Описание
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: Add TLB flush after MES queue eviction/suspension MES (Micro Engine Scheduler) does not perform heavy-weight TLB invalidation after unmapping queues, unlike HWS which does this automatically. This causes a race condition where in-flight DMA descriptors can access memory that has been unmapped, leading to page faults and GPU queue hangs during SVM page migration. The issue manifests as KFDSVMRangeTest.MultiThreadMigrationTest failures on gfx1151 (Strix Point) with XNACK mode 1 enabled - the GPU compute queue hangs with packets submitted but never consumed. Add kfd_flush_tlb() calls after MES queue removal in two locations:
- evict_process_queues_cpsch(): after all queues removed during eviction
- suspend_queues(): after debug/criu queue suspension (with mem_fence barrier)
This ensures all in-flight memory accesses from unmapped queues are
flushed before memory is freed or migrated.
(cherry picked from commit f5c4f88e0f9c45a8fb9dfac0c1df726c95e41b77)
A flaw was found in the Linux kernel's drm/amdkfd component. This vulnerability arises from a race condition where the Micro Engine Scheduler (MES) fails to perform a necessary Translation Lookaside Buffer (TLB) invalidation after unmapping queues. This oversight allows in-flight Direct Memory Access (DMA) descriptors to access memory that has already been unmapped, leading to page faults and potential GPU queue hangs, particularly during Shared Virtual Memory (SVM) page migration.
Отчет
This vulnerability is rated as an Important severity because an authenticated local user with access to GPU compute interfaces could trigger kernel memory corruption or induce a denial of service. The flaw only affects environments running on compatible AMD graphics hardware that actively load the GPU compute driver subsystem. While exploitation requires local access and specific hardware scheduling features, the potential to disrupt host kernel integrity or cause persistent system hangs justifies an Important rating rather than Moderate.
Меры по смягчению последствий
Prevent the amdgpu kernel module from loading:
- Create a configuration file to prevent module loading:
echo "install amdgpu /bin/true" > /etc/modprobe.d/disable-amdgpu.conf
- Rebuild the initial RAM disk:
dracut -f
Warning: A system reboot is required for this change to take effect if the driver is currently loaded. Disabling the amdgpu driver prevents graphical acceleration and GPU compute operations on systems with AMD graphics hardware, and display servers may fail or revert to basic framebuffer mode.
Затронутые пакеты
| Платформа | Пакет | Состояние | Рекомендация | Релиз |
|---|---|---|---|---|
| Red Hat Enterprise Linux 6 | kernel | Affected | ||
| Red Hat Enterprise Linux 7 | kernel | Affected | ||
| Red Hat Enterprise Linux 7 | kernel-rt | Affected | ||
| Red Hat Enterprise Linux 9 | kernel-rt | Affected | ||
| Red Hat OpenShift Container Platform 4 | openshift/ose-rhel-coreos-8 | Affected | ||
| Red Hat OpenShift Container Platform 5 | openshift/ose-rhel-coreos-9 | Affected | ||
| Red Hat Enterprise Linux 10 | kernel | Fixed | RHSA-2026:88477 | 23.10.2026 |
| Red Hat Enterprise Linux 8 | kernel-rt | Fixed | RHSA-2026:82806 | 14.10.2026 |
| Red Hat Enterprise Linux 8 | kernel | Fixed | RHSA-2026:82807 | 14.10.2026 |
| Red Hat Enterprise Linux 9 | kernel | Fixed | RHSA-2026:81552 | 13.10.2026 |
Показывать по
Дополнительная информация
Статус:
EPSS
7.8 High
CVSS3
Связанные уязвимости
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: Add TLB flush after MES queue eviction/suspension MES (Micro Engine Scheduler) does not perform heavy-weight TLB invalidation after unmapping queues, unlike HWS which does this automatically. This causes a race condition where in-flight DMA descriptors can access memory that has been unmapped, leading to page faults and GPU queue hangs during SVM page migration. The issue manifests as KFDSVMRangeTest.MultiThreadMigrationTest failures on gfx1151 (Strix Point) with XNACK mode 1 enabled - the GPU compute queue hangs with packets submitted but never consumed. Add kfd_flush_tlb() calls after MES queue removal in two locations: - evict_process_queues_cpsch(): after all queues removed during eviction - suspend_queues(): after debug/criu queue suspension (with mem_fence barrier) This ensures all in-flight memory accesses from unmapped queues are flushed before memory is freed or migrated. (cherry picked from com...
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: Add TLB flush after MES queue eviction/suspension MES (Micro Engine Scheduler) does not perform heavy-weight TLB invalidation after unmapping queues, unlike HWS which does this automatically. This causes a race condition where in-flight DMA descriptors can access memory that has been unmapped, leading to page faults and GPU queue hangs during SVM page migration. The issue manifests as KFDSVMRangeTest.MultiThreadMigrationTest failures on gfx1151 (Strix Point) with XNACK mode 1 enabled - the GPU compute queue hangs with packets submitted but never consumed. Add kfd_flush_tlb() calls after MES queue removal in two locations: - evict_process_queues_cpsch(): after all queues removed during eviction - suspend_queues(): after debug/criu queue suspension (with mem_fence barrier) This ensures all in-flight memory accesses from unmapped queues are flushed before memory is freed or migrated. (cherry picked from
drm/amdkfd: Add TLB flush after MES queue eviction/suspension
In the Linux kernel, the following vulnerability has been resolved: d ...
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: Add TLB flush after MES queue eviction/suspension MES (Micro Engine Scheduler) does not perform heavy-weight TLB invalidation after unmapping queues, unlike HWS which does this automatically. This causes a race condition where in-flight DMA descriptors can access memory that has been unmapped, leading to page faults and GPU queue hangs during SVM page migration. The issue manifests as KFDSVMRangeTest.MultiThreadMigrationTest failures on gfx1151 (Strix Point) with XNACK mode 1 enabled - the GPU compute queue hangs with packets submitted but never consumed. Add kfd_flush_tlb() calls after MES queue removal in two locations: - evict_process_queues_cpsch(): after all queues removed during eviction - suspend_queues(): after debug/criu queue suspension (with mem_fence barrier) This ensures all in-flight memory accesses from unmapped queues are flushed before memory is freed or migrated. (cherry picked fr...
EPSS
7.8 High
CVSS3