DF-2849 race2 runs (churn harness)
===================================

run.2.log attempt (STOCK kernel #0): 4 racers (cpus 1-5) + detacher
(cpu 2), 2M cycles x 8 tasks.  The ssh command hit its local timeout
and the guest became unresponsive with QEMU still running at ~230%
CPU; serial console showed no panic trace, last line:

  gtq_race2: 4 racers + detacher running

ATTRIBUTION (important for re-verifiers): this death could NOT be
attributed to the bug.  The harness racers are pinned LWKT
daemon-priority threads spinning in `while (!stop)`; such spinners
starve user/protocol threads on their CPUs and wedge the guest by
themselves (lwkt_yield() only yields to equal-priority threads).  The
same wedge reproduced with the yield-equipped harness on the FIXED
kernel (QEMU alive at 466% CPU, no panic on serial for 15+ minutes),
while the captured stock-kernel panic (panic.txt) came from the
lightweight gtq_null variant.  Treat the race2 wedge runs as
harness-induced starvation, NOT as corruption evidence.

Fixed-kernel lightweight churn (gtq_race, 1 racer, 200k cycles) after
the fix build (#1 kernel):

  gtq_race: detacher finished 200000 attach/detach cycles
  gtq_race: enq_ok=1685324 enq_eagain=1807005 enq_null=1086196 fn_count=6109 -> 6109
  gtq_race: no post-stop growth observed this run
  FIXED_KERNEL_UNLOAD_OK