Johannes Berg [off-list ref] writes:

On Wed, 2018-09-05 at 14:50 +0200, Toke Høiland-Jørgensen wrote:

(uh, sorry, bit late ...)

quoted

quoted

The problem isn't so much how we handle it in mac80211 for the queueing,
but how we deal with things like A-MSDU and how we present it to the
driver ... for iwlwifi at least we'd really like to have only data
frames so we can map it directly to the hardware queue ...

Ah, I see. No, then just putting them at the head of a different TXQ
probably won't work...

Are you mapping TXQs to hardware queues dynamically as they empty and
re-fill? Presumably you'll have cases where you don't have enough HWQs?

Depends on the hardware. Newer hardware has basically unlimited HWQs
(something on the order of 512 IIRC).

Older hardware sort of maps them dynamically, but not too dynamic, we
also have 32 (ish) queues there. We only free them if we need a new one
and don't have one, and yes, theoretically we can run out and then we
may have to share a single hardware queue for multiple TXQs, but it
basically never happens in practice.

Right, gotcha, thanks for the explanation :)

-Toke