CVE-2026-99629: drm/amdgpu: clear RB_OVERFLOW bit when enabling interrupts for navi10_ih
In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: clear RBOVERFLOW bit when enabling interrupts for navi10ih
Port this change to navi10ih.c: commit afbf7955ff01 ("drm/amdgpu: clear RBOVERFLOW bit when enabling interrupts")
Setting IHRBWPTR to 0 does not clear RBOVERFLOW if RBENABLE is not set. Pulse WPTROVERFLOWCLEAR after RBENABLE (and with WPTROVERFLOWENABLE) so overflow is cleared when interrupts are enabled.
(cherry picked from commit 713fb4ab04f4d2af6120f8eb607c5673b2759b38)
Affected Software
Frequently Asked Questions
Which systems are affected by this issue?
The issue is in the Linux kernel amdgpu driver’s navi10 interrupt-handler path. Systems using that driver path are the relevant exposure scope described by the fix.
What condition causes the problem?
The RB_OVERFLOW status bit can remain set because writing IH_RB_WPTR to zero does not clear it while RB_ENABLE is unset. The fix clears the overflow condition only after enabling the ring buffer and with WPTR_OVERFLOW_ENABLE set.
How can I determine whether a kernel includes the fix?
Check whether the kernel source or applied stable patches include one of the referenced stable commits: 53f0c61efb5ff1f203659f3ebb8ccacc39ae28bd, 00ae61183bff39238018de3db81d835d87618b42, or d7a7d21bebae5e2bf6f0b89d3dd6828de46c8614. The corrected code pulses WPTR_OVERFLOW_CLEAR after RB_ENABLE is set.