$ dmesg | tail -5   (after both runs — only benign noise, no corruption/slab messages)
vinumioctl: invalid ioctl from process 1265 (ls): 40906468
vinumioctl: invalid ioctl from process 1265 (ls): 40906468
vinumioctl: invalid ioctl from process 1272 (df5094): 40906468
vinumioctl: invalid ioctl from process 1285 (df5094): 40906468
vinumioctl: invalid ioctl from process 1316 (df5094): 40906468

(0x40906468 = _IOR('d', 104, 144) = DIOCGPART — fstat/stat probes of the
vinum control device by ls and by the PoC processes' own fd handling; rejected
with EINVAL by the superdev ioctl switch. Unrelated to the race under test.)
No "Duplicate free", no slab/allocator complaints, no panics across 46.5M
raced elements.