{"id":"CVE-2026-97900","published":"2026-09-25T11:17:16.787","lastModified":"2026-09-25T11:17:16.787","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/drm_exec: fix up contended obj when num_objects is 0\n\ndrm_exec_prepare_array() silently returns success without calling\ndrm_exec_lock_contended() when num_objects is zero. This breaks the\ninvariant upheld by drm_exec_lock_obj(), where every entry point into\nthe locking sequence must first attempt to lock any previously\ncontended object before proceeding.\n\nDrivers that chain multiple drm_exec_prepare_array() calls per\ndrm_exec_until_all_locked() iteration (e.g. amdgpu's userq signal/wait\nioctls, which prepare separate read and write BO arrays) can pass an\nempty array for one of the two calls. If contention is hit while\npreparing the non-empty array, exec->contended is set and the loop\nretries; on retry, the empty-array call preceding it is a no-op that\nnever clears exec->contended, so drm_exec_retry_on_contention()\nimmediately jumps back to the top of the loop without ever reaching\nthe call that would resolve the contention. This spins forever.\n\nFix it by having drm_exec_prepare_array() call drm_exec_lock_contended()\ndirectly when num_objects is zero, so a pending contended object dont\nloop infinitely.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/159720704d9d652b64390c11fb971e15b0a78d23","tags":[]},{"url":"https://git.kernel.org/stable/c/831cd124ce99c3e61266bed30519d52f7c26abed","tags":[]},{"url":"https://git.kernel.org/stable/c/a565022c02f218ec9789c2baf4690792e7a48cbb","tags":[]},{"url":"https://git.kernel.org/stable/c/f3e74866018dab793ebee1fdf0ef34f7271d1f8c","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}