Hi Ron,
Quick question: I was looking through the new tx queue rework davem did
and couldn't immediately tell whether or not the right thing happens
when a STA is removed while aggregation is in effect. Say mac80211 is
operating as an AP and a STA disassociates in the middle of an
aggregation session, I can see nothing that guarantees the aggregation
queue will be drained and returned to the pool etc.
Am I missing something?
johannes
On Fri, Jul 11, 2008 at 12:55 PM, Johannes Berg
<[email protected]> wrote:
> Hi Ron,
>
> Quick question: I was looking through the new tx queue rework davem did
> and couldn't immediately tell whether or not the right thing happens
> when a STA is removed while aggregation is in effect. Say mac80211 is
> operating as an AP and a STA disassociates in the middle of an
> aggregation session, I can see nothing that guarantees the aggregation
> queue will be drained and returned to the pool etc.
>
> Am I missing something?
>
We found that too. We will soon release dissociation patches for review.
Tomas
> johannes
>
> Quick question: I was looking through the new tx queue rework davem did
> and couldn't immediately tell whether or not the right thing happens
> when a STA is removed while aggregation is in effect. Say mac80211 is
> operating as an AP and a STA disassociates in the middle of an
> aggregation session, I can see nothing that guarantees the aggregation
> queue will be drained and returned to the pool etc.
>
> Am I missing something?
funny, but by chance this is a flow i am dealing with right now in STA
mode as part of patch meant to clean a bit the dissassoc flow. as to
AP mode - i did not look at the rework as well and i will have to
check this part, but IMHO this is more to do with the general flow of
assoc/dissassoc. if the MLME decides to remove an STA - by all means
it has to free its resources first - aggregation queues is one if them
of course.
Ron
> funny, but by chance this is a flow i am dealing with right now in STA
> mode as part of patch meant to clean a bit the dissassoc flow. as to
> AP mode - i did not look at the rework as well and i will have to
> check this part, but IMHO this is more to do with the general flow of
> assoc/dissassoc. if the MLME decides to remove an STA - by all means
> it has to free its resources first - aggregation queues is one if them
> of course.
True, but it's actually quite different in AP mode since there we just
get the request by userspace to remove the STA and never know what's
happening until we do. Whereas in STA mode, I'm not even sure we
completely remove the sta_info entry for the AP when we disassociate.
johannes