[patch] if_ath_tx: change interrupt scheduling deferral

[Adrian Chadd]

On 13 September 2011 18:14, Berislav Purgar <bpurgar at gmail.com> wrote:

> channel 6:ht/20
>
> AR5416(iperf -c) ---> AR9280(iperf -s)  i get TCP ~ 9MB/s  UDP ~ 13MB/s
> AR9280(iperf -c) ---> AR5416(iperf -s) i  get TCP ~ 4MB/s  UDP ~ 8MB/s

Hm, 90/130 mbit; that's what it should be. But I wonder why it's so bad one way.
Time to do a bit of debugging.

> channel 6:ht/40
>
> AR5416(iperf -c) ----> AR9280(iperf -s) i get TCP ~ 7MB/s UDP ~ 10MB/s
> AR9280(ipef -c) ----> AR5416(iperf -s) i get TCP ~ 15MB/s UDP ~ 24MB/s

Nice. :)

> after testing it in both direction i look at log files and found these error
> on AR5416 side
>
> Sep 13 11:54:25 rut kernel: ath0: device timeout
> Sep 13 11:54:25 rut kernel: ath0: ath_tx_tid_drain: node 0xc421e000: tid 0:
> txq_
> depth=0, txq_aggr_depth=0, sched=1, paused=1, hwq_depth=0, incomp=0,
> baw_head=0,
>  baw_tail=0 txa_start=395, ni_txseqs=5266
> Sep 13 11:54:30 rut kernel: ath0: device timeout
> Sep 13 11:54:30 rut kernel: ath0: ath_tx_tid_drain: node 0xc421e000: tid 0:
> txq_
> depth=0, txq_aggr_depth=0, sched=1, paused=1, hwq_depth=0, incomp=0,
> baw_head=0,
>  baw_tail=0 txa_start=51382, ni_txseqs=51383
>
>
> and on AR9280
>
> Sep 13 11:55:34 pila kernel: ath2: device timeout
> Sep 13 11:55:34 pila kernel: ath2: ath_tx_tid_drain: node 0xc3065000: tid 0:
> txq
> _depth=0, txq_aggr_depth=0, sched=1, paused=1, hwq_depth=0, incomp=0,
> baw_head=0
> , baw_tail=0 txa_start=2826, ni_txseqs=4496
> Sep 13 11:55:39 pila kernel: ath2: device timeout
> Sep 13 11:55:39 pila kernel: ath2: ath_tx_tid_drain: node 0xc3065000: tid 0:
> txq
> _depth=0, txq_aggr_depth=0, sched=1, paused=1, hwq_depth=0, incomp=0,
> baw_head=0
> , baw_tail=0 txa_start=42270, ni_txseqs=42271

There shouldn't be errors on the AR5416, unless that chip also has the
errant behaviour. I know the AR9280.
Are you able to try the previous svn revision (from a couple days ago)
and see if they have errors?

I think I now know what the problem is, I'll work on trying to fix it.

Also, paused=1 is a bit odd. I'll have to do some digging.

The good news is that I do have an AR5416/AR9280 combination to test
with. I'll just have to whack the AR5416 in a hostap mips board.


Adrian