{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99630","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-25T10:25:14.425Z","datePublished":"2026-11-20T05:49:48.468Z","dateUpdated":"2026-11-22T11:09:21.841Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-11-22T11:09:21.841Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu/userq: validate the moved individual BOs on restore\n\nAn individual BO of a user queue VM, i.e. one which doesn't share the\nroot reservation, goes to the individual needs_update list when it is\nevicted. The user queue restore looked at vm->invalidated for them, which\nthe VM state machine rework turned into the always_valid evicted list and\ncommit 18decfc9e8cc (\"drm/amdgpu/userq: actually validate userq BOs\")\ninto the individual evicted one. That list only holds imported DMA-bufs of\ncompute VMs though, so for a user queue VM nothing is validated.\n\nThe restore then neither locks such a BO nor moves it back. Its PTEs are\nupdated to wherever it is now, cleared if it went to system memory, and\nit doesn't get the new eviction fence, so it can move again under the\nrunning queue.\n\nOn navi48, once the BOs of an SDMA user queue are evicted from GTT\n(debugfs amdgpu_evict_gtt) the queue faults on its ring and rptr:\n\n  amdgpu 0000:03:00.0:   in page starting at address 0x0000000100000000 from client 10\n  amdgpu 0000:03:00.0: \t Faulty UTCL2 client ID: SDMA0 (0xd)\n  amdgpu 0000:03:00.0:   in page starting at address 0x0000000100101000 from client 10\n\nValidate the BOs on the individual needs_update list like before.\n\n(cherry picked from commit eb0d94765d4be18d41e93ad09947dd9cc06ed31a)"}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - The buggy amdgpu_userq_bo_validate runs from amdgpu_userq_restore_worker, which resume_work schedules after TTM evicts individual BOs of a process that created the queue through the local DRM_IOCTL_AMDGPU_USERQ (amdgpu_userq_ioctl). The moved pages are that process's GEM buffers.\nAC:H - Reaching amdgpu_userq_bo_validate requires usermode queues, but gfx_v12_0_early_init and sdma_v7_0_early_init set disable_uq for the default amdgpu.user_queue=-1 and only install userq_funcs when an admin sets the parameter to 1 or 2. The attacker cannot create that state.\nPR:L - DRM_IOCTL_AMDGPU_USERQ is DRM_AUTH|DRM_RENDER_ALLOW, and drm_ioctl_permit allows a render-node client without CAP_SYS_ADMIN, so an unprivileged local user who can open the render node can create the queue and allocate the BOs that TTM evicts.\nUI:N - The attacker opens the render node, creates the user queue in amdgpu_userq_create, and forces the TTM eviction that signals the eviction fence and schedules resume_work. No other principal has to act.\nS:U - The corrupted mapping is a GPUVM PTE in the caller's fpriv->vm, rewritten by amdgpu_vm_bo_update and consumed by that same process's user queue. Impact stays in the host kernel that owns the GPU.\nC:H - amdgpu_evf_mgr_rearm fences only drm_exec-locked objects, so the missed needs_update BO gets no new eviction fence. A later move leaves the VALID PTE from amdgpu_vm_bo_update pointing at a recycled DMA address the running queue can still fetch or SDMA-copy from.\nI:H - That PTE stays writable: amdgpu_vm_bo_invalidate only sets moved and links the bo_va onto individual.needs_update, so an SDMA or GFX store, or rptr writeback, through the still-mapped user queue writes the old page after TTM has recycled it.\nA:H - A BO left in TTM_PL_SYSTEM loses AMDGPU_PTE_VALID in amdgpu_ttm_tt_pde_flags, and the restored queue faults in gmc_v12_0_process_interrupt (the navi48 SDMA0 ring/rptr fault). The unfenced BO can also be moved under the mapped queue, which is a use-after-move of the backing page."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c"],"versions":[{"version":"4cdbba5a16aaf16513b69cc332b1d6f971a44b52","lessThan":"2664f497d5e35384736e42fc59e5b9ef27ca791a","status":"affected","versionType":"git"},{"version":"4cdbba5a16aaf16513b69cc332b1d6f971a44b52","lessThan":"ce5f2a25bdf6b2bccfc4ffb5e4c5f12e97eeace1","status":"affected","versionType":"git"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c"],"versions":[{"version":"7.2","status":"affected"},{"version":"0","lessThan":"7.2","status":"unaffected","versionType":"semver"},{"version":"7.3.5","lessThanOrEqual":"7.3.*","status":"unaffected","versionType":"semver"},{"version":"7.4-rc2","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":"7.2","versionEndExcluding":"7.3.5"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"7.2","versionEndExcluding":"7.4-rc2"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/2664f497d5e35384736e42fc59e5b9ef27ca791a"},{"url":"https://git.kernel.org/stable/c/ce5f2a25bdf6b2bccfc4ffb5e4c5f12e97eeace1"}],"title":"drm/amdgpu/userq: validate the moved individual BOs on restore","x_generator":{"engine":"bippy-1.2.0"}}}}