[PATCH 2/2] ath10k: fix scheduling while atomic bug

[Michal Kazior]

On 1 October 2013 18:35, Kalle Valo <kvalo at qca.qualcomm.com> wrote:
> Michal Kazior <michal.kazior at tieto.com> writes:
>
>> Recent WMI/HTC changes broke WEP with multiple
>> keys. If WMI had no HTC TX credits to submit
>> command for default wep index update it would
>> trigger a bug.
>>
>> This simply moves the wep key index update to a
>> worker.
>>
>> The key update may happen some time after first
>> frame with a different wep key has been sent (i.e.
>> some frames will be sent with old key). This was
>> the case before too as WMI commands were
>> asynchronous.
>>
>> Signed-off-by: Michal Kazior <michal.kazior at tieto.com>
>
> This looks problematic. Basically you just delay sending the WMI
> command, but there's no guarantee that we actually have free credits at
> the time of transmission. So to me it looks like this fixes the issue
> just by luck.

One thing at a time.

This patch fixes 'scheduling while atomic' bug that was introduced
with recent HTC/WMI changes.

Multiple wep key handling had the same issue you point out before
HTC/WMI changes.

To fix the issue you mention we'd need to stop mac80211 queues, queue
frame(s) with a new key on internal queue, schedule a work. The work
should set new wep key index, send pending frames from internal queue
and wake mac80211 queues. This is a bit more complex than it sounds
though.


Michał