{"id":"CVE-2026-89828","published":"2026-09-16T11:16:49.493","lastModified":"2026-09-16T11:16:49.493","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu: Fix init ordering in amdgpu_vram_mgr_init()\n\ndrmm_cgroup_register_region() is called before INIT_LIST_HEAD() and\ngpu_buddy_init() in amdgpu_vram_mgr_init(). If it fails, the function\nreturns early and bypasses those initializations.\n\nSince adev->mman.initialized is set to true before amdgpu_vram_mgr_init()\nis called, a failure triggers amdgpu_ttm_fini(), which calls\namdgpu_vram_mgr_fini(), which then:\n\n - Calls list_for_each_entry_safe() on reservations_pending and\n   reserved_pages, whose list_head::next pointers are zero-initialized\n   (NULL). The loop does not recognize them as empty and dereferences NULL.\n\n - Calls gpu_buddy_fini(), which iterates free_trees[] unconditionally\n   via for_each_free_tree(). Since mm->free_trees is NULL\n   (never allocated), this dereferences NULL.\n\nBoth result in a kernel panic on the module load error path.\n\nFix by moving drmm_cgroup_register_region() to after the list and buddy\nallocator are fully initialized, so the teardown path is safe to run.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/3e234c6face8651045f46895dfe9c086ea64f03b","tags":[]},{"url":"https://git.kernel.org/stable/c/e184e46ca1ba652ab8053a10a51b39aad06d3f5c","tags":[]},{"url":"https://git.kernel.org/stable/c/e773798e14ac0aea54ca9676083b91f445e5bc59","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}