On 14.03.23 13:32, Johannes Berg wrote:
On Tue, 2023-03-14 at 13:28 +0100, Felix Fietkau wrote:
If you want to address this in the least invasive way, add [...],
a global lock to iwlwifi
I'm already fixing this, see
https://lore.kernel.org/r/5674c40151267fea1333f0eda1701b141bbaa170.camel@xxxxxxxxxxxxxxxx
, and a
per-AC lock inside ieee80211_handle_wake_tx_queue().
I'm not *entirely* sure per AC is sufficient given that you could
technically map two ACs to the same HW queue with vif->hw_queue[]?
But again, not really sure all that complexity is still worth it.
Per AC should be sufficient even in that case, because scheduling
happens per AC regardless of vif hw queue mapping.
- Felix
