{"id":"CVE-2026-98391","published":"2026-10-14T14:17:49.980","lastModified":"2026-10-14T14:17:49.980","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/xe: Guard page-fault worker with runtime PM check\n\nDuring VM teardown, the VM's runtime PM reference is dropped\nasynchronously, allowing the device to autosuspend while stale page\nfaults belonging to the now-dead VM are still queued. When the\npage-fault worker later tries to ack one of these, it calls into\nguc_ct_send_locked() on an already-suspended device, tripping:\n\n  Assertion `!xe_pm_runtime_suspended(xe)` failed!\n  WARNING at xe_device.c:1267 xe_device_assert_mem_access+0x11c/0x140 [xe]\n\nA live VM/exec queue always holds a PM reference while it has\noutstanding work, so if the device is suspended at ack time, the\nowning context is already gone and the fault is stale.\n\nTake a runtime PM reference across the entire pagefault\nqueue worker to safely deliver acks for torn-down VMs.\n\nv3:\n - Move PM ref to the generic xe_pagefault_queue_work using\n   guard(xe_pm_runtime)(xe) instead of tracking it in the GuC\n   backend(Matt Brost).\n\nv2:\n - Hold PM ref across the entire batch (begin/end) instead of per-ack.\n   This prevents the device from autosuspending mid-batch, which would\n   leave write_only acks written but the end flush skipped, and skip\n   counter++, desyncing the cadence check.(Himal)\n - Add a comment explaining stale faults.(Himal)\n\n(cherry picked from commit fcc2431d2213dc4d04250c4f1ae87d9c3ae0d455)\n[Rodrigo: Added xe_device struct for compatibility while cherry-picking]","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/20fce5b34b21a995839743b4917a1edd2fd503ba","tags":[]},{"url":"https://git.kernel.org/stable/c/25764620ec64c0142e2c87c41787f45540cf09ff","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}