linux.git/drivers/gpu/drm/xe/xe_guc_submit.c, branch master

drm/xe/guc: Hold device ref until queue teardown completes

2026-07-16T14:33:08Z

GuC exec queue destruction can run asynchronously. If the final device put happens from a destroy worker, drmm cleanup can end up draining the same workqueue and deadlock. Hold a drm_device reference for the queue lifetime and drop it after queue teardown completes. This keeps drmm cleanup from running while async destroy work is still pending. Move GuC destroy work to a module-lifetime Xe workqueue and flush it on PCI remove so hot-unbind/rebind still waits for pending destroy work. With queue-held device refs, guc_submit_sw_fini() cannot run with live GuC IDs. Replace the fini wait with an assertion and remove the unused fini_wq. v2: - Rebase v3: - Switch to queue-lifetime drm_dev_get()/drm_dev_put() model. (Matt) - Queue async teardown on system_dfl_wq instead of xe->destroy_wq. (Matt) - Drop separate deferred drm_dev_put worker. - Remove stale drain_workqueue(xe->destroy_wq) from guc_submit_sw_fini(). v4: - Replace the guc_submit_sw_fini() wait with an assertion and remove the now-unused fini_wq. (sashiko) v5: - Move destroy work to a module-lifetime Xe workqueue instead of system_dfl_wq. (Matt) - Flush the module-lifetime destroy workqueue during PCI remove to preserve the old device-remove wait semantics. v6: - Keep SVM pagemap destroy work on the per-device destroy_wq to avoid letting it outlive the xe_device/drm_device. (Sashiko) - Use WQ_MEM_RECLAIM for xe->destroy_wq because SVM pagemap destroy work can be queued from the reclaim path. v7: - Drop the per-device xe->destroy_wq and use the module-level destroy WQ for SVM pagemap destroy as well. (Matt) - Rename xe_exec_queue_destroy_wq_*() helpers to xe_destroy_wq_*() helpers because the WQ is no longer exec-queue specific. (Matt) v8: - Rebase. v9: - Keep SVM pagemap destroy work on the per-device WQ_MEM_RECLAIM destroy_wq because it can be queued from reclaim and embeds the dev_pagemap used by devres teardown. (Sashiko) - Keep the module-level destroy WQ GuC-only and drop WQ_MEM_RECLAIM from it. - Update the module-WQ kdoc to document the GuC/SVM split. v10: - Keep xe->destroy_wq per-cpu while adding WQ_MEM_RECLAIM to fix the workqueue allocation warning. v11: - Drop the SVM pagemap destroy comment as it was revision-specific. (Thomas) v12: - Rebase. Fixes: 2d2be279f1ca ("drm/xe: fix UAF around queue destruction") Cc: Thomas Hellström Cc: Rodrigo Vivi Cc: Himal Prasad Ghimiray Cc: Tejas Upadhyay Reviewed-by: Matthew Brost Signed-off-by: Arvind Yadav Link: https://patch.msgid.link/20260716062624.211396-1-arvind.yadav@intel.com Signed-off-by: Tejas Upadhyay (cherry picked from commit da1124abac689cc2b1d8995e5f0a816f8a122edb) Signed-off-by: Thomas Hellström

drm/xe/guc: Keep scheduler timeline name alive

2026-07-16T14:33:02Z

The scheduler keeps a pointer to the timeline name, but q->name is freed with the exec queue while scheduler fences can still reference it. Store the name in struct xe_guc_exec_queue so it shares the scheduler's RCU-deferred lifetime. Fixes: 6bd90e700b42 ("drm/xe: Make dma-fences compliant with the safe access rules") Cc: Thomas Hellström Cc: Rodrigo Vivi Cc: Himal Prasad Ghimiray Cc: Matthew Brost Signed-off-by: Arvind Yadav Reviewed-by: Tvrtko Ursulin Acked-by: Matthew Brost Link: https://patch.msgid.link/20260714064402.2457257-1-arvind.yadav@intel.com Signed-off-by: Tejas Upadhyay (cherry picked from commit 41075f0eb5dcbd3b065d15f15ef7bbe9315188e8) Signed-off-by: Thomas Hellström

drm/xe: Remove redundant exec_queue_suspended() check in submit_exec_queue()

2026-07-01T16:01:16Z

There already has a check for exec_queue_suspended(q) that returns early if suspended. Fixes: 65280af331aa ("drm/xe/multi_queue: skip submit when primary queue is suspended") Signed-off-by: Lu Yao Reviewed-by: Rodrigo Vivi Link: https://patch.msgid.link/20260617012516.19930-1-yaolu@kylinos.cn Signed-off-by: Rodrigo Vivi (cherry picked from commit 173202a5a3a9e6590194ce0f5880d1529a71ade7) Signed-off-by: Thomas Hellström

drm/xe: wedge from the timeout handler only after releasing the queue

2026-07-01T16:01:02Z

A kernel job that exhausts its recovery attempts called xe_device_declare_wedged() directly from guc_exec_queue_timedout_job(), while the handler still owned the timed-out job and the queue scheduler (sched = &q->guc->sched, stopped at the top of the handler). In the default wedged mode (XE_WEDGED_MODE_UPON_CRITICAL_ERROR), xe_device_declare_wedged() takes the destructive path in xe_guc_submit_wedge(): guc_submit_reset_prepare(), xe_guc_submit_stop() - which calls guc_exec_queue_stop() on every queue, including this one - softreset and pause-abort. That tears submission down, signals the in-flight fences and restarts the schedulers. This is the correct behaviour when the wedge originates outside the TDR, but not when the TDR itself triggers it: every queue should be torn down except the one the TDR is currently operating on, which it still owns. Control then returned to the handler, which kept using the now stale job and scheduler: xe_sched_job_set_error(job, err); drm_sched_for_each_pending_job(tmp_job, &sched->base, NULL) xe_sched_job_set_error(to_xe_sched_job(tmp_job), -ECANCELED); drm_sched_for_each_pending_job() warns because the scheduler is no longer stopped (WARN_ON(!drm_sched_is_stopped())) and the iteration then dereferences a freed job, faulting on the slab poison: Oops: general protection fault ... 0x6b6b6b6b6b6b6c3b RIP: guc_exec_queue_timedout_job+... Defer the wedge until the handler has finished operating on the queue, right before returning DRM_GPU_SCHED_STAT_NO_HANG, so the teardown no longer races with this handler's use of @q. Fixes: 770031ec2312 ("drm/xe: fix job timeout recovery for unstarted jobs and kernel queues") Suggested-by: Matthew Brost Cc: Matthew Brost Cc: Thomas Hellström Cc: Himal Prasad Ghimiray Cc: Sanjay Yadav Assisted-by: GitHub-Copilot:claude-opus-4.8 Reviewed-by: Matthew Brost Link: https://patch.msgid.link/20260612162414.287971-2-rodrigo.vivi@intel.com Signed-off-by: Rodrigo Vivi (cherry picked from commit a889e9b06bfdb375fc88b3b2a4b143f621f930c6) Signed-off-by: Thomas Hellström

Merge tag 'drm-next-2026-06-17' of https://gitlab.freedesktop.org/drm/kernel

2026-06-17T09:21:00Z

Pull drm updates from Dave Airlie: "Highlights: - xe: add initial CRI platform support - amdgpu: initial HDMI 2.1 FRL support - rust: add some new type concepts for device lifetimes - scheduler: moves to a fair algorithm and lots of cleanups But it's mostly the usual mountain of changes across the board. core: - add docbook for DRM_IOCTL_SYNCOBJ_EVENTFD - change signature of drm_connector_attach_hdr_output_metadata_property - dedup counter and timestamp retrieval in vblank code - parse AMD VSDB v3 in CTA extension blocks - add P230, Y7, XYYY2101010, T430, XVUY210101010 formats - don't call drop master on file close if not master - use drm_printf_indent in atomic / bridge - fix 32b format descriptions - docs: fix toctree - hdmi: add common TMDS character rates - fix drm_syncobj_find_fence leak rust: - introduce Higher-Ranked lifetime types - replace drvdata with scoped registration data - add GPUVM immediate mode abstraction for rust GPU drivers - introduce DeviceContext type state for drm::Device bridge: - clarify drm_bridge_get/put - create drm_get_bridge_by_endpoint and use it - analogix_dp: add panel probing - ite-it6211 - use drm audio hdmi helpers buddy: - add lockdep annotations dp: - add PR and VRR updates - mst: fix buffer overflows - add Adaptive Sync SDP decoding support - fix OOB reads in dp-mst ttm: - bump fpfn/lpfn to 64-bit scheduler: - change default to fair scheduler - map runqueue 1:1 with scheduler dma-buf: - port selftests to kunit - convert dma-buf system/heap allocators to module - add separate DMABUF_HEAPS_SYSTEM_CC_SHARED Kconfig udmabuf: - revert hugetlb support - fix error with CONFIG_DMA_API_DEBUG dma-fence: - fix tracepoints lifetime - remove unused signal on any support ras: - add clear error counter netlink command to drm ras gpusvm: - reject VMAs with VM_IO or VM_PFNMAP when creating SVM ranges - use IOVA allocations pagemap: - use IOVA allocations panels: - update to use ref counts - add support for CSW PNB601LS1-2, LGD LP116WHA-SPB1 - add support for waveshare panels - CMN N116BCN-EA1, CMN N140HCA-EEK, IVO M140NWFQ R5, - IVO, R140NWFW R0, BOE NT140*, BOE NV133FHM-N4F, - AUO B140*, AUO B133HAN06.6 and AUO B116XTN02.3 eDP panels - Surface Pro 12 Panel xe: - add CRI PCI-IDs - debugfs add multi-lrc info - engine init cleanup - PF fair scheduling auto provisioning - system controller support for CRI/Xe3p - PXP state machine fixes - Reset/wedge/unload corner case fixes - Wedge path memory allocation fixes - PAT type cleanups - Reject unsafe PAT for CPU cached memory - OA improvements for CRI device memory - kernel doc syntax in xe headers - xe_drm.h documentation fixes - include guard cleanups - VF CCS memory pool - i915/xe step unification - Xe3p GT tuning fixes - forcewake cleanup in GT and GuC - admin-only PF mode - enable hwmon energy attributes for CRI - enable GT_MI_USER_INTERRUPT - refactor emit functions - oa workarounds - multi_queue: allow QUEUE_TIMESTAMP register - convert stolen memory to ttm range manager - use xe2 style blitter as a feature flag - make drm_driver const - add/use IRQ page to HW engine definition - fix oops when display disabled i915: - enable PIPEDMC_ERROR interrupt - more common display code refactoring - restructure DP/HDMI sink format handling - eliminate FB usage from lowlevel pinning code - panel replay bw optimization - integrate sharpness filter into the scaler - new fb_pin abstraction for xe/i915 fb transparent handling - skip inactive MST connectors on HDCP - start switching to display specific registers - use polling when irq unavailable - Adaptive-sync SDP prep amdgpu: - use drm_display_info for AMD VSDB data - Initial HDMI 2.1 FRL support - Initial DCN 4.2.1 support - GART fixes for non-4k pages - GC 11.5.6/SDMA 6.4.0/and other new IPs - GFX9/DCE6/Hawaii/SDMA4/GART/Userq fixes - Finish support for using multiple SDMA queues for TTM operations - SWSMU updates - GC 12.1 updates - SMU 15.0.8 updates - DCN 4.2 updates - DC type conversion fixes - Enable DC power module - Replay/PSR updates - SMU 13.x updates - Compute queue quantum MQD updates - ASPM fix - Align VKMS with common implementation - DC analog support fixes - UVD 3 fixes - TCC harvesting fixes for SI - GC 11 APU module reload fix - NBIO 6.3.2 support - IH 7.1 updates - DC cursor fixes - VCN/JPEG user fence fixes - DC support for connectors without DDC - Prefer ROM BAR for default VGA device - DC bandwidth fixes - Add PTL support for profiler - Introduce dc_plane_cm and migrate surface update color path - Add FRL registers for HDMI 2.1 - Restructure VM state machine - Auxless ALPM support - GEM_OP locking/warning fixes - switch to system_dfl_wq amdkfd: - GPUVM TLB flush fix - Hotplug fix - Boundary check fixes - SVM fixes - CRIU fixes - add profiler API - MES 12.1 updates msm: - core: - fix shrinker documentation - IFPC enabled for gen8 - PERFCNTR_CONFIG ioctl support - GPU: - reworked UBWC handling - a810 support - MDSS: - add support for Milos platform - reworked UBWC handling - DisplayPort: - reworked HPD handling as prep for MST - DPU: - Milos platform support - reworked UBWC handling - DSI: - Milos platform support nova: - Hopper/Blackwell enablement (GH100/GB100/GB202) - FSP support - 32-bit firmware support - HAL functions - refactor GSP boot/unload - GA100 support - VBIOS hardening/refactoring - Adopt higher order lifetime types tyr: - define register blocks - add shmem backed GEM objects - adopt higher order lifetime types - move clock cleanup into Drop radeon: - Hawaii SMU fixes - CS parser fix - use struct drm_edid instead of edid amdxdna: - export per-client BO memory via fdinfo - AIE4 device support - support medium/lower power modes - expandable device heap support - revert read-only user-pointer BO mappings ivpu: - support frequency limiting panthor: - enable GEM shrinker support - add eviction and reclaim info to fdinfo v3d: - enable runtime PM mgag200: - support XRGB1555 + C8 ast: - support XRGB1555 + C8 - use constants for lots of registers - fix register handling imagination: - fence handling refactoring nouveau: - fix sched double call - expose VBIOS on GSP-RM systems - add GA100 support virtio: - add VIRTIO_GPU_F_BLOB_ALIGNMENT flag - add deferred mapping support gud: - add RCade Display Adapter hibmc: - fix no connectors usage mediatek: - hdmi: convert error handling - simplify mtk_crtc allocation exynos: - move fbdev emulation to drm client buffers - use drm format helpers for geometry/size - adopt core DMA tracking - fix framebuffer offset handling renesas: - add RZ/T2H SOC support versilicon: - add cursor plane support tegra: - use drm client for framebuffer" * tag 'drm-next-2026-06-17' of https://gitlab.freedesktop.org/drm/kernel: (1731 commits) dma-buf: move system_cc_shared heap under separate Kconfig accel/amdxdna: Clear sva pointer after unbind agp/amd64: Fix broken error propagation in agp_amd64_probe() accel/amdxdna: Require carveout when PASID and force_iova are disabled drm/amdkfd: always resume_all after suspend_all drm/amdgpu/gfx: move fault and EOP IRQ get/put to hw_init/hw_fini drm/amd/display: Consult MCCS FreeSync cap only if requested & supported drm/amd/pm: Use strscpy in profile mode parsing drm/amdkfd: Fix infinite loop parsing CRAT with zero subtype length drm/amdkfd: fix sysfs topology prop length on buffer truncation drm/amdgpu: drop retry loop in amdgpu_hmm_range_get_pages drm/amd/pm: bound OD parameter parsing to stack array size drm/amd/pm: Stop pp_od_clk_voltage emit at PAGE_SIZE drm/amdkfd: Unwind debug trap enable on copy_to_user failure drm/amdgpu: validate the mes firmware version for gfx12.1 drm/amdgpu: validate the mes firmware version for gfx12 drm/amdgpu: compare MES firmware version ucode for gfx11 drm/amdkfd: Add bounds check for AMDKFD_IOC_WAIT_EVENTS drm/amdgpu: restart the CS if some parts of the VM are still invalidated drm/amd/display: use unsigned types for local pipe and REG_GET counters ...

drm/xe: fix job timeout recovery for unstarted jobs and kernel queues

2026-06-11T13:39:43Z

A job that GuC never scheduled (never started) indicates a GuC scheduling failure; previously such jobs were silently errored out instead of triggering a GT reset to recover. Trigger a GT reset and resubmit them, but only when the queue was not already killed or banned: an unstarted job on an already banned queue is the ban working as intended and must neither clear the ban nor kick off a reset, otherwise a banned userspace queue could be resurrected and spam GT resets. Kernel queues are always recovered this way and wedge the device once recovery attempts are exhausted, since kernel work must not silently fail. A started job that times out on a userspace VM bind queue stays banned rather than being reset and retried. The queue is banned early in the timeout handler to signal the G2H scheduling-done handler so it wakes the disable-scheduling waiter; without it the waiter sleeps the full 5s timeout. When a reset is warranted the ban is cleared before rearming so that guc_exec_queue_start() can resubmit jobs after the GT reset - a still-banned queue would block resubmission and cause an infinite TDR loop. The already-banned case is gated out before this point via skip_timeout_check, so it is unaffected. v2: (Himal) Do it for any queue type, not just kernel/migration v3: - (Sashiko and Sanjay): don't clear the ban / GT reset for already killed/banned queues on unstarted-job timeout - Update commit message - (Matt) Add Fixes tag Fixes: fe05cee4d953 ("drm/xe: Don't short circuit TDR on jobs not started") Cc: Matthew Auld Cc: Matthew Brost Cc: Sanjay Yadav Cc: Himal Prasad Ghimiray Assisted-by: GitHub-Copilot:claude-sonnet-4.6 Assisted-by: GitHub-Copilot:claude-opus-4.8 Tested-by: Sanjay Yadav Reviewed-by: Sanjay Yadav Reviewed-by: Matthew Brost Reviewed-by: Himal Prasad Ghimiray Link: https://patch.msgid.link/20260610152548.404575-3-rodrigo.vivi@intel.com Signed-off-by: Rodrigo Vivi (cherry picked from commit b1107d085e7e8ed15ba6f80c102528a9c8a6cb0e) Signed-off-by: Matthew Brost

drm/xe/multi_queue: skip submit when primary queue is suspended

2026-06-04T13:04:09Z

Return early in submit path when the multi-queue primary exec queue is suspended to avoid submitting while suspended. v2: Remove idle_skip_suspend fix as that feature is being reverted here https://patchwork.freedesktop.org/series/167262/ Fixes: bc5775c59258 ("drm/xe/multi_queue: Add GuC interface for multi queue support") Cc: stable@vger.kernel.org # v7.0+ Assisted-by: GitHub-Copilot:claude-sonnet-4.6 Reviewed-by: Daniele Ceraolo Spurio Signed-off-by: Niranjana Vishwanathapura Link: https://patch.msgid.link/20260603233946.863663-2-niranjana.vishwanathapura@intel.com (cherry picked from commit b7fb55cc3364ca128cfff9d50649ffd4327cd01e) Signed-off-by: Rodrigo Vivi

drm/xe: Clear pending_disable before signaling suspend fence

2026-06-04T13:04:03Z

In the schedule-disable done path for suspend, we signal the suspend fence before clearing pending_disable. That wakeup can let suspend_wait complete and resume be queued immediately. The resume path may then reach enable_scheduling() while pending_disable is still set and hit the !exec_queue_pending_disable(q) assertion. Fix this by clearing pending_disable before signaling the suspend fence, so any resumed transition observes a consistent state. Fixes: 87651f31ae4e ("drm/xe/guc_submit: fix race around suspend_pending") Cc: stable@vger.kernel.org # v7.0+ Signed-off-by: Tangudu Tilak Tirumalesh Reviewed-by: Thomas Hellstrom Signed-off-by: Daniele Ceraolo Spurio Link: https://patch.msgid.link/20260603065217.3131066-3-tilak.tirumalesh.tangudu@intel.com (cherry picked from commit 4b1ae138b0e103d753773956a84eebc2edbf62c4) Signed-off-by: Rodrigo Vivi

Revert "drm/xe: Skip exec queue schedule toggle if queue is idle during suspend"

2026-06-04T13:03:57Z

This reverts commit 8533051ce92015e9cc6f75e0d52119b9d91610b6. The idle-skip optimization bypasses GuC suspend, so the GPU may not perform the context switch that flushes TLB entries for invalidated userptr VMAs. In LR/preempt-fence VM mode, this can lead to missed TLB invalidation and page faults during userptr invalidation tests. Restore unconditional schedule toggling on suspend so the context-switch TLB flush is always performed. This optimization will be reintroduced with a fix that does not skip suspend in LR/preempt-fence VM mode. Fixes: 8533051ce920 ("drm/xe: Skip exec queue schedule toggle if queue is idle during suspend") Cc: stable@vger.kernel.org # v7.0+ Suggested-by: Thomas Hellstrom Signed-off-by: Tangudu Tilak Tirumalesh Reviewed-by: Thomas Hellstrom Signed-off-by: Daniele Ceraolo Spurio Link: https://patch.msgid.link/20260603065217.3131066-2-tilak.tirumalesh.tangudu@intel.com (cherry picked from commit 6a1e7934d9a6cf46aecae00a99c2603d1295e170) Signed-off-by: Rodrigo Vivi

drm/xe: Assign queue name in time for drm_sched_init

2026-05-26T19:31:34Z

Currently the queue name is only assigned after the drm scheduler instance has been created. This loses information with all logging or debug workqueue facilities so lets re-order things a bit so the name gets assigned in time. To be able to assign a GuC ID early we split the allocation into reservation and publish phases. First, with the submission state lock held, we reserve the ID in the GuC ID manager, which serves as an authoritative source of truth. Then we can drop the lock and reserve entries in the exec queue lookup XArray. This can be lockless since the NULL entries are invisible both to the kernel and userspace. Only after the queue has been fully created we replace the reserved entries with the queue pointer, which can be done locklessly for single width queues. Signed-off-by: Tvrtko Ursulin Cc: Matthew Brost Cc: Rodrigo Vivi Cc: Thomas Hellstrom Reviewed-by: Rodrigo Vivi Link: https://patch.msgid.link/20260523103418.61832-1-tvrtko.ursulin@igalia.com Signed-off-by: Rodrigo Vivi