This patch implements two things:
* it refactors out the cabq hardware TXQ code from the cabq list setup code;
* it restores the per-TXQ lock (for hardware TXQs and the VAP
multicast TXQs) and wraps list operations around those.
The main reason for the second part of the patch is to try and reduce
latency/contention on grabbing locks when setting up the CABQ for
transmit. I saw the TX lock getting acquired and contended on when
also sending out CABQ frames and I didn't want to have this lead to
missed / stuck beacons.
I've not yet tested adhoc, EDMA NICs or legacy+TDMA.
I'd appreciate whatever testing people can give it. Please let me know
if you end up with more stuck beacons.
email@example.com mailing list
To unsubscribe, send any mail to "freebsd-wireless-unsubscr...@freebsd.org"