{"id":"CVE-2026-97424","published":"2026-09-24T17:17:20.340","lastModified":"2026-09-25T13:17:25.963","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu/ras: add ras_suspend callback and use it for cp_ecc_error_irq\n\ncp_ecc_error_irq is acquired in amdgpu_gfx_ras_late_init() but\nreleased in gfx_v9_0_hw_fini(), so the put site has to query\namdgpu_irq_enabled() because the get is skipped on SR-IOV VF.\n\nras_late_init / ras_fini have no suspend counterpart, so move the\nput to amdgpu_gfx_ras_suspend() / amdgpu_gfx_ras_fini() and add a\nmatching ras_suspend callback that is invoked from\namdgpu_ras_suspend() before disable_all_features().  The get and\nput now sit in the same place and check the same condition (not\nVF, funcs registered), no refcount querying needed.\n\nAn active flag gates ras_fini so the\nsuspend-then-unload-without-resume path falls into\namdgpu_ras_block_late_fini_default() instead of double-releasing\nwhat ras_suspend already cleaned up.\n\nDrop the cp_ecc_error_irq put from gfx_v9_0_hw_fini().  gfx_v8_0\nmanages cp_ecc_error_irq locally and is unaffected; no other GFX\ngeneration has this IRQ.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/001824103f2fe6ffd2380afddf81f43b446035b2","tags":[]},{"url":"https://git.kernel.org/stable/c/e3829992dd9fa0a82511af4f01733fc854cd15a5","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}