CVE-2026-99622: drm/amdgpu: drop the dma-buf reference when a KFD import fails
In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: drop the dma-buf reference when a KFD import fails
importobjcreate() takes a reference on the dma-buf for the new kgdmem, but the error path after validating the BO doesn't drop it. Since the validation reserves the BO interruptibly, an import failing with -ERESTARTSYS leaks the exported BO with its memory, and keeps the module in use.
Importing a 64 MiB VRAM BO in a loop while getting SIGUSR1 every 20 us and another thread changes the BO's mappings:
INFO: 171821 imports, VRAM used before 65 MiB, after 129 MiB modprobe: FATAL: Module amdgpu is in use.
Drop the reference, and free the sync object and the mutex of the kgdmem as well. With that VRAM usage returns to 65 MiB after the same test.
(cherry picked from commit b4d1eca2729840148ff2ea957724ce764ca5a94e)
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 b4d1eca2729840148ff2ea957724ce764ca5a94e
Frequently Asked Questions
What conditions are needed to trigger the resource leak?
A KFD dma-buf import must fail after BO validation, specifically when the interruptible BO reservation is interrupted and returns -ERESTARTSYS. The demonstrated trigger repeatedly imports a VRAM BO while SIGUSR1 is delivered frequently and another thread changes the BO's mappings.
What is the operational impact of repeated failures?
Each affected failure can retain the exported BO and its memory because the dma-buf reference is not released. Repeated imports can increase VRAM consumption and keep the amdgpu module in use, preventing module unload.
How can an administrator recognize a likely affected system?
Look for VRAM usage that continues to grow after failed KFD dma-buf import activity rather than returning to its prior level. Another indicator is modprobe reporting that the amdgpu module is in use after the workload has completed.
What can be done if the fix cannot be applied immediately?
Avoid or limit workloads that repeatedly import dma-bufs into KFD while imports may be interrupted by signals or concurrent BO mapping changes. Monitor VRAM consumption and avoid repeated failed import attempts to reduce accumulation of retained BOs.