TX hang and EAPOL stuff fixed in HEAD; please test!

[Raphael Kubo da Costa]

Adrian Chadd <adrian at freebsd.org> writes:

> Would those using atheros NICs on -HEAD please update and give it a
> good thrashing?
>
> I've fixed a couple of things:
>
> * The driver was not able to transmit EAPOL frames if the send queue
> ran out of space. This is why my UDP TX tests were failing - the group
> rekey would kick in, fail to transmit the EAPOL "Yes I've done that"
> ack frame back to the AP and the AP would kick the station off.
> * I've modified the transmit path to always keep a link pointer around
> rather than NULL'ing it out and reinitialising TX DMA every time. This
> is the "blessed" way to handle DMA. However, I'm still trying to
> verify that I've actually done this correctly in all ways so I'd
> really like this tested.
>
> I'd like to start merging in my power save and reassociation fixes
> from my test branch but to do that I'd really like to have this code
> fully tested.

As of r250406, I get a kernel panic whenever I bring wlan0 up because
the assertion in if_ath.c:4317 gets triggered.

I have reverted r250391 for now, removed sys/contrib/sys/dev/ath now
that sys/contrib/dev/ath is in place and commented out the "ath0:
ath_edma_recv_proc_queue: handled npkts 0" messages that Joshua also
mentioned. Everything seems to be working fine (ie. connecting to my
crappy WPA2 router).