CVE-2026-99627: drm/amdgpu/gfx12: disallow GFXOFF while updating the SPM VMID
In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu/gfx12: disallow GFXOFF while updating the SPM VMID
gfxv120updatespmvmid() does a read-modify-write of RLCSPMMCCNTL over MMIO without keeping GFX powered, unlike gfx11. When GFX is in GFXOFF at that point the update doesn't stick, and with the SPM VMID programmed on submission again, the job of a process with a reserved VMID can still run with the SPM writing through VMID 0.
On navi48, after a few other submissions and a VM fault on the same VM, a COPYDATA of RLCSPMMCCNTL from such a job reads 0 although the update was called for it:
$ amduapiregression --run-subtest cs-secure-ib-compute,gem-mapping-info-high-va,gpuvm-fault-high-va,reserved-vmid-spm-vmid amdgpu 0000:03:00.0: [gfxhub] page fault (srcid:0 ring:40 vmid:7 pasid:185) amdgpu 0000:03:00.0: Faulty UTCL2 client ID: SDMA1 (0xe) RLCSPMMCCNTL in the job: 0x00000000
With GFXOFF disallowed around the update it reads the reserved VMID 15.
(cherry picked from commit 907aebc5578ae09fa5c0e2f0784201295837e60b)
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.12.118 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.18.64 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 7.3.5 - Upgrade
Upgrade to a fixed release to a version that resolves this vulnerability.
Patch 907aebc5578ae09fa5c0e2f0784201295837e60b
Frequently Asked Questions
Which systems are most clearly affected by the reported behavior?
The issue is reported in the Linux AMDGPU GFX12 path and was reproduced on Navi48 hardware. The failure involves jobs from a process using a reserved VMID while the SPM is programmed during submission.
What conditions are needed for the incorrect SPM VMID state to occur?
GFX must enter GFXOFF while gfx_v12_0_update_spm_vmid() performs its MMIO read-modify-write of RLC_SPM_MC_CNTL. In that state, the register update may not persist, allowing SPM writes to continue through VMID 0.
How can an administrator identify the affected behavior?
The supplied reproduction produced an AMDGPU gfxhub page fault and observed RLC_SPM_MC_CNTL as 0x00000000 in the affected job, despite an attempted VMID update. The corrected behavior reads the reserved VMID value, shown as 15 in the report.
What remediation is described?
Use a kernel containing the fix that disallows GFXOFF around the SPM VMID update. The provided stable-kernel references identify commits carrying that change.