Re: [PATCH 1/2] e1000: fix netpoll with NAPI

[Jeff Moyer <jmoyer@redhat.com>]

==> Regarding Re: [PATCH 1/2] e1000: fix netpoll with NAPI; Auke Kok <auke-jan.h.kok@intel.com> adds:

auke-jan.h.kok> Hi,

auke-jan.h.kok> we're not too happy with this as it puts a branch right in
auke-jan.h.kok> the regular receive path. We haven't ran the numbers on it
auke-jan.h.kok> yet but it is likely that this will lower performance
auke-jan.h.kok> significantly during normal receives for something that is
auke-jan.h.kok> not common use.

auke-jan.h.kok> Attached below a (revised) patch that adds proper locking
auke-jan.h.kok> around the rx_clean to prevent the race.

That patch locks around the tx clean routine.  As such, it doesn't prevent
the problem.

> +	disable_irq(adapter->pdev->irq);
> +	if (likely(netif_rx_schedule_prep(&adapter->polling_netdev[0]))) {
> +		if (spin_trylock(&adapter->tx_queue_lock)) {
> +			e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
> +			spin_unlock(&adapter->tx_queue_lock);
> +		}
> +		adapter->clean_rx(adapter, adapter->rx_ring,
> +				&budget, netdev->weight);
> +		clear_bit(__LINK_STATE_RX_SCHED,
> +				&adapter->polling_netdev[0].state);

-Jeff