Re: TXQ real_num_tx_queues comments/questions

[Johannes Berg <johannes@sipsolutions.net>]

[-- Attachment #1: Type: text/plain, Size: 1011 bytes --]

On Tue, 2008-07-15 at 02:28 -0700, David Miller wrote:
> From: Johannes Berg <johannes@sipsolutions.net>
> Date: Tue, 15 Jul 2008 11:25:23 +0200
> 
> > > It seems to me that a simple synchronize_net() call near the end of
> > > agg queue removal would solve your problem as-is, wouldn't it?
> > 
> > Well, as far as I can tell we can't do that there because it's in
> > tasklet context/called under spinlocks so it has to be atomic. I think.
> 
> Right, my bad.

Also, I don't actually think it is sufficient since select_queue isn't
called under rcu lock, so even if we synchronize_net() somewhere there
we can still end up not having waited long enough, no?

> It seems that there can end up being holes in the queue
> space on mac80211, which slightly complicates matters.

Yes, but why does it complicate matters? From mac80211's POV it's just
that select_queue must have known about the removed queue when the
requeue is done. And we can defer the requeue to a workqueue.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]