On Sunday 26 July 2009 19:01:35 Larry Finger wrote:
> Michael Buesch wrote:
> > On Sunday 26 July 2009 17:37:10 Larry Finger wrote:
> >>
> >> The revised printk shows
> >>
> >> b43-phy0 warning: DMA queue overflow with free_slots = 0
> >
> > Ok, it's a mac80211 bug then.
>
> Perhaps not. I also got the ring->stopped WARN_ON.
Yeah. That just indicates that it _is_ a mac80211 bug.
> Is it possible that
> a second transmission was held at the spin_lock_irqsave() when the
> queues were stopped. If that were the case, the following patch should
> fix the problem.
Nah, that patch cannot fix anything. You just move the check inside of the lock.
>From the outside view it stays the same. The lock makes it atomic.
> ---
>
> Index: wireless-testing/drivers/net/wireless/b43/dma.c
> ===================================================================
> --- wireless-testing.orig/drivers/net/wireless/b43/dma.c
> +++ wireless-testing/drivers/net/wireless/b43/dma.c
> @@ -1334,17 +1334,20 @@ int b43_dma_tx(struct b43_wldev *dev, st
> spin_lock_irqsave(&ring->lock, flags);
>
> B43_WARN_ON(!ring->tx);
> +
> + if (unlikely(free_slots(ring) < TX_SLOTS_PER_FRAME)) {
> + if (b43_debug(dev, B43_DBG_DMAVERBOSE)) {
> + b43warn(dev->wl, "DMA queue overflow with %d"
> + " free slots\n", free_slots(ring));
> + err = -ENOSPC;
> + goto out_unlock;
> + }
> + }
> /* Check if the queue was stopped in mac80211,
> * but we got called nevertheless.
> * That would be a mac80211 bug. */
> B43_WARN_ON(ring->stopped);
>
> - if (unlikely(free_slots(ring) < TX_SLOTS_PER_FRAME)) {
> - b43warn(dev->wl, "DMA queue overflow\n");
> - err = -ENOSPC;
> - goto out_unlock;
> - }
> -
> /* Assign the queue number to the ring (if not already done before)
> * so TX status handling can use it. The queue to ring mapping is
> * static, so we don't need to store it per frame. */
>
> ---
>
> I do have one question. Should the "DMA queue overflow" message ever be
> printed?
> It seems to me that we have a 'no harm - no foul' situation.
Yes. This is a situation that should not happen.
If we stop the queue in mac80211, we expect that mac80211 does not queue more
frames from then on.
--
Greetings, Michael.
_______________________________________________
Bcm43xx-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev
