Re: Another fragmentation multiqueue kludge

["Tomas Winkler" <tomasw@gmail.com>]

On Thu, Jul 24, 2008 at 5:59 PM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Thu, 2008-07-24 at 17:55 +0300, Tomas Winkler wrote:
>
>> > What happens if you just invert the
>> >                if (__netif_subqueue_stopped(local->mdev, i))
>> >                        continue;
>>
>> > check in ieee80211_tx_pending to read
>> >                if (!__netif_subqueue_stopped(local->mdev, i))
>> >                        continue;
>> >
>> > as I suggested yesterday?
>>
>> This is wrong since you it will actually starts queues that driver
>> didn't ask to start in the next line.
>
> Ok.
>
>> As I see it any failure in the driver's tx path will cause moving to
>> pending queue, except packets on AMPDU
>> queues that will be dropped.
>
>> That's correct for stopping because of queue overhead, not for errors.
>
> Tough luck, just drop packets on errors. If that matters to you, fix the
> driver to not run into these errors.

I don't really mind. I just want to be environment friendly and don't
want to be asked by other driver maintainers to fix their drivers :)

>> So till someone fix adm driver consider this patch It worked quite
>> well. I will also send patch that fixes the behavior in the iwlwifi
>> driver.
>
> If you fix iwlwifi we don't need this patch, and then I can instead
> remove the code from mac80211, but we can just as well merge this first
> since we'll get rid of it again.

Okay I will post both patches.
Thanks for review
Tomas