Hi Miri,

You're right -- cancel_delayed_work_sync() handles re-arming by design,
and the work is scheduled with a 1-hour delay (IWL_PTP_WRAP_TIME =
3600 * HZ), so it cannot realistically fire again during the
microsecond-scale teardown window.

The real issue is simply that the delayed work can execute between
ptp_clock_unregister() and cancel_delayed_work_sync(), observing
partially cleared PTP state.

I'll send a v2 with a corrected commit message, and also split the
patch into two (one per sub-driver) as Simon suggested.

Thanks for the review!

Junjie