From ca9e5115e870b9a531deb02752055a8a587904e3 Mon Sep 17 00:00:00 2001 From: Matthew Brost Date: Wed, 7 Jan 2026 12:57:32 -0800 Subject: drm/xe: Adjust page count tracepoints in shrinker MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Page accounting can change via the shrinker without calling xe_ttm_tt_unpopulate(), which normally updates page count tracepoints through update_global_total_pages. Add a call to update_global_total_pages when the shrinker successfully shrinks a BO. v2: - Don't adjust global accounting when pinning (Stuart) Cc: stable@vger.kernel.org Fixes: ce3d39fae3d3 ("drm/xe/bo: add GPU memory trace points") Signed-off-by: Matthew Brost Reviewed-by: Stuart Summers Link: https://patch.msgid.link/20260107205732.2267541-1-matthew.brost@intel.com (cherry picked from commit cc54eabdfbf0c5b6638edc50002cfafac1f1e18b) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_bo.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index bf4ee976b680..71acd45aa33b 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -1055,6 +1055,7 @@ static long xe_bo_shrink_purge(struct ttm_operation_ctx *ctx, unsigned long *scanned) { struct xe_device *xe = ttm_to_xe_device(bo->bdev); + struct ttm_tt *tt = bo->ttm; long lret; /* Fake move to system, without copying data. */ @@ -1079,8 +1080,10 @@ static long xe_bo_shrink_purge(struct ttm_operation_ctx *ctx, .writeback = false, .allow_move = false}); - if (lret > 0) + if (lret > 0) { xe_ttm_tt_account_subtract(xe, bo->ttm); + update_global_total_pages(bo->bdev, -(long)tt->num_pages); + } return lret; } @@ -1166,8 +1169,10 @@ long xe_bo_shrink(struct ttm_operation_ctx *ctx, struct ttm_buffer_object *bo, if (needs_rpm) xe_pm_runtime_put(xe); - if (lret > 0) + if (lret > 0) { xe_ttm_tt_account_subtract(xe, tt); + update_global_total_pages(bo->bdev, -(long)tt->num_pages); + } out_unref: xe_bo_put(xe_bo); -- cgit v1.2.3 From 6f287b1c8d0e255e94e54116ebbe126515f5c911 Mon Sep 17 00:00:00 2001 From: Marco Crivellari Date: Thu, 8 Jan 2026 19:01:48 +0100 Subject: drm/xe: fix WQ_MEM_RECLAIM passed as max_active to alloc_workqueue() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Workqueue xe-ggtt-wq has been allocated using WQ_MEM_RECLAIM, but the flag has been passed as 3rd parameter (max_active) instead of 2nd (flags) creating the workqueue as per-cpu with max_active = 8 (the WQ_MEM_RECLAIM value). So change this by set WQ_MEM_RECLAIM as the 2nd parameter with a default max_active. Fixes: 60df57e496e4 ("drm/xe: Mark GGTT work queue with WQ_MEM_RECLAIM") Cc: stable@vger.kernel.org Signed-off-by: Marco Crivellari Reviewed-by: Matthew Brost Signed-off-by: Matthew Brost Link: https://patch.msgid.link/20260108180148.423062-1-marco.crivellari@suse.com (cherry picked from commit aa39abc08e77d66ebb0c8c9ec4cc8d38ded34dc9) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_ggtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c index ef481b334af4..793d7324a395 100644 --- a/drivers/gpu/drm/xe/xe_ggtt.c +++ b/drivers/gpu/drm/xe/xe_ggtt.c @@ -322,7 +322,7 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt) else ggtt->pt_ops = &xelp_pt_ops; - ggtt->wq = alloc_workqueue("xe-ggtt-wq", 0, WQ_MEM_RECLAIM); + ggtt->wq = alloc_workqueue("xe-ggtt-wq", WQ_MEM_RECLAIM, 0); if (!ggtt->wq) return -ENOMEM; -- cgit v1.2.3 From a009bbb9586119047a071971a48b093fcc65f33d Mon Sep 17 00:00:00 2001 From: Jani Nikula Date: Wed, 7 Jan 2026 17:53:58 +0200 Subject: drm/xe/vf: fix struct xe_gt_sriov_vf_migration kernel-doc MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix kernel-doc warnings on struct xe_gt_sriov_vf_migration: Warning: ../drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h:47 cannot understand function prototype: 'struct xe_gt_sriov_vf_migration' Fixes: e1d2e2d878bf ("drm/xe/vf: Add xe_gt_recovery_pending helper") Cc: Matthew Brost Cc: Michal Wajdeczko Cc: Tomasz Lis Reviewed-by: Matt Roper Link: https://patch.msgid.link/20260107155401.2379127-2-jani.nikula@intel.com Signed-off-by: Jani Nikula (cherry picked from commit 44393331c79f5df14c1ff25f4a355f439a2dc8a2) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h b/drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h index 420b0e6089de..e8897a77ba19 100644 --- a/drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h +++ b/drivers/gpu/drm/xe/xe_gt_sriov_vf_types.h @@ -41,10 +41,10 @@ struct xe_gt_sriov_vf_runtime { }; /** - * xe_gt_sriov_vf_migration - VF migration data. + * struct xe_gt_sriov_vf_migration - VF migration data. */ struct xe_gt_sriov_vf_migration { - /** @migration: VF migration recovery worker */ + /** @worker: VF migration recovery worker */ struct work_struct worker; /** @lock: Protects recovery_queued, teardown */ spinlock_t lock; -- cgit v1.2.3 From dc1d0ffee09740088eb190af84a2c470d279bad9 Mon Sep 17 00:00:00 2001 From: Jani Nikula Date: Wed, 7 Jan 2026 17:53:59 +0200 Subject: drm/xe/xe_late_bind_fw: fix enum xe_late_bind_fw_id kernel-doc MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix kernel-doc warnings on enum xe_late_bind_fw_id: Warning: ../drivers/gpu/drm/xe/xe_late_bind_fw_types.h:19 cannot understand function prototype: 'enum xe_late_bind_fw_id' Fixes: 45832bf9c10f ("drm/xe/xe_late_bind_fw: Initialize late binding firmware") Cc: Badal Nilawar Cc: Daniele Ceraolo Spurio Cc: Rodrigo Vivi Reviewed-by: Badal Nilawar Link: https://patch.msgid.link/20260107155401.2379127-3-jani.nikula@intel.com Signed-off-by: Jani Nikula (cherry picked from commit a857e6102970c7bd8f2db967fe02d76741179d14) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_late_bind_fw_types.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_late_bind_fw_types.h b/drivers/gpu/drm/xe/xe_late_bind_fw_types.h index 0f5da89ce98b..2a8a985c37e7 100644 --- a/drivers/gpu/drm/xe/xe_late_bind_fw_types.h +++ b/drivers/gpu/drm/xe/xe_late_bind_fw_types.h @@ -15,10 +15,12 @@ #define XE_LB_MAX_PAYLOAD_SIZE SZ_4K /** - * xe_late_bind_fw_id - enum to determine late binding fw index + * enum xe_late_bind_fw_id - enum to determine late binding fw index */ enum xe_late_bind_fw_id { + /** @XE_LB_FW_FAN_CONTROL: Fan control */ XE_LB_FW_FAN_CONTROL = 0, + /** @XE_LB_FW_MAX_ID: Number of IDs */ XE_LB_FW_MAX_ID }; -- cgit v1.2.3 From 47bf28e22a121b807a9a9680c4209846a78a98a6 Mon Sep 17 00:00:00 2001 From: Jani Nikula Date: Wed, 7 Jan 2026 17:54:00 +0200 Subject: drm/xe/vm: fix xe_vm_validation_exec() kernel-doc MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix kernel-doc warnings on xe_vm_validation_exec(): Warning: ../drivers/gpu/drm/xe/xe_vm.h:392 expecting prototype for xe_vm_set_validation_exec(). Prototype was for xe_vm_validation_exec() instead Fixes: 0131514f9789 ("drm/xe: Pass down drm_exec context to validation") Cc: Thomas Hellström Cc: Matthew Brost Reviewed-by: Matt Roper Link: https://patch.msgid.link/20260107155401.2379127-4-jani.nikula@intel.com Signed-off-by: Jani Nikula (cherry picked from commit b3a7767989e6519127ac5e0cde682c50ad587f3b) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_vm.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h index ef8a5019574e..016f6786134c 100644 --- a/drivers/gpu/drm/xe/xe_vm.h +++ b/drivers/gpu/drm/xe/xe_vm.h @@ -379,7 +379,7 @@ static inline void xe_vm_set_validation_exec(struct xe_vm *vm, struct drm_exec * } /** - * xe_vm_set_validation_exec() - Accessor to read the drm_exec object + * xe_vm_validation_exec() - Accessor to read the drm_exec object * @vm: The vm we want to register a drm_exec object with. * * Return: The drm_exec object used to lock the vm's resv. The value -- cgit v1.2.3 From b886aa65eafe3098bbd691f0ca4a9abce03f9d03 Mon Sep 17 00:00:00 2001 From: Matthew Brost Date: Fri, 9 Jan 2026 17:27:38 -0800 Subject: drm/xe: Disable timestamp WA on VFs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The timestamp WA does not work on a VF because it requires reading MMIO registers, which are inaccessible on a VF. This timestamp WA confuses LRC sampling on a VF during TDR, as the LRC timestamp would always read as 1 for any active context. Disable the timestamp WA on VFs to avoid this confusion. Signed-off-by: Matthew Brost Reviewed-by: Umesh Nerlige Ramappa Fixes: 617d824c5323 ("drm/xe: Add WA BB to capture active context utilization") Link: https://patch.msgid.link/20260110012739.2888434-7-matthew.brost@intel.com (cherry picked from commit efffd56e4bd894e0935eea00e437f233b6cebc0d) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_lrc.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c index b5083c99dd50..281286f2b5f9 100644 --- a/drivers/gpu/drm/xe/xe_lrc.c +++ b/drivers/gpu/drm/xe/xe_lrc.c @@ -1050,6 +1050,9 @@ static ssize_t setup_utilization_wa(struct xe_lrc *lrc, { u32 *cmd = batch; + if (IS_SRIOV_VF(gt_to_xe(lrc->gt))) + return 0; + if (xe_gt_WARN_ON(lrc->gt, max_len < 12)) return -ENOSPC; -- cgit v1.2.3 From 6f4b7aed61817624250e590ba0ef304146d34614 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Tue, 20 Jan 2026 11:06:10 +0000 Subject: drm/xe/uapi: disallow bind queue sharing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Currently this is very broken if someone attempts to create a bind queue and share it across multiple VMs. For example currently we assume it is safe to acquire the user VM lock to protect some of the bind queue state, but if allow sharing the bind queue with multiple VMs then this quickly breaks down. To fix this reject using a bind queue with any VM that is not the same VM that was originally passed when creating the bind queue. This a uAPI change, however this was more of an oversight on kernel side that we didn't reject this, and expectation is that userspace shouldn't be using bind queues in this way, so in theory this change should go unnoticed. Based on a patch from Matt Brost. v2 (Matt B): - Hold the vm lock over queue create, to ensure it can't be closed as we attach the user_vm to the queue. - Make sure we actually check for NULL user_vm in destruction path. v3: - Fix error path handling. Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs") Reported-by: Thomas Hellström Signed-off-by: Matthew Auld Cc: José Roberto de Souza Cc: Matthew Brost Cc: Michal Mrozek Cc: Carl Zhang Cc: # v6.8+ Acked-by: José Roberto de Souza Reviewed-by: Matthew Brost Reviewed-by: Arvind Yadav Acked-by: Michal Mrozek Link: https://patch.msgid.link/20260120110609.77958-3-matthew.auld@intel.com (cherry picked from commit 9dd08fdecc0c98d6516c2d2d1fa189c1332f8dab) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_exec_queue.c | 32 +++++++++++++++++++++++++++++++- drivers/gpu/drm/xe/xe_exec_queue.h | 1 + drivers/gpu/drm/xe/xe_exec_queue_types.h | 6 ++++++ drivers/gpu/drm/xe/xe_sriov_vf_ccs.c | 2 +- drivers/gpu/drm/xe/xe_vm.c | 7 ++++++- 5 files changed, 45 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_exec_queue.c b/drivers/gpu/drm/xe/xe_exec_queue.c index 8724f8de67e2..779d7e7e2d2e 100644 --- a/drivers/gpu/drm/xe/xe_exec_queue.c +++ b/drivers/gpu/drm/xe/xe_exec_queue.c @@ -328,6 +328,7 @@ struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe * @xe: Xe device. * @tile: tile which bind exec queue belongs to. * @flags: exec queue creation flags + * @user_vm: The user VM which this exec queue belongs to * @extensions: exec queue creation extensions * * Normalize bind exec queue creation. Bind exec queue is tied to migration VM @@ -341,6 +342,7 @@ struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe */ struct xe_exec_queue *xe_exec_queue_create_bind(struct xe_device *xe, struct xe_tile *tile, + struct xe_vm *user_vm, u32 flags, u64 extensions) { struct xe_gt *gt = tile->primary_gt; @@ -377,6 +379,9 @@ struct xe_exec_queue *xe_exec_queue_create_bind(struct xe_device *xe, xe_exec_queue_put(q); return ERR_PTR(err); } + + if (user_vm) + q->user_vm = xe_vm_get(user_vm); } return q; @@ -407,6 +412,11 @@ void xe_exec_queue_destroy(struct kref *ref) xe_exec_queue_put(eq); } + if (q->user_vm) { + xe_vm_put(q->user_vm); + q->user_vm = NULL; + } + q->ops->destroy(q); } @@ -742,6 +752,22 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data, XE_IOCTL_DBG(xe, eci[0].engine_instance != 0)) return -EINVAL; + vm = xe_vm_lookup(xef, args->vm_id); + if (XE_IOCTL_DBG(xe, !vm)) + return -ENOENT; + + err = down_read_interruptible(&vm->lock); + if (err) { + xe_vm_put(vm); + return err; + } + + if (XE_IOCTL_DBG(xe, xe_vm_is_closed_or_banned(vm))) { + up_read(&vm->lock); + xe_vm_put(vm); + return -ENOENT; + } + for_each_tile(tile, xe, id) { struct xe_exec_queue *new; @@ -749,9 +775,11 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data, if (id) flags |= EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD; - new = xe_exec_queue_create_bind(xe, tile, flags, + new = xe_exec_queue_create_bind(xe, tile, vm, flags, args->extensions); if (IS_ERR(new)) { + up_read(&vm->lock); + xe_vm_put(vm); err = PTR_ERR(new); if (q) goto put_exec_queue; @@ -763,6 +791,8 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data, list_add_tail(&new->multi_gt_list, &q->multi_gt_link); } + up_read(&vm->lock); + xe_vm_put(vm); } else { logical_mask = calc_validate_logical_mask(xe, eci, args->width, diff --git a/drivers/gpu/drm/xe/xe_exec_queue.h b/drivers/gpu/drm/xe/xe_exec_queue.h index fda4d4f9bda8..37a9da22f420 100644 --- a/drivers/gpu/drm/xe/xe_exec_queue.h +++ b/drivers/gpu/drm/xe/xe_exec_queue.h @@ -28,6 +28,7 @@ struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe u32 flags, u64 extensions); struct xe_exec_queue *xe_exec_queue_create_bind(struct xe_device *xe, struct xe_tile *tile, + struct xe_vm *user_vm, u32 flags, u64 extensions); void xe_exec_queue_fini(struct xe_exec_queue *q); diff --git a/drivers/gpu/drm/xe/xe_exec_queue_types.h b/drivers/gpu/drm/xe/xe_exec_queue_types.h index 771ffe35cd0c..3a4263c92b3d 100644 --- a/drivers/gpu/drm/xe/xe_exec_queue_types.h +++ b/drivers/gpu/drm/xe/xe_exec_queue_types.h @@ -54,6 +54,12 @@ struct xe_exec_queue { struct kref refcount; /** @vm: VM (address space) for this exec queue */ struct xe_vm *vm; + /** + * @user_vm: User VM (address space) for this exec queue (bind queues + * only) + */ + struct xe_vm *user_vm; + /** @class: class of this exec queue */ enum xe_engine_class class; /** diff --git a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c index 797a4b866226..d963231b5135 100644 --- a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c +++ b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c @@ -346,7 +346,7 @@ int xe_sriov_vf_ccs_init(struct xe_device *xe) flags = EXEC_QUEUE_FLAG_KERNEL | EXEC_QUEUE_FLAG_PERMANENT | EXEC_QUEUE_FLAG_MIGRATE; - q = xe_exec_queue_create_bind(xe, tile, flags, 0); + q = xe_exec_queue_create_bind(xe, tile, NULL, flags, 0); if (IS_ERR(q)) { err = PTR_ERR(q); goto err_ret; diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 79ab6c512d3e..095bb197e8b0 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -1617,7 +1617,7 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) if (!vm->pt_root[id]) continue; - q = xe_exec_queue_create_bind(xe, tile, create_flags, 0); + q = xe_exec_queue_create_bind(xe, tile, vm, create_flags, 0); if (IS_ERR(q)) { err = PTR_ERR(q); goto err_close; @@ -3578,6 +3578,11 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } } + if (XE_IOCTL_DBG(xe, q && vm != q->user_vm)) { + err = -EINVAL; + goto put_exec_queue; + } + /* Ensure all UNMAPs visible */ xe_svm_flush(vm); -- cgit v1.2.3 From 772157f626d0e1a7c6d49dffb0bbe4b2343a1d44 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Tue, 20 Jan 2026 11:06:11 +0000 Subject: drm/xe/migrate: fix job lock assert MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We are meant to be checking the user vm for the bind queue, but actually we are checking the migrate vm. For various reasons this is not currently firing but this will likely change in the future. Now that we have the user_vm attached to the bind queue, we can fix this by directly checking that here. Fixes: dba89840a920 ("drm/xe: Add GT TLB invalidation jobs") Signed-off-by: Matthew Auld Cc: Thomas Hellström Cc: Matthew Brost Reviewed-by: Matthew Brost Reviewed-by: Arvind Yadav Link: https://patch.msgid.link/20260120110609.77958-4-matthew.auld@intel.com (cherry picked from commit 9dd1048bca4fe2aa67c7a286bafb3947537adedb) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_migrate.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c index 5a95b08a4723..d8ee76aab4e4 100644 --- a/drivers/gpu/drm/xe/xe_migrate.c +++ b/drivers/gpu/drm/xe/xe_migrate.c @@ -2445,7 +2445,7 @@ void xe_migrate_job_lock(struct xe_migrate *m, struct xe_exec_queue *q) if (is_migrate) mutex_lock(&m->job_mutex); else - xe_vm_assert_held(q->vm); /* User queues VM's should be locked */ + xe_vm_assert_held(q->user_vm); /* User queues VM's should be locked */ } /** @@ -2463,7 +2463,7 @@ void xe_migrate_job_unlock(struct xe_migrate *m, struct xe_exec_queue *q) if (is_migrate) mutex_unlock(&m->job_mutex); else - xe_vm_assert_held(q->vm); /* User queues VM's should be locked */ + xe_vm_assert_held(q->user_vm); /* User queues VM's should be locked */ } #if IS_ENABLED(CONFIG_PROVE_LOCKING) -- cgit v1.2.3 From f262015b9797effdec15e8a81c81b2158ede9578 Mon Sep 17 00:00:00 2001 From: Lukasz Laguna Date: Wed, 21 Jan 2026 15:33:04 +0100 Subject: drm/xe: Update wedged.mode only after successful reset policy change MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Previously, the driver's internal wedged.mode state was updated without verifying whether the corresponding engine reset policy update in GuC succeeded. This could leave the driver reporting a wedged.mode state that doesn't match the actual reset behavior programmed in GuC. With this change, the reset policy is updated first, and the driver's wedged.mode state is modified only if the policy update succeeds on all available GTs. This patch also introduces two functional improvements: - The policy is sent to GuC only when a change is required. An update is needed only when entering or leaving XE_WEDGED_MODE_UPON_ANY_HANG, because only in that case the reset policy changes. For example, switching between XE_WEDGED_MODE_UPON_CRITICAL_ERROR and XE_WEDGED_MODE_NEVER doesn't affect the reset policy, so there is no need to send the same value to GuC. - An inconsistent_reset flag is added to track cases where reset policy update succeeds only on a subset of GTs. If such inconsistency is detected, future wedged mode configuration will force a retry of the reset policy update to restore a consistent state across all GTs. Fixes: 6b8ef44cc0a9 ("drm/xe: Introduce the wedged_mode debugfs") Signed-off-by: Lukasz Laguna Link: https://patch.msgid.link/20260107174741.29163-3-lukasz.laguna@intel.com Reviewed-by: Rodrigo Vivi Signed-off-by: Rodrigo Vivi (cherry picked from commit 0f13dead4e0385859f5c9c3625a19df116b389d3) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/xe_debugfs.c | 72 ++++++++++++++++++++++++++++-------- drivers/gpu/drm/xe/xe_device_types.h | 18 +++++++++ drivers/gpu/drm/xe/xe_guc_ads.c | 14 ++++--- drivers/gpu/drm/xe/xe_guc_ads.h | 5 ++- 4 files changed, 87 insertions(+), 22 deletions(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/xe_debugfs.c b/drivers/gpu/drm/xe/xe_debugfs.c index e91da9589c5f..63fd8bf13c70 100644 --- a/drivers/gpu/drm/xe/xe_debugfs.c +++ b/drivers/gpu/drm/xe/xe_debugfs.c @@ -256,14 +256,64 @@ static ssize_t wedged_mode_show(struct file *f, char __user *ubuf, return simple_read_from_buffer(ubuf, size, pos, buf, len); } +static int __wedged_mode_set_reset_policy(struct xe_gt *gt, enum xe_wedged_mode mode) +{ + bool enable_engine_reset; + int ret; + + enable_engine_reset = (mode != XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET); + ret = xe_guc_ads_scheduler_policy_toggle_reset(>->uc.guc.ads, + enable_engine_reset); + if (ret) + xe_gt_err(gt, "Failed to update GuC ADS scheduler policy (%pe)\n", ERR_PTR(ret)); + + return ret; +} + +static int wedged_mode_set_reset_policy(struct xe_device *xe, enum xe_wedged_mode mode) +{ + struct xe_gt *gt; + int ret; + u8 id; + + guard(xe_pm_runtime)(xe); + for_each_gt(gt, xe, id) { + ret = __wedged_mode_set_reset_policy(gt, mode); + if (ret) { + if (id > 0) { + xe->wedged.inconsistent_reset = true; + drm_err(&xe->drm, "Inconsistent reset policy state between GTs\n"); + } + return ret; + } + } + + xe->wedged.inconsistent_reset = false; + + return 0; +} + +static bool wedged_mode_needs_policy_update(struct xe_device *xe, enum xe_wedged_mode mode) +{ + if (xe->wedged.inconsistent_reset) + return true; + + if (xe->wedged.mode == mode) + return false; + + if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET || + mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET) + return true; + + return false; +} + static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf, size_t size, loff_t *pos) { struct xe_device *xe = file_inode(f)->i_private; - struct xe_gt *gt; u32 wedged_mode; ssize_t ret; - u8 id; ret = kstrtouint_from_user(ubuf, size, 0, &wedged_mode); if (ret) @@ -272,22 +322,14 @@ static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf, if (wedged_mode > 2) return -EINVAL; - if (xe->wedged.mode == wedged_mode) - return size; + if (wedged_mode_needs_policy_update(xe, wedged_mode)) { + ret = wedged_mode_set_reset_policy(xe, wedged_mode); + if (ret) + return ret; + } xe->wedged.mode = wedged_mode; - xe_pm_runtime_get(xe); - for_each_gt(gt, xe, id) { - ret = xe_guc_ads_scheduler_policy_toggle_reset(>->uc.guc.ads); - if (ret) { - xe_gt_err(gt, "Failed to update GuC ADS scheduler policy. GuC may still cause engine reset even with wedged_mode=2\n"); - xe_pm_runtime_put(xe); - return -EIO; - } - } - xe_pm_runtime_put(xe); - return size; } diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h index 0b2fa7c56d38..047e86e22133 100644 --- a/drivers/gpu/drm/xe/xe_device_types.h +++ b/drivers/gpu/drm/xe/xe_device_types.h @@ -44,6 +44,22 @@ struct xe_pat_ops; struct xe_pxp; struct xe_vram_region; +/** + * enum xe_wedged_mode - possible wedged modes + * @XE_WEDGED_MODE_NEVER: Device will never be declared wedged. + * @XE_WEDGED_MODE_UPON_CRITICAL_ERROR: Device will be declared wedged only + * when critical error occurs like GT reset failure or firmware failure. + * This is the default mode. + * @XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET: Device will be declared wedged on + * any hang. In this mode, engine resets are disabled to avoid automatic + * recovery attempts. This mode is primarily intended for debugging hangs. + */ +enum xe_wedged_mode { + XE_WEDGED_MODE_NEVER = 0, + XE_WEDGED_MODE_UPON_CRITICAL_ERROR = 1, + XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET = 2, +}; + #define XE_BO_INVALID_OFFSET LONG_MAX #define GRAPHICS_VER(xe) ((xe)->info.graphics_verx100 / 100) @@ -587,6 +603,8 @@ struct xe_device { int mode; /** @wedged.method: Recovery method to be sent in the drm device wedged uevent */ unsigned long method; + /** @wedged.inconsistent_reset: Inconsistent reset policy state between GTs */ + bool inconsistent_reset; } wedged; /** @bo_device: Struct to control async free of BOs */ diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c index bcb85a1bf26d..3f7f1b5602d5 100644 --- a/drivers/gpu/drm/xe/xe_guc_ads.c +++ b/drivers/gpu/drm/xe/xe_guc_ads.c @@ -983,16 +983,17 @@ static int guc_ads_action_update_policies(struct xe_guc_ads *ads, u32 policy_off /** * xe_guc_ads_scheduler_policy_toggle_reset - Toggle reset policy * @ads: Additional data structures object + * @enable_engine_reset: true to enable engine resets, false otherwise * - * This function update the GuC's engine reset policy based on wedged.mode. + * This function update the GuC's engine reset policy. * * Return: 0 on success, and negative error code otherwise. */ -int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads) +int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads, + bool enable_engine_reset) { struct guc_policies *policies; struct xe_guc *guc = ads_to_guc(ads); - struct xe_device *xe = ads_to_xe(ads); CLASS(xe_guc_buf, buf)(&guc->buf, sizeof(*policies)); if (!xe_guc_buf_is_valid(buf)) @@ -1004,10 +1005,11 @@ int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads) policies->dpc_promote_time = ads_blob_read(ads, policies.dpc_promote_time); policies->max_num_work_items = ads_blob_read(ads, policies.max_num_work_items); policies->is_valid = 1; - if (xe->wedged.mode == 2) - policies->global_flags |= GLOBAL_POLICY_DISABLE_ENGINE_RESET; - else + + if (enable_engine_reset) policies->global_flags &= ~GLOBAL_POLICY_DISABLE_ENGINE_RESET; + else + policies->global_flags |= GLOBAL_POLICY_DISABLE_ENGINE_RESET; return guc_ads_action_update_policies(ads, xe_guc_buf_flush(buf)); } diff --git a/drivers/gpu/drm/xe/xe_guc_ads.h b/drivers/gpu/drm/xe/xe_guc_ads.h index 2e6674c760ff..7a39f361cb17 100644 --- a/drivers/gpu/drm/xe/xe_guc_ads.h +++ b/drivers/gpu/drm/xe/xe_guc_ads.h @@ -6,6 +6,8 @@ #ifndef _XE_GUC_ADS_H_ #define _XE_GUC_ADS_H_ +#include + struct xe_guc_ads; int xe_guc_ads_init(struct xe_guc_ads *ads); @@ -13,6 +15,7 @@ int xe_guc_ads_init_post_hwconfig(struct xe_guc_ads *ads); void xe_guc_ads_populate(struct xe_guc_ads *ads); void xe_guc_ads_populate_minimal(struct xe_guc_ads *ads); void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads); -int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads); +int xe_guc_ads_scheduler_policy_toggle_reset(struct xe_guc_ads *ads, + bool enable_engine_reset); #endif -- cgit v1.2.3 From bdcdf968be314b6fc8835b99fb4519e7619671e6 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= Date: Wed, 21 Jan 2026 10:10:47 +0100 Subject: drm, drm/xe: Fix xe userptr in the absence of CONFIG_DEVICE_PRIVATE MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CONFIG_DEVICE_PRIVATE is not selected by default by some distros, for example Fedora, and that leads to a regression in the xe driver since userptr support gets compiled out. It turns out that DRM_GPUSVM, which is needed for xe userptr support compiles also without CONFIG_DEVICE_PRIVATE, but doesn't compile without CONFIG_ZONE_DEVICE. Exclude the drm_pagemap files from compilation with !CONFIG_ZONE_DEVICE, and remove the CONFIG_DEVICE_PRIVATE dependency from CONFIG_DRM_GPUSVM and the xe driver's selection of it, re-enabling xe userptr for those configs. v2: - Don't compile the drm_pagemap files unless CONFIG_ZONE_DEVICE is set. - Adjust the drm_pagemap.h header accordingly. Fixes: 9e9787414882 ("drm/xe/userptr: replace xe_hmm with gpusvm") Cc: Matthew Auld Cc: Himal Prasad Ghimiray Cc: Thomas Hellström Cc: Matthew Brost Cc: "Thomas Hellström" Cc: Rodrigo Vivi Cc: dri-devel@lists.freedesktop.org Cc: # v6.18+ Signed-off-by: Thomas Hellström Reviewed-by: Matthew Auld Acked-by: Maarten Lankhorst Link: https://patch.msgid.link/20260121091048.41371-2-thomas.hellstrom@linux.intel.com (cherry picked from commit 1e372b246199ca7a35f930177fea91b557dac16e) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig index 4b288eb3f5b0..c34be1be155b 100644 --- a/drivers/gpu/drm/xe/Kconfig +++ b/drivers/gpu/drm/xe/Kconfig @@ -39,7 +39,7 @@ config DRM_XE select DRM_TTM select DRM_TTM_HELPER select DRM_EXEC - select DRM_GPUSVM if !UML && DEVICE_PRIVATE + select DRM_GPUSVM if !UML select DRM_GPUVM select DRM_SCHED select MMU_NOTIFIER -- cgit v1.2.3 From e27ada4f19e7ffda4c05ce8633daf6daed667eea Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= Date: Wed, 21 Jan 2026 10:10:48 +0100 Subject: drm/xe: Select CONFIG_DEVICE_PRIVATE when DRM_XE_GPUSVM is selected MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CONFIG_DEVICE_PRIVATE is a prerequisite for DRM_XE_GPUSVM. Explicitly select it so that DRM_XE_GPUSVM is not unintentionally left out from distro configs not explicitly enabling CONFIG_DEVICE_PRIVATE. v2: - Select also CONFIG_ZONE_DEVICE since it's needed by CONFIG_DEVICE_PRIVATE. v3: - Depend on CONFIG_ZONE_DEVICE rather than selecting it. Cc: Matthew Auld Cc: Matthew Brost Cc: Rodrigo Vivi Cc: Signed-off-by: Thomas Hellström Reviewed-by: Matthew Auld Link: https://patch.msgid.link/20260121091048.41371-3-thomas.hellstrom@linux.intel.com (cherry picked from commit 9386f49316074d2d76fd78d6bd359996de42fb7f) Signed-off-by: Thomas Hellström --- drivers/gpu/drm/xe/Kconfig | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/xe') diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig index c34be1be155b..4d7dcaff2b91 100644 --- a/drivers/gpu/drm/xe/Kconfig +++ b/drivers/gpu/drm/xe/Kconfig @@ -80,8 +80,9 @@ config DRM_XE_GPUSVM bool "Enable CPU to GPU address mirroring" depends on DRM_XE depends on !UML - depends on DEVICE_PRIVATE + depends on ZONE_DEVICE default y + select DEVICE_PRIVATE select DRM_GPUSVM help Enable this option if you want support for CPU to GPU address -- cgit v1.2.3