{ "dataType": "CVE_RECORD", "dataVersion": "5.2", "cveMetadata": { "cveId": "CVE-2024-57921", "assignerOrgId": "416baaa9-dc9f-4396-8d5f-8c081fb06d67", "state": "PUBLISHED", "assignerShortName": "Linux", "dateReserved": "2025-01-19T11:50:08.375Z", "datePublished": "2025-01-19T11:52:40.516Z", "dateUpdated": "2026-08-05T11:46:55.489Z" }, "containers": { "cna": { "providerMetadata": { "orgId": "416baaa9-dc9f-4396-8d5f-8c081fb06d67", "shortName": "Linux", "dateUpdated": "2026-08-05T11:46:55.489Z" }, "descriptions": [ { "lang": "en", "value": "In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu: Add a lock when accessing the buddy trim function\n\nWhen running YouTube videos and Steam games simultaneously,\nthe tester found a system hang / race condition issue with\nthe multi-display configuration setting. Adding a lock to\nthe buddy allocator's trim function would be the solution.\n\n\n[ 7197.250436] general protection fault, probably for non-canonical address 0xdead000000000108\n[ 7197.250447] RIP: 0010:__alloc_range+0x8b/0x340 [amddrm_buddy]\n[ 7197.250470] Call Trace:\n[ 7197.250472] \n[ 7197.250475] ? show_regs+0x6d/0x80\n[ 7197.250481] ? die_addr+0x37/0xa0\n[ 7197.250483] ? exc_general_protection+0x1db/0x480\n[ 7197.250488] ? drm_suballoc_new+0x13c/0x93d [drm_suballoc_helper]\n[ 7197.250493] ? asm_exc_general_protection+0x27/0x30\n[ 7197.250498] ? __alloc_range+0x8b/0x340 [amddrm_buddy]\n[ 7197.250501] ? __alloc_range+0x109/0x340 [amddrm_buddy]\n[ 7197.250506] amddrm_buddy_block_trim+0x1b5/0x260 [amddrm_buddy]\n[ 7197.250511] amdgpu_vram_mgr_new+0x4f5/0x590 [amdgpu]\n[ 7197.250682] amdttm_resource_alloc+0x46/0xb0 [amdttm]\n[ 7197.250689] ttm_bo_alloc_resource+0xe4/0x370 [amdttm]\n[ 7197.250696] amdttm_bo_validate+0x9d/0x180 [amdttm]\n[ 7197.250701] amdgpu_bo_pin+0x15a/0x2f0 [amdgpu]\n[ 7197.250831] amdgpu_dm_plane_helper_prepare_fb+0xb2/0x360 [amdgpu]\n[ 7197.251025] ? try_wait_for_completion+0x59/0x70\n[ 7197.251030] drm_atomic_helper_prepare_planes.part.0+0x2f/0x1e0\n[ 7197.251035] drm_atomic_helper_prepare_planes+0x5d/0x70\n[ 7197.251037] drm_atomic_helper_commit+0x84/0x160\n[ 7197.251040] drm_atomic_nonblocking_commit+0x59/0x70\n[ 7197.251043] drm_mode_atomic_ioctl+0x720/0x850\n[ 7197.251047] ? __pfx_drm_mode_atomic_ioctl+0x10/0x10\n[ 7197.251049] drm_ioctl_kernel+0xb9/0x120\n[ 7197.251053] ? srso_alias_return_thunk+0x5/0xfbef5\n[ 7197.251056] drm_ioctl+0x2d4/0x550\n[ 7197.251058] ? __pfx_drm_mode_atomic_ioctl+0x10/0x10\n[ 7197.251063] amdgpu_drm_ioctl+0x4e/0x90 [amdgpu]\n[ 7197.251186] __x64_sys_ioctl+0xa0/0xf0\n[ 7197.251190] x64_sys_call+0x143b/0x25c0\n[ 7197.251193] do_syscall_64+0x7f/0x180\n[ 7197.251197] ? srso_alias_return_thunk+0x5/0xfbef5\n[ 7197.251199] ? amdgpu_display_user_framebuffer_create+0x215/0x320 [amdgpu]\n[ 7197.251329] ? drm_internal_framebuffer_create+0xb7/0x1a0\n[ 7197.251332] ? srso_alias_return_thunk+0x5/0xfbef5\n\n(cherry picked from commit 3318ba94e56b9183d0304577c74b33b6b01ce516)" } ], "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 path is entered through amdgpu DRM ioctls (DRM_IOCTL_AMDGPU_GEM_CREATE / AMDGPU_CS / MODE_ATOMIC) on a local device node such as /dev/dri/renderD128. No network or adjacent-network exposure exists.\nAC:L - The attacker controls both sides of the race — one thread repeatedly allocates GFX12-DCC contiguous VRAM buffers driving the unlocked drm_buddy_block_trim(), while other threads hammer ordinary VRAM alloc/free through the locked buddy path — so the window can be hit reliably by simply looping. The original report shows it triggering under two ordinary concurrent workloads with no attacker effort at all.\nPR:L - DRM_IOCTL_AMDGPU_GEM_CREATE is DRM_AUTH|DRM_RENDER_ALLOW and accepts AMDGPU_GEM_CREATE_GFX12_DCC directly, and amdgpu_cs_find_mapping() forces AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS from the equally unprivileged CS ioctl — so an ordinary local user with render-node access (the logged-in desktop user, or any process in the render/video group) can reach it without DRM master or any capability.\nUI:N - A single attacker process spawns both the DCC-allocating thread and the concurrent VRAM alloc/free threads; no action by another user is required.\nS:U - The corruption is confined to the kernel's own drm_buddy allocator state and slab objects within the same kernel security authority; no hypervisor, IOMMU, or DMA isolation boundary is crossed.\nC:H - Free-list corruption yields a use-after-free on kmem_cache-backed drm_buddy_block objects whose contents the attacker can groom via their own VRAM allocations, and double-allocation of the same VRAM range to two different DRM clients allows reading another process's GPU buffer contents (framebuffers, decoded video, compute data).\nI:H - The racing list_del()/list_insert_sorted()/split_block() sequences give a controlled linked-list write primitive over freed and live buddy blocks, and aliased VRAM lets the attacker overwrite another client's buffers — memory corruption of this class is leveragable into arbitrary kernel write and control-flow hijack.\nA:H - The bug was reported as a system hang and reproduced as a general protection fault on LIST_POISON (0xdead000000000108) in __alloc_range, i.e. a hard kernel oops that an unprivileged local user can trigger repeatedly." } ] } ], "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_vram_mgr.c" ], "versions": [ { "version": "4a5ad08f537703c35cf7cc29845381805c891d9b", "lessThan": "758e3c3054b65336cf0c5f240221f63b4fb98478", "status": "affected", "versionType": "git" }, { "version": "4a5ad08f537703c35cf7cc29845381805c891d9b", "lessThan": "75c8b703e5bded1e33b08fb09b829e7c2c1ed50a", "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_vram_mgr.c" ], "versions": [ { "version": "6.11", "status": "affected" }, { "version": "0", "lessThan": "6.11", "status": "unaffected", "versionType": "semver" }, { "version": "6.12.10", "lessThanOrEqual": "6.12.*", "status": "unaffected", "versionType": "semver" }, { "version": "6.13", "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": "6.11", "versionEndExcluding": "6.12.10" }, { "vulnerable": true, "criteria": "cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*", "versionStartIncluding": "6.11", "versionEndExcluding": "6.13" } ] } ] } ], "references": [ { "url": "https://git.kernel.org/stable/c/758e3c3054b65336cf0c5f240221f63b4fb98478" }, { "url": "https://git.kernel.org/stable/c/75c8b703e5bded1e33b08fb09b829e7c2c1ed50a" } ], "title": "drm/amdgpu: Add a lock when accessing the buddy trim function", "x_generator": { "engine": "bippy-1.2.0" } } } }