Re: [ath9k-devel] ath9k: massive unexplained latency in 2.6.27 (rc5, rc6, probably others)

["Steven Noonan" <steven@uplinklabs.net>]

On Fri, Sep 19, 2008 at 4:53 PM, Luis R. Rodriguez
<lrodriguez@atheros.com> wrote:
> On Fri, Sep 19, 2008 at 04:31:20PM -0700, Steven Noonan wrote:
>>
>> I'm seeing the same latency issue I reported originally with the latest
>> -tip kernel and WITHOUT irqpoll.
>
> Damn it.
>
>> Apparently there's something else to this...
>
> Steven,
>
> Can you reproduce this? If so can you tell us how? If you can reproduce
> can you try to test this with wireless-testing as well? That is still
> based on 2.6.26-rc6 but its just has all the new wireless development
> stuff queued up for 2.6.28 so far. Everything else is the same.

It's a weird problem, because I'm able to reproduce it by running a
kernel with ath9k loaded for a few hours. In this particular case, I
only ran this kernel for about 3.5 hours , but the delay when VT
switching was getting noticeable. Unload/reload of ath9k and the
latency disappeared again. It's definitely ath9k's doing, but I don't
know how. I'm finding the delay is most noticeable after 7 hours
uptime with ath9k loaded.

I'll pull wireless-testing and run that kernel for another 8 hours.
What's the URL for the repo? Surely you don't mean the
wireless-testing on git.kernel.org that hasn't been touched for almost
a week?

I kind of wish there was a better debugging facility in ath9k. With
ATH_DBG_INTERRUPTS, I see several hundred messages per second, which
fills log files up like crazy (I switched to ftrace_printk as Ingo
suggested, which has alleviated this somewhat). Is there a way to make
it less verbose (perhaps so that it prints data at a rate of no more
than 2 or 3 messages per second)? Or maybe the rate at which it prints
this stuff is indicative something's wrong?

- Steven