Re: [RFT 0/3] ath9k: more PCU locking enhancements

[Ben Greear <greearb@candelatech.com>]

On 10/26/2010 03:03 PM, Luis R. Rodriguez wrote:
> On Tue, Oct 26, 2010 at 2:59 PM, Ben Greear<greearb@candelatech.com>  wrote:
>> On 10/26/2010 01:40 AM, Luis R. Rodriguez wrote:
>>>
>>> Here is some more PCU locking enhancements I tested today
>>> while trying to resolve the WARN() that happens when we
>>> try to stop RX DMA and fail. While working on that I figured
>>> I'd work on the TX DMA stuff too, here's a shot at it. I
>>> can no longer get TX / RX DMA rants, please test and let
>>> me know if you do. I only tried some basic testing like
>>> rmmoding while scannign, which typicallly produced some
>>> errors. Now I don't get squat.
>>>
>>> Ben if you can test wit your super proprietary application
>>> that'd be great.
>>>
>>> This also simplifies locking considerably.
>>>
>>> This doesn't break suspend so I'm happy. It also depends
>>> on the last RX DMA fixes I had posted earlier. If you'd
>>> like to get an all-in-one patch of all my patches pending
>>> you can wget this file and git am it:
>>>
>>>
>>> http://www.kernel.org/pub/linux/kernel/people/mcgrof/patches/tmp/pending-mcgrof-2010-10-26-v1.patch
>>> sha1sum: 874a3cc1a57f7e26ad191cd7b5045315f94c5823
>>
>> I have done some initial testing on the combined patch on top of today's
>> wireless-testing tree.  I also have the memory-barrier patch applied to
>> ath9k, as that is still not upstream.  I have no idea if it has any affect
>> or not (I'm on x86..seems that wmb() stuff was mostly for other platforms?).
>>
>> So far, it is showing zero problems, certainly no memory poison issues.
>>
>> The wireless-testing tree has some lockdep warning related to a mouse driver
>> that disables lockdep early, so it's possible there are lockdep issues
>> waiting.
>>
>> I will let this test run for a while, but it already looks more stable
>> than before, so:
>>
>> Tested-by:  Ben Greear<greearb@candelatech.com>
>
> Awesome! Thanks for testing. So how about the TX dma rants, do you
> still get those?

I've seen no rants at all.

I'm using my standard 130 STAs trying to associate with an AP that will only
take around 30 associations at a time..leads to 100 or so supplications constantly
trying to scan and associate, and I've never seen this test run more than about
10 minutes without poison warnings and/or lockups.  I used to be happy when it got that far :)

We'll do some more interesting tests with APs + STAs, as well as throughput tests, etc
as soon as we get time.  (Basically, hoping for same support that we pushed into ath5k
recently).

If you are aware of any existing/expected issues with multiple APs + STAs,
please let me know.

Thanks,
Ben

>
>    Luis


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com