use at all then it must be run regularly as part of a CI type system and any issues it finds must be fixed up by the developer's that own the relevant code. Where fixing means either fixing genuine bugs, re-working the code to not hit a false positive or annotating the code to explain to lockdep why it is a safe operation.
Quoted 3 lines
It's even more annoying to reduce our CI pass rates, especially when in BAT tests, with such false deadlocks.
Maybe. But it is even more annoying when you have a genuine locking issue that you don't notice because you have disabled lockdep and just have some random hang issue that is impossible to reproduce or debug.
Quoted 4 lines
It's the developer's responsibility to test its code with
debug_lockdepand fix all the potential deadlocks and ignore the false ones.
You seem to have this backwards. Developers are not expected to run every possible test on every possible platform in every possible configuration. That is the job of CI.
John.
Quoted 5 lines
I sent a patch for this[*] already.
Andi
[*] https://gitlab.freedesktop.org/gfx-ci/i915-infra/-/merge_requests/128
Hi Zhanjun,
Quoted 9 lines · John Harrison
On 8/11/2023 11:20, Zhanjun Dong wrote:
This attempts to avoid circular locking dependency between flush delayed work and
intel_gt_reset.
Whenintel_gt_resetwas called, task will hold a lock.
To cacel delayed work here, the_syncversion will also acquire a lock, which might trigger the possible cirular locking dependency warning. Whenintel_gt_resetcalled,reset_in_progressflag will be set, add code to check the flag, call async verion if reset is in progress.
I liked the previous commit, it just needed to be wrapped (not in the dmesg copy-paste part).
Quoted 14 lines
Signed-off-by Zhanjun Dong <[email protected]>Cc John Harrison <[email protected]>Cc Andi Shyti <[email protected]>Cc Daniel Vetter <[email protected]>---
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c| 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)diff --git
a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cb/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.cindexa0e3ef1c65d2..600388c849f7100644
---a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1359,7 +1359,16 @@ static voidguc_enable_busyness_worker(struct intel_guc *guc)
Quoted 9 lines
static void
guc_cancel_busyness_worker(struct intel_guc *guc){
-cancel_delayed_work_sync(&guc->timestamp.work);
+ /*
+ * Whenintel_gt_resetwas called, task will hold a lock.
+ * To cacel delayed work here, the_syncversion will also acquire a lock, which might
+ * trigger the possible cirular locking dependency warning.
+ * Check thereset_in_progressflag, call async verion if reset is in progress.
+ */
Indeed the commit message is a bit misleading and it raises some alarms if explained it this way.
Quoted 19 lines
This needs to explain in much more detail what is going on and why it is not a problem. E.g.:
The busyness worker needs to be cancelled. In general that means using the synchronous cancel version to ensure that an in-progress worker will not keep executing beyond whatever is happening that needs the cancel. E.g. suspend, driver unload, etc. However, in the case of a reset, the synchronous version is not required and can trigger a false deadlock detection warning.
The business worker takes the reset mutex to protect against resets interfering with it. However, it does a trylock and bails out if the reset lock is already acquired. Thus there is no actual deadlock or other concern with the worker running concurrently with a reset. So an asynchronous cancel is safe in the case of a reset rather than a driver unload or suspend type operation. On the other hand, if the
cancel_syncversion is used when a reset is in progress then the mutex deadlock detection sees the mutex being acquired through multiple paths and complains.
Quoted 2 lines
So just don't bother. That keeps the detection code happy and is safe because of the trylock code described above.
Can you please update the commit message with John's suggestion?
Is there any further question on this?
Andi