Linux kernel unsafe BO eviction in amdgpu handle closure
Linux
Affected versions
| Product | Affected | Fixed in |
|---|---|---|
| Linux Linux |
| No fix yet |
| Linux Linux |
|
|
Linux
| Product | Affected | Fixed in |
|---|---|---|
| Linux Linux |
| No fix yet |
| Linux Linux |
|
|
6.18.597.2.137.3In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: keep the eviction fence until the last handle is closed
amdgpu_gem_object_close() detaches the user queue eviction fence from the BO before it checks bo_va->ref_count, so closing any handle of a BO removes the fence, even when the file still has another handle to the BO and the bo_va with its GPU mappings stays. The BO can then be evicted without suspending the user queues of the file, which keep accessing it through mappings that are only marked for update.
Seen on navi31 with a SDMA user queue: a BO opened twice in the same file (GEM_FLINK + GEM_OPEN) loses its eviction fence when the second handle is closed (traced attach/detach of the fence):
attach bo=ffff8c78a6c84c00 GEM_CLOSE handle=7 detach bo=ffff8c78a6c84c00
while handle 6 and the mapping are still in use.
Only detach the fence when the last reference to the bo_va goes away, matching amdgpu_gem_object_open(), which attaches it only when the bo_va is created. With this the fence stays until the file is closed.
(cherry picked from commit 8387982f03d4026b7dc11b0b5e3118298eee4d6a)
From the CVE record · Linux
In the Linux kernel's amdgpu driver, closing one of multiple handles to a buffer object can remove its eviction fence while another handle and GPU mapping remain in use. A local user using an AMD GPU user queue can trigger eviction without suspending the queue, which may continue accessing the buffer through mappings marked for update. This requires multiple handles to the same buffer object; no separate user interaction is described.
Operators running Linux kernels with the amdgpu driver should check whether their kernel includes the fix; affected versions were not specified.