On Mon, Aug 07, 2006 at 03:34:30PM -0400, jamal wrote:
> On Mon, 2006-07-08 at 21:14 +0200, Edgar E. Iglesias wrote:
> 
> > > If yes, what is the likelihood they will sit there forever? I think
> > > perhaps some TX interupts will happen, no?
> > 
> > with jamal undefined, absolutely. With jamal defined, TX interrupts will 
> > come
> > but I couldnt find a way into e1000_prune_tx_ring unless fdesc met the 
> > conditions. Correct? 
> > 
> 
> Forgive me since i am still missing something ..
> 
> Observe that the same threshold used in two different ways:
> 
> 1) in tx path tx_ring->prunet is to check on when we should _start_ to
> prune.
> 2) on rx path tx_ring->prunet is to check when to _stop_ pruning.
> 

I can see two calls to e1000_prune_tx_ring with jamal _defined_.

1. tx path
+#ifdef jamal
+       {
+               int fdesc = E1000_DESC_UNUSED(tx_ring);
+               if (unlikely(fdesc < tx_ring->waket))
+                       e1000_prune_tx_ring(adapter,tx_ring);
+       }
+#endif

2. tx and rx path
+#ifdef jamal
+       spin_lock(&tx_ring->tx_lock);
+       {
+               int fdesc = E1000_DESC_UNUSED(tx_ring);
+               if (fdesc < tx_ring->prunet) {
+                       if (e1000_prune_tx_ring(adapter,tx_ring))
+                               cleaned = TRUE;
                }
        }
+       spin_unlock(&tx_ring->tx_lock);
+#else
+       if (e1000_prune_tx_ring(adapter,tx_ring))
+               cleaned = TRUE;
+#endif

Assume a ring of 64 entries, prunet of 16, waket of 8. Now host sends 40 skbs
and stops. tx-ring holds 40 skbs, has 24 free. TX interrupts hit you, you may
even be receiveing packets but I don't see how you enter prune_tx_ring without
more packets going out via hard_start_xmit? skb's will sit on the ring until 
more packets are sent from the quiet host.

As you can see, with jamal _undefined_ e1000_prune_tx_ring is called
unconditionally and I beleive things will work ok.

I am not familiar with this code nor the hw so I'm probably missing something
fundamental. 

Best regards
-- 
        Programmer
        Edgar E. Iglesias <[EMAIL PROTECTED]> 46.46.272.1946
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to