Changes in amdgpu_eviction_fence.c

a3057aa926ad024e79c6652d891f311d6d0faa5c - drm/amdgpu/userq: schedule_delayed_work should be after fence signalled

Tue, 24 Mar 2026 16:16:34 +0100

drm/amdgpu/userq: schedule_delayed_work should be after fence signalledReorganise the amdgpu_eviction_fence_suspend_worker code soschedule_delayed_work is the last thing we do after amdgpu_userq_evictis complete and the eviction fence is signalled.Suggested-by: Christian König Signed-off-by: Sunil Khatri Reviewed-by: Christian König Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

fc61df1516178958d7e61ec52c05257d92e9f2e8 - drm/amdgpu: annotate eviction fence signaling path

Mon, 02 Feb 2026 13:06:25 +0100

drm/amdgpu: annotate eviction fence signaling pathMake sure lockdep sees the dependencies here.Signed-off-by: Christian König Reviewed-by: Sunil Khatri Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

2c192b06f1e6bd9f770138a55b26c4ea04246e2f - drm/amdgpu: fix adding eviction fence

Wed, 28 Jan 2026 16:07:03 +0100

drm/amdgpu: fix adding eviction fenceWe can't add the eviction fence without validating the BO.Signed-off-by: Christian König Reviewed-by: Sunil Khatri Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

99f30a0607c6161309613b714e14470f5fc7ce41 - drm/amdgpu: fix eviction fence and userq manager shutdown

Fri, 30 Jan 2026 16:46:36 +0100

drm/amdgpu: fix eviction fence and userq manager shutdownThat is a really complicated dance and wasn't implemented fully correct.Signed-off-by: Christian König Reviewed-by: Sunil Khatri Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

2cd7284ba54b22869c301fba7aff9ec96a12f8c0 - drm/amdgpu: completely rework eviction fence handling v2

Wed, 28 Jan 2026 13:58:14 +0100

drm/amdgpu: completely rework eviction fence handling v2Well that was broken on multiple levels.First of all a lot of checks were placed at incorrect locations, especially ifthe resume worker should run or not.Then a bunch of code was just mid-layering because of incorrect assignment whoshould do what.And finally comments explaining what happens instead of why.Just re-write it from scratch, that should at least fix some of the hangs weare seeing.Use RCU for the eviction fence pointer in the manager, the spinlock usage wasmostly incorrect as well. Then finally remove all the nonsense checks andactually add them in the correct locations.v2: some typo fixes and cleanups suggested by SunilSigned-off-by: Christian König Reviewed-by: Sunil Khatri Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

bf4afc53b77aeaa48b5409da5c8da6bb4eff7f43 - Convert 'alloc_obj' family to use the new default GFP_KERNEL argument

Sun, 22 Feb 2026 01:37:42 +0100

Convert 'alloc_obj' family to use the new default GFP_KERNEL argumentThis was done entirely with mindless brute force, using git grep -l '\ List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

69050f8d6d075dc01af7a5f2f550a8067510366f - treewide: Replace kmalloc with kmalloc_obj for non-scalar types

Sat, 21 Feb 2026 08:49:23 +0100

treewide: Replace kmalloc with kmalloc_obj for non-scalar typesThis is the result of running the Coccinelle script fromscripts/coccinelle/api/kmalloc_objs.cocci. The script is designed toavoid scalar types (which need careful case-by-case checking), andinstead replace kmalloc-family calls that allocate struct or unionobject instances:Single allocations: kmalloc(sizeof(TYPE), ...)are replaced with: kmalloc_obj(TYPE, ...)Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)are replaced with: kmalloc_objs(TYPE, COUNT, ...)Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)(where TYPE may also be *VAR)The resulting allocations no longer return "void *", instead returning"TYPE *".Signed-off-by: Kees Cook List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

c598d5eb9fb331ba17bc9ad67ae9a2231ca5aca5 - Merge drm/drm-next into drm-misc-next

Wed, 11 Jun 2025 09:01:34 +0200

Merge drm/drm-next into drm-misc-nextBackmerging to forward to v6.16-rc1Signed-off-by: Thomas Zimmermann List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

bf33a0003d9e3b0546f2d7e91bebfd67af59f275 - dma-fence: Use a flag for 64-bit seqnos

Thu, 15 May 2025 11:49:57 +0200

dma-fence: Use a flag for 64-bit seqnosWith the goal of reducing the need for drivers to touch (and dereference)fence->ops, we move the 64-bit seqnos flag from struct dma_fence_ops tothe fence->flags.Drivers which were setting this flag are changed to use newdma_fence_init64() instead of dma_fence_init().v2: * Streamlined init and added kerneldoc. * Rebase for amdgpu userq which landed since.Signed-off-by: Tvrtko Ursulin Reviewed-by: Christian König # v1Signed-off-by: Tvrtko Ursulin Link: https://lore.kernel.org/r/20250515095004.28318-3-tvrtko.ursulin@igalia.com List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

b2c11e27080d8556664c20c30ca3527ffa99bec4 - drm/amdgpu: lock the eviction fence for wq signals it

Wed, 30 Apr 2025 04:34:14 +0200

drm/amdgpu: lock the eviction fence for wq signals itLock and refer to the eviction fence before the eviction fenceschedules work queue tries to signal it.Suggested-by: Christian König Signed-off-by: Prike Liang Acked-by: Alex Deucher Reviewed-by: Arvind Yadav Reviewed-by: Christian König Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

13d0724f0ffe92971ef5a0da819f75c76c01d92e - drm/amdgpu: fix use-after-unlock in eviction fence destroy

Thu, 15 May 2025 09:14:46 +0200

drm/amdgpu: fix use-after-unlock in eviction fence destroyThe eviction fence destroy path incorrectly calls dma_fence_put() onevf_mgr->ev_fence after releasing the ev_fence_lock. This introduces apotential use-after-unlock or race because another thread concurrentlymodifies evf_mgr->ev_fence.Fix this by grabbing a local reference to evf_mgr->ev_fence under thelock and using that for dma_fence_put() after waiting.Cc: Alex Deucher Reviewed-by: Christian König Reviewed-by: Sunil Khatri Signed-off-by: Arvind Yadav Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

af7160c25c68520299db40d608892810904fa064 - drm/amdgpu: fix the eviction fence dereference

Tue, 15 Apr 2025 04:42:19 +0200

drm/amdgpu: fix the eviction fence dereferenceThe dma_resv_add_fence() already refers to the added fence.So when attaching the evciton fence to the gem bo, it needn'trefer to it anymore.Signed-off-by: Prike Liang Reviewed-by: Christian König Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

e125a6e8ce0d747f4ae62234587308cfbbdcf4fa - drm/amdgpu: set the evf name to identify the userq case

Tue, 15 Apr 2025 04:31:15 +0200

drm/amdgpu: set the evf name to identify the userq caseThe evf fence name can clearly identify the userq usage.Signed-off-by: Prike Liang Reviewed-by: Christian König Reviewed-by: Arvind Yadav Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

42a66677805d03df9e2600fab82d0cbe855500e1 - drm/amdgpu/userq: use consistent function naming

Wed, 16 Apr 2025 23:49:45 +0200

drm/amdgpu/userq: use consistent function namings/userqueue/userq/1. remove the mix of amdgpu_userqueue and amdgpu_userq2. to be consistent with other amdgpu_userq_fence.c3. it's shorterReviewed-by: Prike Liang Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

4fdbe3a623b264d0f361a81a41baff16b4aa6bb1 - drm/amdgpu/userq: rename eviction helpers

Wed, 16 Apr 2025 23:27:41 +0200

drm/amdgpu/userq: rename eviction helperssuspend/resume -> evict/restoreRename to avoid confusion with the system suspendand resume helpers.v2: update error messagesReviewed-by: Prike Liang Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

32bd8b3ea7071ca96d4adcd0817046e09c5df415 - drm/amdgpu: Fix display freezing issue when resizing apps

Tue, 18 Mar 2025 14:15:40 +0100

drm/amdgpu: Fix display freezing issue when resizing appsThe display is freezing because the amdgpu_userq_wait_ioctl()is waiting for a non-user queue fence(specifically, the PT update fence).RootCause:The resume_work is initiated by both amdgpu_userq_suspend andamdgpu_userqueue_ensure_ev_fence at same time. The amdgpu_userq_suspendsignals a dma-fence and subsequently triggers the resume_work, which isintended to replace the existing fence by creating new dma-fence. However,following this, the amdgpu_userqueue_ensure_ev_fence schedules anotherresume_work that generates a new dma-fence, thereby replacing the onecreated by amdgpu_userq_suspend. Consequently, the original fence willnever be signaled.Cc: Alex Deucher Cc: Christian König Cc: Shashank Sharma Cc: Sunil Khatri Reviewed-by: Christian König Reviewed-by: Shashank Sharma Signed-off-by: Arvind Yadav Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

31f7efcdca4d1caaa3a0babc33377e27e6f9b593 - drm/amdgpu: enable eviction fence

Wed, 20 Nov 2024 18:45:33 +0100

drm/amdgpu: enable eviction fenceThis patch enables attachment and detachment of eviction fences.This is just a fork of eviction fence enabling code from the firstpatch of the series so that the CI testing can happen on fullyfledged code.Cc: Alex Deucher Cc: Christian Koenig Reviewed-by: Christian Koenig Signed-off-by: Shashank Sharma Signed-off-by: Arvind Yadav Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

a242a3e4b5be22ffd85fb768d8c96df79b018136 - drm/amdgpu: simplify eviction fence suspend/resume

Wed, 11 Dec 2024 12:09:00 +0100

drm/amdgpu: simplify eviction fence suspend/resumeThe basic idea in this redesign is to add an eviction fence only in UQresume path. When userqueue is not present, keep ev_fence as NULLMain changes are: - do not create the eviction fence during evf_mgr_init, keeping evf_mgr->ev_fence=NULL until UQ get active. - do not replace the ev_fence in evf_resume path, but replace it only in uq_resume path, so remove all the unnecessary code from ev_fence_resume. - add a new helper function (amdgpu_userqueue_ensure_ev_fence) which will do the following: - flush any pending uq_resume work, so that it could create an eviction_fence - if there is no pending uq_resume_work, add a uq_resume work and wait for it to execute so that we always have a valid ev_fence - call this helper function from two places, to ensure we have a valid ev_fence: - when a new uq is created - when a new uq completion fence is createdv2: Worked on review comments by Christian.v3: Addressed few more review comments by Christian.v4: Move mutex lock outside of the amdgpu_userqueue_suspend() function (Christian).v5: squash in build fix (Alex)Cc: Alex Deucher Reviewed-by: Christian König Signed-off-by: Arvind Yadav Signed-off-by: Shashank Sharma Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

f2234816a31d0ec85ab63899762ec962ab682704 - drm/amdgpu: fix IGT CI regression with eviction fence

Wed, 27 Nov 2024 17:06:45 +0100

drm/amdgpu: fix IGT CI regression with eviction fenceThis patch fixes one of the regressions in eviction fence code withIGT tests.Reviewed-by: Shashank Sharma Signed-off-by: Amaranath Somalapuram Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c

b8e6d3f68c3bd1ac54492e210ece87475e7f862b - drm/amdgpu: handle eviction fence race

Wed, 20 Nov 2024 18:04:33 +0100

drm/amdgpu: handle eviction fence raceThe eviction process can get into a race condition between the evictionfence suspend work (which replaces the old fence with new) and kms_close(which destroys the fence and doesn't expect a new one).This patch:- adds a flag to indicate that fd is closing, so fence replacement is not required (evf_mgr->fd_closing)- adds a flush_work() during the ev_fence_destroy routineV2: Addressed review comments from Christian: - Do not use mutex to sync - Use flush_work and wait for suspend_work to be doneV3: Fixed state machine for queue->active, which adds into race between suspend/resume and queue opsCc: Alex Deucher Cc: Christian König Reviewed-by: Christian König Signed-off-by: Shashank Sharma Signed-off-by: Arvind Yadav Signed-off-by: Alex Deucher List of files: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c