{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-68382","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-07-30T09:28:09.387Z","datePublished":"2026-08-10T12:04:01.187Z","dateUpdated":"2026-08-17T05:04:28.645Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-08-17T05:04:28.645Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/xe/guc: Hold device ref until queue teardown completes\n\nGuC exec queue destruction can run asynchronously. If the final device\nput happens from a destroy worker, drmm cleanup can end up draining\nthe same workqueue and deadlock.\n\nHold a drm_device reference for the queue lifetime and drop it after\nqueue teardown completes. This keeps drmm cleanup from running while\nasync destroy work is still pending.\n\nMove GuC destroy work to a module-lifetime Xe workqueue and flush it\non PCI remove so hot-unbind/rebind still waits for pending destroy work.\n\nWith queue-held device refs, guc_submit_sw_fini() cannot run with live\nGuC IDs. Replace the fini wait with an assertion and remove the unused\nfini_wq.\n\nv2:\n  - Rebase\n\nv3:\n  - Switch to queue-lifetime drm_dev_get()/drm_dev_put() model. (Matt)\n  - Queue async teardown on system_dfl_wq instead of xe->destroy_wq. (Matt)\n  - Drop separate deferred drm_dev_put worker.\n  - Remove stale drain_workqueue(xe->destroy_wq) from guc_submit_sw_fini().\n\nv4:\n  - Replace the guc_submit_sw_fini() wait with an assertion and remove\n    the now-unused fini_wq. (sashiko)\n\nv5:\n  - Move destroy work to a module-lifetime Xe workqueue instead of\n    system_dfl_wq. (Matt)\n  - Flush the module-lifetime destroy workqueue during PCI remove to\n    preserve the old device-remove wait semantics.\n\nv6:\n  - Keep SVM pagemap destroy work on the per-device destroy_wq to avoid\n    letting it outlive the xe_device/drm_device. (Sashiko)\n  - Use WQ_MEM_RECLAIM for xe->destroy_wq because SVM pagemap destroy work\n    can be queued from the reclaim path.\n\nv7:\n  - Drop the per-device xe->destroy_wq and use the module-level destroy WQ\n    for SVM pagemap destroy as well. (Matt)\n  - Rename xe_exec_queue_destroy_wq_*() helpers to xe_destroy_wq_*()\n    helpers because the WQ is no longer exec-queue specific. (Matt)\n\nv8:\n  - Rebase.\n\nv9:\n  - Keep SVM pagemap destroy work on the per-device WQ_MEM_RECLAIM\n    destroy_wq because it can be queued from reclaim and embeds\n    the dev_pagemap used by devres teardown. (Sashiko)\n  - Keep the module-level destroy WQ GuC-only and drop WQ_MEM_RECLAIM\n    from it.\n  - Update the module-WQ kdoc to document the GuC/SVM split.\n\nv10:\n  - Keep xe->destroy_wq per-cpu while adding WQ_MEM_RECLAIM to fix the\n    workqueue allocation warning.\n\nv11:\n  - Drop the SVM pagemap destroy comment as it was revision-specific.\n    (Thomas)\n\nv12:\n  - Rebase.\n\n(cherry picked from commit da1124abac689cc2b1d8995e5f0a816f8a122edb)"}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7.8,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - The vulnerable state is created through the Intel Xe DRM render node (/dev/dri/renderD*) via DRM_RENDER_ALLOW exec-queue create/destroy ioctls and manifests during local device teardown; no network or adjacent-network input is involved.\nAC:L - The attacker deterministically controls the decisive side of the race: GuC queue teardown blocks on a GuC deregister round-trip and cancel_delayed_work_sync() of the TDR, so long-running contexts with large preempt/job timeouts reliably keep exec_queue_lookup non-empty past guc_submit_sw_fini()'s fixed 5s wait, and the attempt can be repeated freely.\nPR:L - Exec queue creation and destruction are DRM_RENDER_ALLOW ioctls reachable by any unprivileged user holding an open render node, which is granted to logged-in users by default udev/ACL rules on systems with Intel GPUs; no capability or root privilege is needed to leave destroy work pending.\nUI:N - The attacker acts entirely on its own through render-node ioctls and GPU submission; no victim must open a file, mount a filesystem, or take any other action for the stale teardown state to be created.\nS:U - The corruption is confined to kernel memory managed by the same security authority as the xe driver; there is no VM escape, IOMMU/DMA boundary bypass, or other crossing into a different security scope.\nC:H - After the 5s wait times out (the assert is a no-op on production builds), drm_dev_release() kfree()s the whole xe_device while destroy work is pending, so the worker dereferences a freed slab object; the attacker can spray the reclaimed allocation and read back kernel data through the stale xe/guc pointers and xarray.\nI:H - The pending worker writes through the freed xe_device — xa_erase() on the destroyed lookup, mutex operations on submission_state.lock, and xe_guc_id_mgr_release_locked() bitmap updates — giving controlled writes into a reclaimed heap object and a path to control-flow hijack.\nA:H - Use-after-free on the freed device structure reliably oopses or panics the kernel; additionally the drain_workqueue() from a work item on the same destroy workqueue self-deadlocks, hanging device removal indefinitely."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/xe/xe_device.c","drivers/gpu/drm/xe/xe_device_types.h","drivers/gpu/drm/xe/xe_guc_submit.c","drivers/gpu/drm/xe/xe_guc_types.h","drivers/gpu/drm/xe/xe_module.c","drivers/gpu/drm/xe/xe_module.h","drivers/gpu/drm/xe/xe_pci.c"],"versions":[{"version":"2d2be279f1ca9e7288282d4214f16eea8a727cdb","lessThan":"03d6f83979b0d75a0b0893dfe1735ec93facf515","status":"affected","versionType":"git"},{"version":"2d2be279f1ca9e7288282d4214f16eea8a727cdb","lessThan":"9b7e60184f4b22e893d4ae95234d5f26261a430c","status":"affected","versionType":"git"},{"version":"272b0e78874586d6ccae04079d75b27b47705544","status":"affected","versionType":"git"},{"version":"421c74670b0f9d5c007f1276d3647aa58f407fde","status":"affected","versionType":"git"},{"version":"6.10.14","lessThan":"6.11","status":"affected","versionType":"semver"},{"version":"6.11.3","lessThan":"6.12","status":"affected","versionType":"semver"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/xe/xe_device.c","drivers/gpu/drm/xe/xe_device_types.h","drivers/gpu/drm/xe/xe_guc_submit.c","drivers/gpu/drm/xe/xe_guc_types.h","drivers/gpu/drm/xe/xe_module.c","drivers/gpu/drm/xe/xe_module.h","drivers/gpu/drm/xe/xe_pci.c"],"versions":[{"version":"6.12","status":"affected"},{"version":"0","lessThan":"6.12","status":"unaffected","versionType":"semver"},{"version":"7.1.6","lessThanOrEqual":"7.1.*","status":"unaffected","versionType":"semver"},{"version":"7.2","lessThanOrEqual":"*","status":"unaffected","versionType":"original_commit_for_fix"}]}],"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.12","versionEndExcluding":"7.1.6"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.12","versionEndExcluding":"7.2"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.10.14"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.11.3"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/03d6f83979b0d75a0b0893dfe1735ec93facf515"},{"url":"https://git.kernel.org/stable/c/9b7e60184f4b22e893d4ae95234d5f26261a430c"}],"title":"drm/xe/guc: Hold device ref until queue teardown completes","x_generator":{"engine":"bippy-1.2.0"}}}}