From 0ef5c4e4dbbfcebaa9b2eca18097b43016727dfe Mon Sep 17 00:00:00 2001 From: Dave Airlie Date: Fri, 29 Aug 2025 12:16:32 +1000 Subject: nouveau: fix disabling the nonstall irq due to storm code Nouveau has code that when it gets an IRQ with no allowed handler it disables it to avoid storms. However with nonstall interrupts, we often disable them from the drm driver, but still request their emission via the push submission. Just don't disable nonstall irqs ever in normal operation, the event handling code will filter them out, and the driver will just enable/disable them at load time. This fixes timeouts we've been seeing on/off for a long time, but they became a lot more noticeable on Blackwell. This doesn't fix all of them, there is a subsequent fence emission fix to fix the last few. Fixes: 3ebd64aa3c4f ("drm/nouveau/intr: support multiple trees, and explicit interfaces") Cc: stable@vger.kernel.org Signed-off-by: Dave Airlie Link: https://lore.kernel.org/r/20250829021633.1674524-1-airlied@gmail.com [ Fix a typo and a minor checkpatch.pl warning; remove "v2" from commit subject. - Danilo ] Signed-off-by: Danilo Krummrich --- drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c | 2 ++ drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c | 23 ++++++++++++++-------- drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c | 1 + drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h | 2 ++ .../gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fifo.c | 1 + 5 files changed, 21 insertions(+), 8 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c index fdffa0391b31..6fd4e60634fb 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c +++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c @@ -350,6 +350,8 @@ nvkm_fifo_dtor(struct nvkm_engine *engine) nvkm_chid_unref(&fifo->chid); nvkm_event_fini(&fifo->nonstall.event); + if (fifo->func->nonstall_dtor) + fifo->func->nonstall_dtor(fifo); mutex_destroy(&fifo->mutex); if (fifo->func->dtor) diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c index e74493a4569e..6848a56f20c0 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c +++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c @@ -517,19 +517,11 @@ ga100_fifo_nonstall_intr(struct nvkm_inth *inth) static void ga100_fifo_nonstall_block(struct nvkm_event *event, int type, int index) { - struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event); - struct nvkm_runl *runl = nvkm_runl_get(fifo, index, 0); - - nvkm_inth_block(&runl->nonstall.inth); } static void ga100_fifo_nonstall_allow(struct nvkm_event *event, int type, int index) { - struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event); - struct nvkm_runl *runl = nvkm_runl_get(fifo, index, 0); - - nvkm_inth_allow(&runl->nonstall.inth); } const struct nvkm_event_func @@ -564,12 +556,26 @@ ga100_fifo_nonstall_ctor(struct nvkm_fifo *fifo) if (ret) return ret; + nvkm_inth_allow(&runl->nonstall.inth); + nr = max(nr, runl->id + 1); } return nr; } +void +ga100_fifo_nonstall_dtor(struct nvkm_fifo *fifo) +{ + struct nvkm_runl *runl; + + nvkm_runl_foreach(runl, fifo) { + if (runl->nonstall.vector < 0) + continue; + nvkm_inth_block(&runl->nonstall.inth); + } +} + int ga100_fifo_runl_ctor(struct nvkm_fifo *fifo) { @@ -599,6 +605,7 @@ ga100_fifo = { .runl_ctor = ga100_fifo_runl_ctor, .mmu_fault = &tu102_fifo_mmu_fault, .nonstall_ctor = ga100_fifo_nonstall_ctor, + .nonstall_dtor = ga100_fifo_nonstall_dtor, .nonstall = &ga100_fifo_nonstall, .runl = &ga100_runl, .runq = &ga100_runq, diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c index 755235f55b3a..18a0b1f4eab7 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c +++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c @@ -30,6 +30,7 @@ ga102_fifo = { .runl_ctor = ga100_fifo_runl_ctor, .mmu_fault = &tu102_fifo_mmu_fault, .nonstall_ctor = ga100_fifo_nonstall_ctor, + .nonstall_dtor = ga100_fifo_nonstall_dtor, .nonstall = &ga100_fifo_nonstall, .runl = &ga100_runl, .runq = &ga100_runq, diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h index 5e81ae195329..fff1428ef267 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h +++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h @@ -41,6 +41,7 @@ struct nvkm_fifo_func { void (*start)(struct nvkm_fifo *, unsigned long *); int (*nonstall_ctor)(struct nvkm_fifo *); + void (*nonstall_dtor)(struct nvkm_fifo *); const struct nvkm_event_func *nonstall; const struct nvkm_runl_func *runl; @@ -200,6 +201,7 @@ u32 tu102_chan_doorbell_handle(struct nvkm_chan *); int ga100_fifo_runl_ctor(struct nvkm_fifo *); int ga100_fifo_nonstall_ctor(struct nvkm_fifo *); +void ga100_fifo_nonstall_dtor(struct nvkm_fifo *); extern const struct nvkm_event_func ga100_fifo_nonstall; extern const struct nvkm_runl_func ga100_runl; extern const struct nvkm_runq_func ga100_runq; diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fifo.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fifo.c index 1ac5628c5140..4ed54b386a60 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fifo.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fifo.c @@ -601,6 +601,7 @@ r535_fifo_new(const struct nvkm_fifo_func *hw, struct nvkm_device *device, rm->chan.func = &r535_chan; rm->nonstall = &ga100_fifo_nonstall; rm->nonstall_ctor = ga100_fifo_nonstall_ctor; + rm->nonstall_dtor = ga100_fifo_nonstall_dtor; return nvkm_fifo_new_(rm, device, type, inst, pfifo); } -- cgit v1.2.3 From 2cb66ae6040fd3cb058c3391b180f378fc0e3e2f Mon Sep 17 00:00:00 2001 From: Faith Ekstrand Date: Fri, 29 Aug 2025 12:16:33 +1000 Subject: nouveau: Membar before between semaphore writes and the interrupt This ensures that the memory write and the interrupt are properly ordered and we won't wake up the kernel before the semaphore write has hit memory. Fixes: b1ca384772b6 ("drm/nouveau/gv100-: switch to volta semaphore methods") Cc: stable@vger.kernel.org Signed-off-by: Faith Ekstrand Signed-off-by: Dave Airlie Link: https://lore.kernel.org/r/20250829021633.1674524-2-airlied@gmail.com Signed-off-by: Danilo Krummrich --- drivers/gpu/drm/nouveau/gv100_fence.c | 7 +- .../gpu/drm/nouveau/include/nvhw/class/clc36f.h | 85 ++++++++++++++++++++++ 2 files changed, 91 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/nouveau/gv100_fence.c b/drivers/gpu/drm/nouveau/gv100_fence.c index cccdeca72002..317e516c4ec7 100644 --- a/drivers/gpu/drm/nouveau/gv100_fence.c +++ b/drivers/gpu/drm/nouveau/gv100_fence.c @@ -18,7 +18,7 @@ gv100_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence) struct nvif_push *push = &chan->chan.push; int ret; - ret = PUSH_WAIT(push, 8); + ret = PUSH_WAIT(push, 13); if (ret) return ret; @@ -32,6 +32,11 @@ gv100_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence) NVDEF(NVC36F, SEM_EXECUTE, PAYLOAD_SIZE, 32BIT) | NVDEF(NVC36F, SEM_EXECUTE, RELEASE_TIMESTAMP, DIS)); + PUSH_MTHD(push, NVC36F, MEM_OP_A, 0, + MEM_OP_B, 0, + MEM_OP_C, NVDEF(NVC36F, MEM_OP_C, MEMBAR_TYPE, SYS_MEMBAR), + MEM_OP_D, NVDEF(NVC36F, MEM_OP_D, OPERATION, MEMBAR)); + PUSH_MTHD(push, NVC36F, NON_STALL_INTERRUPT, 0); PUSH_KICK(push); diff --git a/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h index 8735dda4c8a7..338f74b9f501 100644 --- a/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h +++ b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h @@ -7,6 +7,91 @@ #define NVC36F_NON_STALL_INTERRUPT (0x00000020) #define NVC36F_NON_STALL_INTERRUPT_HANDLE 31:0 +// NOTE - MEM_OP_A and MEM_OP_B have been replaced in gp100 with methods for +// specifying the page address for a targeted TLB invalidate and the uTLB for +// a targeted REPLAY_CANCEL for UVM. +// The previous MEM_OP_A/B functionality is in MEM_OP_C/D, with slightly +// rearranged fields. +#define NVC36F_MEM_OP_A (0x00000028) +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_TARGET_CLIENT_UNIT_ID 5:0 // only relevant for REPLAY_CANCEL_TARGETED +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_INVALIDATION_SIZE 5:0 // Used to specify size of invalidate, used for invalidates which are not of the REPLAY_CANCEL_TARGETED type +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_TARGET_GPC_ID 10:6 // only relevant for REPLAY_CANCEL_TARGETED +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_CANCEL_MMU_ENGINE_ID 6:0 // only relevant for REPLAY_CANCEL_VA_GLOBAL +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR 11:11 +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR_EN 0x00000001 +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_SYSMEMBAR_DIS 0x00000000 +#define NVC36F_MEM_OP_A_TLB_INVALIDATE_TARGET_ADDR_LO 31:12 +#define NVC36F_MEM_OP_B (0x0000002c) +#define NVC36F_MEM_OP_B_TLB_INVALIDATE_TARGET_ADDR_HI 31:0 +#define NVC36F_MEM_OP_C (0x00000030) +#define NVC36F_MEM_OP_C_MEMBAR_TYPE 2:0 +#define NVC36F_MEM_OP_C_MEMBAR_TYPE_SYS_MEMBAR 0x00000000 +#define NVC36F_MEM_OP_C_MEMBAR_TYPE_MEMBAR 0x00000001 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB 0:0 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ONE 0x00000000 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ALL 0x00000001 // Probably nonsensical for MMU_TLB_INVALIDATE_TARGETED +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC 1:1 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC_ENABLE 0x00000000 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_GPC_DISABLE 0x00000001 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY 4:2 // only relevant if GPC ENABLE +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_NONE 0x00000000 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_START 0x00000001 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_START_ACK_ALL 0x00000002 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_TARGETED 0x00000003 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_GLOBAL 0x00000004 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_REPLAY_CANCEL_VA_GLOBAL 0x00000005 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE 6:5 // only relevant if GPC ENABLE +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_NONE 0x00000000 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_GLOBALLY 0x00000001 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACK_TYPE_INTRANODE 0x00000002 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE 9:7 //only relevant for REPLAY_CANCEL_VA_GLOBAL +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_READ 0 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_WRITE 1 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_STRONG 2 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_RSVRVD 3 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_WEAK 4 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ATOMIC_ALL 5 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_WRITE_AND_ATOMIC 6 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_ACCESS_TYPE_VIRT_ALL 7 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL 9:7 // Invalidate affects this level and all below +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_ALL 0x00000000 // Invalidate tlb caches at all levels of the page table +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_PTE_ONLY 0x00000001 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE0 0x00000002 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE1 0x00000003 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE2 0x00000004 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE3 0x00000005 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE4 0x00000006 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PAGE_TABLE_LEVEL_UP_TO_PDE5 0x00000007 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE 11:10 // only relevant if PDB_ONE +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_VID_MEM 0x00000000 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_SYS_MEM_COHERENT 0x00000002 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_APERTURE_SYS_MEM_NONCOHERENT 0x00000003 +#define NVC36F_MEM_OP_C_TLB_INVALIDATE_PDB_ADDR_LO 31:12 // only relevant if PDB_ONE +#define NVC36F_MEM_OP_C_ACCESS_COUNTER_CLR_TARGETED_NOTIFY_TAG 19:0 +// MEM_OP_D MUST be preceded by MEM_OPs A-C. +#define NVC36F_MEM_OP_D (0x00000034) +#define NVC36F_MEM_OP_D_TLB_INVALIDATE_PDB_ADDR_HI 26:0 // only relevant if PDB_ONE +#define NVC36F_MEM_OP_D_OPERATION 31:27 +#define NVC36F_MEM_OP_D_OPERATION_MEMBAR 0x00000005 +#define NVC36F_MEM_OP_D_OPERATION_MMU_TLB_INVALIDATE 0x00000009 +#define NVC36F_MEM_OP_D_OPERATION_MMU_TLB_INVALIDATE_TARGETED 0x0000000a +#define NVC36F_MEM_OP_D_OPERATION_L2_PEERMEM_INVALIDATE 0x0000000d +#define NVC36F_MEM_OP_D_OPERATION_L2_SYSMEM_INVALIDATE 0x0000000e +// CLEAN_LINES is an alias for Tegra/GPU IP usage +#define NVC36F_MEM_OP_B_OPERATION_L2_INVALIDATE_CLEAN_LINES 0x0000000e +#define NVC36F_MEM_OP_D_OPERATION_L2_CLEAN_COMPTAGS 0x0000000f +#define NVC36F_MEM_OP_D_OPERATION_L2_FLUSH_DIRTY 0x00000010 +#define NVC36F_MEM_OP_D_OPERATION_L2_WAIT_FOR_SYS_PENDING_READS 0x00000015 +#define NVC36F_MEM_OP_D_OPERATION_ACCESS_COUNTER_CLR 0x00000016 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE 1:0 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_MIMC 0x00000000 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_MOMC 0x00000001 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_ALL 0x00000002 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TYPE_TARGETED 0x00000003 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE 2:2 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE_MIMC 0x00000000 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_TYPE_MOMC 0x00000001 +#define NVC36F_MEM_OP_D_ACCESS_COUNTER_CLR_TARGETED_BANK 6:3 #define NVC36F_SEM_ADDR_LO (0x0000005c) #define NVC36F_SEM_ADDR_LO_OFFSET 31:2 #define NVC36F_SEM_ADDR_HI (0x00000060) -- cgit v1.2.3 From 69a79ada8eb034ce016b5b78fb7d08d8687223de Mon Sep 17 00:00:00 2001 From: Karol Wachowski Date: Fri, 8 Aug 2025 13:09:39 +0200 Subject: accel/ivpu: Prevent recovery work from being queued during device removal Use disable_work_sync() instead of cancel_work_sync() in ivpu_dev_fini() to ensure that no new recovery work items can be queued after device removal has started. Previously, recovery work could be scheduled even after canceling existing work, potentially leading to use-after-free bugs if recovery accessed freed resources. Rename ivpu_pm_cancel_recovery() to ivpu_pm_disable_recovery() to better reflect its new behavior. Fixes: 58cde80f45a2 ("accel/ivpu: Use dedicated work for job timeout detection") Cc: stable@vger.kernel.org # v6.8+ Signed-off-by: Karol Wachowski Reviewed-by: Lizhi Hou Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250808110939.328366-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_drv.c | 2 +- drivers/accel/ivpu/ivpu_pm.c | 4 ++-- drivers/accel/ivpu/ivpu_pm.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_drv.c b/drivers/accel/ivpu/ivpu_drv.c index 3d6d52492536..3289751b4757 100644 --- a/drivers/accel/ivpu/ivpu_drv.c +++ b/drivers/accel/ivpu/ivpu_drv.c @@ -677,7 +677,7 @@ static void ivpu_bo_unbind_all_user_contexts(struct ivpu_device *vdev) static void ivpu_dev_fini(struct ivpu_device *vdev) { ivpu_jobs_abort_all(vdev); - ivpu_pm_cancel_recovery(vdev); + ivpu_pm_disable_recovery(vdev); ivpu_pm_disable(vdev); ivpu_prepare_for_reset(vdev); ivpu_shutdown(vdev); diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c index eacda1dbe840..475ddc94f1cf 100644 --- a/drivers/accel/ivpu/ivpu_pm.c +++ b/drivers/accel/ivpu/ivpu_pm.c @@ -417,10 +417,10 @@ void ivpu_pm_init(struct ivpu_device *vdev) ivpu_dbg(vdev, PM, "Autosuspend delay = %d\n", delay); } -void ivpu_pm_cancel_recovery(struct ivpu_device *vdev) +void ivpu_pm_disable_recovery(struct ivpu_device *vdev) { drm_WARN_ON(&vdev->drm, delayed_work_pending(&vdev->pm->job_timeout_work)); - cancel_work_sync(&vdev->pm->recovery_work); + disable_work_sync(&vdev->pm->recovery_work); } void ivpu_pm_enable(struct ivpu_device *vdev) diff --git a/drivers/accel/ivpu/ivpu_pm.h b/drivers/accel/ivpu/ivpu_pm.h index 89b264cc0e3e..a2aa7a27f32e 100644 --- a/drivers/accel/ivpu/ivpu_pm.h +++ b/drivers/accel/ivpu/ivpu_pm.h @@ -25,7 +25,7 @@ struct ivpu_pm_info { void ivpu_pm_init(struct ivpu_device *vdev); void ivpu_pm_enable(struct ivpu_device *vdev); void ivpu_pm_disable(struct ivpu_device *vdev); -void ivpu_pm_cancel_recovery(struct ivpu_device *vdev); +void ivpu_pm_disable_recovery(struct ivpu_device *vdev); int ivpu_pm_suspend_cb(struct device *dev); int ivpu_pm_resume_cb(struct device *dev); -- cgit v1.2.3 From 232674e1a6dd2f7a68b0d496a7ed1a57d79533da Mon Sep 17 00:00:00 2001 From: Pierre-Eric Pelloux-Prayer Date: Mon, 1 Sep 2025 14:40:32 +0200 Subject: drm/sched: Fix racy access to drm_sched_entity.dependency The drm_sched_job_unschedulable trace point can access entity->dependency after it was cleared by the callback installed in drm_sched_entity_add_dependency_cb, causing: BUG: kernel NULL pointer dereference, address: 0000000000000020 [...] Workqueue: comp_1.1.0 drm_sched_run_job_work [gpu_sched] RIP: 0010:trace_event_raw_event_drm_sched_job_unschedulable+0x70/0xd0 [gpu_sched] To fix this we either need to keep a reference to the fence before setting up the callbacks, or move the trace_drm_sched_job_unschedulable calls into drm_sched_entity_add_dependency_cb where they can be done earlier. Fixes: 76d97c870f29 ("drm/sched: Trace dependencies for GPU jobs") Signed-off-by: Pierre-Eric Pelloux-Prayer Reviewed-by: Tvrtko Ursulin Signed-off-by: Philipp Stanner Link: https://lore.kernel.org/r/20250901124032.1955-1-pierre-eric.pelloux-prayer@amd.com (cherry picked from commit b2b8af21fec35be417a3199b5a6c354605dd222a) Signed-off-by: Maxime Ripard --- drivers/gpu/drm/scheduler/sched_entity.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c index 8867b95ab089..3d06f72531ba 100644 --- a/drivers/gpu/drm/scheduler/sched_entity.c +++ b/drivers/gpu/drm/scheduler/sched_entity.c @@ -391,7 +391,8 @@ EXPORT_SYMBOL(drm_sched_entity_set_priority); * Add a callback to the current dependency of the entity to wake up the * scheduler when the entity becomes available. */ -static bool drm_sched_entity_add_dependency_cb(struct drm_sched_entity *entity) +static bool drm_sched_entity_add_dependency_cb(struct drm_sched_entity *entity, + struct drm_sched_job *sched_job) { struct drm_gpu_scheduler *sched = entity->rq->sched; struct dma_fence *fence = entity->dependency; @@ -421,6 +422,10 @@ static bool drm_sched_entity_add_dependency_cb(struct drm_sched_entity *entity) entity->dependency = fence; } + if (trace_drm_sched_job_unschedulable_enabled() && + !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &entity->dependency->flags)) + trace_drm_sched_job_unschedulable(sched_job, entity->dependency); + if (!dma_fence_add_callback(entity->dependency, &entity->cb, drm_sched_entity_wakeup)) return true; @@ -461,10 +466,8 @@ struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity) while ((entity->dependency = drm_sched_job_dependency(sched_job, entity))) { - if (drm_sched_entity_add_dependency_cb(entity)) { - trace_drm_sched_job_unschedulable(sched_job, entity->dependency); + if (drm_sched_entity_add_dependency_cb(entity, sched_job)) return NULL; - } } /* skip jobs from entity that marked guilty */ -- cgit v1.2.3 From 304f455b648d68ba47135583b644ee72ac48d544 Mon Sep 17 00:00:00 2001 From: James Jones Date: Tue, 26 Aug 2025 12:57:16 -0700 Subject: MAINTAINERS: Update git entry for nouveau The gitlab repository previously associated with the nouveau module has fallen out of use. The drm-misc tree here: https://gitlab.freedesktop.org/drm/misc/kernel.git Is now where most nouveau-related patches are applied. This change updates the MAINTAINERS file to reflect this. Signed-off-by: James Jones Link: https://lore.kernel.org/r/20250826195716.1897-1-jajones@nvidia.com Signed-off-by: Danilo Krummrich --- MAINTAINERS | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/MAINTAINERS b/MAINTAINERS index 3f1c4af3f8e0..f94e115a8d32 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -7820,7 +7820,7 @@ Q: https://patchwork.freedesktop.org/project/nouveau/ Q: https://gitlab.freedesktop.org/drm/nouveau/-/merge_requests B: https://gitlab.freedesktop.org/drm/nouveau/-/issues C: irc://irc.oftc.net/nouveau -T: git https://gitlab.freedesktop.org/drm/nouveau.git +T: git https://gitlab.freedesktop.org/drm/misc/kernel.git F: drivers/gpu/drm/nouveau/ F: include/uapi/drm/nouveau_drm.h -- cgit v1.2.3 From bdd5a14e660062114bdebaef9ad52adf04970a89 Mon Sep 17 00:00:00 2001 From: Michael Walle Date: Thu, 21 Aug 2025 14:23:41 +0200 Subject: drm/bridge: ti-sn65dsi86: fix REFCLK setting The bridge has three bootstrap pins which are sampled to determine the frequency of the external reference clock. The driver will also (over)write that setting. But it seems this is racy after the bridge is enabled. It was observed that although the driver write the correct value (by sniffing on the I2C bus), the register has the wrong value. The datasheet states that the GPIO lines have to be stable for at least 5us after asserting the EN signal. Thus, there seems to be some logic which samples the GPIO lines and this logic appears to overwrite the register value which was set by the driver. Waiting 20us after asserting the EN line resolves this issue. Fixes: a095f15c00e2 ("drm/bridge: add support for sn65dsi86 bridge driver") Signed-off-by: Michael Walle Reviewed-by: Douglas Anderson Signed-off-by: Douglas Anderson Link: https://lore.kernel.org/r/20250821122341.1257286-1-mwalle@kernel.org --- drivers/gpu/drm/bridge/ti-sn65dsi86.c | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/drivers/gpu/drm/bridge/ti-sn65dsi86.c b/drivers/gpu/drm/bridge/ti-sn65dsi86.c index 464390372b34..ae0d08e5e960 100644 --- a/drivers/gpu/drm/bridge/ti-sn65dsi86.c +++ b/drivers/gpu/drm/bridge/ti-sn65dsi86.c @@ -392,6 +392,17 @@ static int __maybe_unused ti_sn65dsi86_resume(struct device *dev) gpiod_set_value_cansleep(pdata->enable_gpio, 1); + /* + * After EN is deasserted and an external clock is detected, the bridge + * will sample GPIO3:1 to determine its frequency. The driver will + * overwrite this setting in ti_sn_bridge_set_refclk_freq(). But this is + * racy. Thus we have to wait a couple of us. According to the datasheet + * the GPIO lines has to be stable at least 5 us (td5) but it seems that + * is not enough and the refclk frequency value is still lost or + * overwritten by the bridge itself. Waiting for 20us seems to work. + */ + usleep_range(20, 30); + /* * If we have a reference clock we can enable communication w/ the * panel (including the aux channel) w/out any need for an input clock -- cgit v1.2.3