CVE-2026-98389

In the Linux kernel, the following vulnerability has been resolved:

drm/amdgpu: skip gfx switch power profile during GPU reset

During resume from GPU reset, the gfx idle work may invoke switch power profile before the reset completes. This causes the following assert error because the register access occurs without first releasing the GPU reset semaphore:

[ 1576.768935] CR2: 0000559ea133ead0 CR3: 00000002e6c42000 CR4: 0000000000350ef0 [ 1576.768940] Call Trace: [ 1576.768944] [ 1576.768953] amdgpu device rreg+0x21/0x50 [amdgpu] [ 1576.769158] smu msg v1 send msg+0x1a4/0x6e0 [amdgpu] [ 1576.769437] smu cmn send smc msg with params ext+0xba/0x120 [amdgpu] [ 1576.769721] smu cmn send smc msg with param+0x33/0x40 [amdgpu] [ 1576.769993] smu v13 0 0 set power profile mode+0x192/0x2b0 [amdgpu] [ 1576.770267] smu bump power profile mode+0x5d/0x80 [amdgpu] [ 1576.770538] smu switch power profile+0xa4/0xf0 [amdgpu] [ 1576.770839] amdgpu dpm switch power profile+0x6f/0x90 [amdgpu] [ 1576.771210] amdgpu gfx profile idle work handler+0xe9/0x130 [amdgpu] [ 1576.771460] process one work+0x23e/0x6f0 [ 1576.771491] worker thread+0x1c4/0x380 [ 1576.771506] kthread+0x10c/0x150 [ 1576.771512] ? pfx worker thread+0x10/0x10 [ 1576.771518] ? pfx kthread+0x10/0x10 [ 1576.771530] ret from fork+0x314/0x390 [ 1576.771537] ? pfx kthread+0x10/0x10 [ 1576.771546] ret from fork asm+0x1a/0x30

(cherry picked from commit d93b1ff538ce9750c01e0dd0aa62575579c0fc08)