{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99627","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-25T10:25:14.425Z","datePublished":"2026-11-20T05:49:46.013Z","dateUpdated":"2026-11-22T11:09:20.645Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-11-22T11:09:20.645Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu/gfx12: disallow GFXOFF while updating the SPM VMID\n\ngfx_v12_0_update_spm_vmid() does a read-modify-write of RLC_SPM_MC_CNTL\nover MMIO without keeping GFX powered, unlike gfx11. When GFX is in\nGFXOFF at that point the update doesn't stick, and with the SPM VMID\nprogrammed on submission again, the job of a process with a reserved\nVMID can still run with the SPM writing through VMID 0.\n\nOn navi48, after a few other submissions and a VM fault on the same VM,\na COPY_DATA of RLC_SPM_MC_CNTL from such a job reads 0 although the\nupdate was called for it:\n\n  $ amd_uapi_regression --run-subtest cs-secure-ib-compute,gem-mapping-info-high-va,gpuvm-fault-high-va,reserved-vmid-spm-vmid\n  amdgpu 0000:03:00.0: [gfxhub] page fault (src_id:0 ring:40 vmid:7 pasid:185)\n  amdgpu 0000:03:00.0: \t Faulty UTCL2 client ID: SDMA1 (0xe)\n  RLC_SPM_MC_CNTL in the job: 0x00000000\n\nWith GFXOFF disallowed around the update it reads the reserved VMID 15.\n\n(cherry picked from commit 907aebc5578ae09fa5c0e2f0784201295837e60b)"}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:H","baseScore":7.1,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - gfx_v12_0_update_spm_vmid runs from amdgpu_ib_schedule inside amdgpu_job_run, for an amdgpu_cs_ioctl whose VMID was reserved by amdgpu_vm_ioctl(AMDGPU_VM_OP_RESERVE_VMID). The RLC_SPM_MC_CNTL update is lost on that local submission while GFX is in GFXOFF.\nAC:L - PP_GFXOFF_MASK is set in the default amdgpu_pp_feature_mask, and amdgpu_gfx_do_off_ctrl enters GFXOFF 100ms after idle. The attacker waits for that, then submits; the MMIO read-modify-write in gfx_v12_0_update_spm_vmid happens before the doorbell and does not land while GFX is powered off.\nPR:L - DRM_IOCTL_DEF_DRV marks AMDGPU_VM and AMDGPU_CS DRM_AUTH|DRM_RENDER_ALLOW, and amdgpu_vm_ioctl's AMDGPU_VM_OP_RESERVE_VMID case calls amdgpu_vmid_alloc_reserved with no capability check. Any local render-node opener can reserve the VMID and submit.\nUI:N - The attacker opens the render node, reserves the VMID, and submits the IB that enables SPM. No other user has to mount media or take any action.\nS:U - The stores land in VMID 0, which gmc_v12_0 documents as the kernel's physical GPU addresses and which gfxhub_v12_0_init_system_aperture_regs maps with an AGP window. That is the same host kernel authority as the driver.\nC:N - The RLC SPM block stores its own counter samples to RLC_SPM_PERFMON_RING_BASE using the VMID in RLC_SPM_MC_CNTL. A stuck VMID of 0 redirects those stores and does not return the target's previous bytes to the caller.\nI:H - The IB enables SPM and sets RLC_SPM_PERFMON_RING_BASE_LO/HI, a 48-bit destination the kernel never programs. gfx_v12_0_update_spm_vmid fails to replace VMID 0, so the stores hit the kernel aperture (GCMC_VM_AGP_BASE is 0). Commit 460c484f2411 calls that VMID 0 overwrite the dangerous case.\nA:H - A VMID 0 store through the AGP or GART aperture corrupts kernel memory and can oops the host; an unmapped destination faults the gfxhub for the rest of the job. On navi48 the repro left RLC_SPM_MC_CNTL at 0 for the whole submission, so the bad writes continue until SPM is stopped."}]}],"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/gfx_v12_0.c"],"versions":[{"version":"52cb80c12e8a647aa9c903903d58916e6c1c4d38","lessThan":"117e6ac89e730d647aeaf53b77420efece331f4d","status":"affected","versionType":"git"},{"version":"52cb80c12e8a647aa9c903903d58916e6c1c4d38","lessThan":"cd38a2b45edf84324ee9572c09c279fd4ebec5fc","status":"affected","versionType":"git"},{"version":"52cb80c12e8a647aa9c903903d58916e6c1c4d38","lessThan":"374aba4bef461462a9640f8355ebff5d5021c459","status":"affected","versionType":"git"},{"version":"52cb80c12e8a647aa9c903903d58916e6c1c4d38","lessThan":"d1551515a3258b43cba60bd7f45bbdd1f4e9e60a","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/gfx_v12_0.c"],"versions":[{"version":"6.11","status":"affected"},{"version":"0","lessThan":"6.11","status":"unaffected","versionType":"semver"},{"version":"6.12.118","lessThanOrEqual":"6.12.*","status":"unaffected","versionType":"semver"},{"version":"6.18.64","lessThanOrEqual":"6.18.*","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":"6.11","versionEndExcluding":"6.12.118"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.11","versionEndExcluding":"6.18.64"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.11","versionEndExcluding":"7.3.5"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.11","versionEndExcluding":"7.4-rc2"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/117e6ac89e730d647aeaf53b77420efece331f4d"},{"url":"https://git.kernel.org/stable/c/cd38a2b45edf84324ee9572c09c279fd4ebec5fc"},{"url":"https://git.kernel.org/stable/c/374aba4bef461462a9640f8355ebff5d5021c459"},{"url":"https://git.kernel.org/stable/c/d1551515a3258b43cba60bd7f45bbdd1f4e9e60a"}],"title":"drm/amdgpu/gfx12: disallow GFXOFF while updating the SPM VMID","x_generator":{"engine":"bippy-1.2.0"}}}}