{
  "finding_id": "DF-5183",
  "status": "reproduced",
  "verdict": "REPRODUCED deterministically on the QEMU guest (same lwqdemo.ko as DF-5182): tsleep(timo=(int)MAX_SCHEDULE_TIMEOUT) with NO waker returned EWOULDBLOCK (rv=35) after 2 ticks - the infinite timeout that wait_on_bit() (linux_wait.c:141-144) passes truncates from (unsigned long)LONG_MAX to int -1, and _callout_setup_quick() clamps the negative expiry to a next-tick callout, so every infinite wait_on_bit() is actually a 1 Hz polling loop. wait_on_bit_timeout(MAX_SCHEDULE_TIMEOUT) on a set bit exited only when the bit was cleared 5s later (rv=0, 396 ticks), confirming poll-driven exit. The mode-as-flags half (TASK_INTERRUPTIBLE=1 never maps to PCATCH=0x100, so interruptible waits are silently unkillable; low bits land in PWAKEUP_CPUMASK) is source-certain but not runtime-demonstrable from a KLD (kernel threads take no signals); latent because in-tree callers pass TASK_UNINTERRUPTIBLE only.",
  "reproduced": true,
  "impact": "dos",
  "confidence": "certain",
  "exploit_chain": "kernel-internal (drm drivers via linux/wait_bit.h; i915 reset paths are active callers). Consequences: unkillable TASK_INTERRUPTIBLE waits for any future caller; 1 Hz timer-fire + context-switch per infinite waiter (CPU cost, and it masks stranded waiters); no unprivileged direct trigger on this guest (no drm device).",
  "evidence": "run.log LWQ-F3 lines (two boots, identical); VERDICT.md source chain; kern_timeout.c:971-972 clamp; kern_synch.c:512 int timo.",
  "kernel_refs": [
    "sys/dev/drm/linux_wait.c:105",
    "sys/dev/drm/linux_wait.c:118",
    "sys/dev/drm/linux_wait.c:141",
    "sys/dev/drm/linux_wait.c:143",
    "sys/dev/drm/include/linux/sched.h:72",
    "sys/sys/param.h:324",
    "sys/kern/kern_timeout.c:971"
  ],
  "poc_changes": "No seed; experiment added to the DF-5182 module (deterministic tsleep-level probe + a wait_on_bit_timeout-level confirmation).",
  "attempts": 2,
  "guest_uname": "DragonFly dfbsd 6.5-DEVELOPMENT DragonFly 6.5-DEVELOPMENT #0: Thu Jul  2 06:02:54 UTC 2026  root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC  x86_64",
  "runtime_sec": 480,
  "guest_dirty": 0,
  "build_cmd": "cd /root/lwqdemo && make SYSDIR=/usr/src/sys (build.log, rc=0)",
  "run_cmd": "kldload ./lwqdemo.ko && sleep 8 && dmesg | grep LWQ-F3 (run.log)",
  "code_hash": "1978c8c3c8887476ea99e9125e41040d68b2484aa4895ca87ab0ad09d9850632",
  "notes": "Distinct root causes from DF-2181 (absolute-vs-relative expiry test) and DF-2182 (missing interlock) at the same call site: this covers the mode->flags mapping and the unsigned-long->int timeout truncation. Also corrects the DF-2182 record: a wait_on_bit() waiter polls at 1 Hz rather than sleeping forever.",
  "recommended_fix": "Translate mode to PCATCH and clamp timeout >= INT_MAX / == MAX_SCHEDULE_TIMEOUT to timo=0 before calling tsleep() in wait_on_bit_timeout().",
  "fix_status": "not_tested",
  "fix_baseline_reproduced": true,
  "fix_patched_reproduced": false,
  "fix_verdict": "fix.diff authored (mode->PCATCH translation + timeout clamp); not kernel-rebuild-validated (Low-severity mechanical change).",
  "fix_evidence": "fix.diff in this pack"
}