More than one ath10k NIC will not load (bisected)

[Michal Kazior]

On 4 February 2015 at 16:23, Ben Greear <greearb at candelatech.com> wrote:
> On 02/04/2015 01:35 AM, Michal Kazior wrote:
>>
>> On 4 February 2015 at 10:07, Kalle Valo <kvalo at qca.qualcomm.com> wrote:
>>>
>>> Ben Greear <greearb at candelatech.com> writes:
>>>
>>>>> Hmm.. This removes warm_reset in probe function but I fail to see how
>>>>> this could end up not loading one of the NIC *silently*?
>>>>>
>>>>> Anyway there's a pending patch which adds the reset back:
>>>>>
>>>>> https://github.com/kvalo/ath/commit/bdcd6f4e4ac5d2d2a56da4813f56655e6db0ee45
>>>>> . You might want to try it and see if it helps.
>>>>
>>>>
>>>> Reverting the patch made it work again for me.
>>>>
>>>> I don't understand that code well, but perhaps you are disabling
>>>> a shared interrupt that silently stops the second NIC from
>>>> being able to do it's thing?
>>>>
>>>> Do you have a PC with 2 NICs in it that you could try yourself?
>>>>
>>>> I can grab you the logs of a failure to boot later today.
>>>
>>>
>>> What should we do with this one? I didn't look at the details yet, but
>>> do we have any other option than to revert?
>>
>>
>> I believe this is an issue in Ben's userspace (he sent me logs
>> privately) or some sort of kernel event bug. It basically looked like
>> this: both devices were detected by ath10k and both started
>> register_work. One of the devices loaded all the way while the other
>> tried to load a few non-existing firmware files and it stopped. Few
>> minutes later there was a hung task splat pointing to
>> request_firmware() called from ath10k suggesting userspace didn't
>> handle firmware request.
>>
>> The "offending" patch effectively removed 200ms from probe() in
>> ath10k. This could've change timing on request_firmware() calls on
>> Ben's system. Btw. the 200ms is back again now with
>> 1a7fecb766c83dace747f42b25bbb544b00a0163 ("ath10k: reset chip before
>> reading chip_id in probe").
>>
>> Marek tried running 2 qca988x on his laptop some time ago (with and
>> without the extra timing) and didn't have any issues.
>
>
> I can retry my system with stock Fedora 20 and see if it works there.
>
> If not, then I think it still needs to be worked on...you agree?

In the driver? I'd argue. I don't see how ath10k could make
request_firmware() hang, do you?


Michał