CVE-2026-97424

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

drm/amdgpu/ras: add ras suspend callback and use it for cp ecc error irq

cp ecc error irq is acquired in amdgpu gfx ras late init() but released in gfx v9 0 hw fini(), so the put site has to query amdgpu irq enabled() because the get is skipped on SR-IOV VF.

ras late init / ras fini have no suspend counterpart, so move the put to amdgpu gfx ras suspend() / amdgpu gfx ras fini() and add a matching ras suspend callback that is invoked from amdgpu ras suspend() before disable all features(). The get and put now sit in the same place and check the same condition (not VF, funcs registered), no refcount querying needed.

An active flag gates ras fini so the suspend-then-unload-without-resume path falls into amdgpu ras block late fini default() instead of double-releasing what ras suspend already cleaned up.

Drop the cp ecc error irq put from gfx v9 0 hw fini(). gfx v8 0 manages cp ecc error irq locally and is unaffected; no other GFX generation has this IRQ.