run.3.log — DF-5793 baseline, SIGKILL latency while parked, stock kernel

$ (/home/maxx/dup_directed 60 &) ; sleep 3; PID=$(pgrep dup_directed | head -1)
pid=1315
$ kill -9 1315 ; poll kill -0 every 50ms until gone
SIGKILL-to-death sec: 1
$ dmesg | grep -c "target descriptor"
32

Interpretation: the dup2 thread was inside tsleep(fdp, 0, "fdres", hz)
(flags=0, no PCATCH); SIGKILL could not be delivered until the 1s timeout
expired and the retry loop completed. On a persistent reservation (blocked
kmalloc under memory pressure, SCM_RIGHTS multi-fd receive) the process is
unkillable for as long as the peer holds the reservation.