drm/i915: Fix an incorrect free rather than derefence issue.
Subsystems (2)
PATCH 1/1 drm/i915: Fix an incorrect free rather than derefence issue.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88652
--- drivers/gpu/drm/i915/i915_drv.c | 2 +- drivers/gpu/drm/i915/i915_gem.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +1 -1
@@ -2661,7 +2661,7 @@ static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,intel_lr_context_unpin (ring, submit_req->ctx);i915_gem_context_unreference (submit_req->ctx); -kfree (submit_req); +i915_gem_request_unreference (submit_req); }/*
Signature · 6 lines
2.1.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 3 lines · Nick Hoath
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88652
Signed-off-by Nick Hoath <[email protected]>
Commit message is missing the absolutely crucial detail about which patch introduced this regression:
commit 6d3d8274bc45de4babb62d64562d92af984dd238
Author: Nick Hoath <[email protected]> AuthorDate: Thu Jan 15 13:10:39 2015 +0000
drm/i915: Subsume intel_ctx_submit_request in to drm_i915_gem_request
Another thing I've noticed is that we explicitly drop the context reference for the request before dropping the request reference. Without clearing the req->ctx pointer. That has a very high chance to leading to tears, imo the context unreferenceing should be pushed into i915_gem_request_free.
Except that it's there already, which means we have a double unref now?
Also this patch is for the legacy ringbuffer code, but the referenced bug is for gen8+ execlists. We're definitely not running this code here I think.
Imo step one is to drop all the explicit ctx refcounting for req->ctx and always rely on the implicit reference. Then see what happens.
Cheers, Daniel
Quoted 25 lines
--- drivers/gpu/drm/i915/i915_drv.c | 2 +- drivers/gpu/drm/i915/i915_gem.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2661,7 +2661,7 @@ static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,intel_lr_context_unpin (ring, submit_req->ctx);i915_gem_context_unreference (submit_req->ctx); -kfree (submit_req); +i915_gem_request_unreference (submit_req); }/* Signature · 6 lines
2.1.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 14 lines · Daniel Vetter
Quoted 3 lines · Nick Hoath
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88652
Signed-off-by Nick Hoath <[email protected]>Commit message is missing the absolutely crucial detail about which patch introduced this regression:
commit
6d3d8274bc45de4babb62d64562d92af984dd238Author: Nick Hoath <[email protected]> AuthorDate: Thu Jan 15 13:10:39 2015 +0000drm/i915: Subsume
intel_ctx_submit_requestin todrm_i915_gem_request
Quoted 7 lines
Another thing I've noticed is that we explicitly drop the context reference for the request before dropping the request reference. Without clearing the
req->ctxpointer. That has a very high chance to leading to tears, imo the context unreferenceing should be pushed intoi915_gem_request_free.Except that it's there already, which means we have a double unref now?
Looking at the code, it looks like that's the case.
Quoted 4 lines
Also this patch is for the legacy ringbuffer code, but the referenced bug is for gen8+ execlists. We're definitely not running this code here I think.
i915_gem_reset_ring_cleanup is used in execlists in the hang recovery case.
Quoted 3 lines
Imo step one is to drop all the explicit ctx refcounting for
req->ctxand always rely on the implicit reference. Then see what happens.
I agree that the refcounting needs re-evaluating after the merge of execlist queue entries & requests, however I think the cleanup of the double unref/removing the refcounting should be done in another patchset. This patch is purely to fix the issue raised in 88652. Depends on the relative priorities.
Quoted 21 lines
Cheers, Daniel
Quoted 17 lines
--- drivers/gpu/drm/i915/i915_drv.c | 2 +- drivers/gpu/drm/i915/i915_gem.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/gpu/drm/i915/i915_gem.c +1
-1@@ -2661,7 +2661,7 @@ static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,intel_lr_context_unpin (ring, submit_req->ctx);i915_gem_context_unreference (submit_req->ctx); -kfree (submit_req); +i915_gem_request_unreference (submit_req); }
Quoted 9 lines
Quoted 8 lines
drivers/gpu/drm/i915/i915_gem.c +0
-1/* -- 2.1.1
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 30 lines · Nick Hoath
Quoted 17 lines · Daniel Vetter
Quoted 4 lines · Nick Hoath
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88652
Signed-off-by Nick Hoath <[email protected]>Commit message is missing the absolutely crucial detail about which patch introduced this regression:
commit
6d3d8274bc45de4babb62d64562d92af984dd238Author: Nick Hoath <[email protected]> AuthorDate: Thu Jan 15 13:10:39 2015 +0000drm/i915: Subsume
intel_ctx_submit_requestin todrm_i915_gem_requestQuoted 7 lines
Another thing I've noticed is that we explicitly drop the context reference for the request before dropping the request reference. Without clearing the
req->ctxpointer. That has a very high chance to leading to tears, imo the context unreferenceing should be pushed intoi915_gem_request_free.Except that it's there already, which means we have a double unref now?
Looking at the code, it looks like that's the case.
Quoted 1 line
Quoted 6 lines
Quoted 3 lines
Also this patch is for the legacy ringbuffer code, but the referenced bug is for gen8+ execlists. We're definitely not running this code here I think.
i915_gem_reset_ring_cleanupis used in execlists in the hang recovery case.
Oops I've missed that, somehow I've thought intel_lrc.c has it's own reset recovery code. But I've just mixed up function names a bit.
Quoted 9 lines
Quoted 2 lines
Imo step one is to drop all the explicit ctx refcounting for
req->ctxand always rely on the implicit reference. Then see what happens.I agree that the refcounting needs re-evaluating after the merge of execlist queue entries & requests, however I think the cleanup of the double unref/removing the refcounting should be done in another patchset. This patch is purely to fix the issue raised in 88652. Depends on the relative priorities.
Well this patch here won't work because there's now a double unref. And I spotted that one because intel_execlists_retire_requests seems to have the exact same double unref already. Hence why I think we should fix up that first and then reasses what's left.
The bug here (before your patch) is just a use-after-free, if there's some other reference to the request. And it will also be fixed with the redone req->ctx refcounting.
-Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Task id: 5767 -------------------------------------Summary-------------------------------------
Platform Delta drm-intel-nightly Series Applied
PNV -1 282/282 281/282
ILK 313/313 313/313 SNB 309/323 309/323 IVB 380/380 380/380 BYT 296/296 296/296 HSW -2 425/425 423/425 BDW -1 318/318 317/318 -------------------------------------Detailed------------------------------------- Platform Test drm-intel-nightly Series Applied *PNV igt_gen3_render_linear_blits PASS(5) CRASH(1)PASS(1) HSW igt_kms_flip_plain-flip-fb-recreate TIMEOUT(2)PASS(1) TIMEOUT(1)PASS(1) HSW igt_kms_flip_plain-flip-fb-recreate-interruptible TIMEOUT(2)PASS(2) TIMEOUT(1)PASS(1) *BDW igt_gem_gtt_hog PASS(8) DMESG_WARN(1)PASS(1)
Note: You need to pay more attention to line start with '*'
_______________________________________________
Intel-gfx mailing list [email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx