You would think it would be easier to maintain a separate 2.2.x and
2.4.x version of acenic.c at this point :)
Jes Sorensen wrote:
> @@ -2307,14 +2337,10 @@
> u32 idx, flagsize;
>
> /*
> - * ARGH, there is just no pretty way to do this
> + * This only happens with pre-softnet, ie. 2.2.x kernels.
> */
> -#if (LINUX_VERSION_CODE < 0x02032b)
> - if (test_and_set_bit(0, &dev->tbusy))
> + if (early_stop_netif_stop_queue(dev))
> return 1;
> -#else
> - netif_stop_queue(dev);
> -#endif
>
> idx = ap->tx_prd;
>
> @@ -2358,7 +2384,8 @@
> */
> mod_timer(&ap->timer, jiffies + (3 * HZ));
>
> - /* The following check will fix a race between the interrupt
> + /*
> + * The following check will fix a race between the interrupt
> * handler increasing the tx_ret_csm and testing for tx_full
> * and this tx routine's testing the tx_ret_csm and setting
> * the tx_full; note that this fix makes assumptions on the
> @@ -2369,13 +2396,17 @@
> if (((idx + 2) % TX_RING_ENTRIES != ap->tx_ret_csm)
> && xchg(&ap->tx_full, 0)) {
> del_timer(&ap->timer);
> + /*
> + * We may not need this one in the post softnet era
> + * in this case this can be changed to a
> + * early_stop_netif_wake_queue(dev);
> + */
> netif_wake_queue(dev);
nope, you don't need that. if you are start_xmit, and you didn't stop
the queue yourself, there's no need to start it, and definitely no need
to wake...
you can probably remove 'tx_full' as well, as it is usually redundant to
tbusy/netif_{start,stop}_queue.
Jeff
-- Jeff Garzik | Building 1024 | The chief enemy of creativity is "good" sense MandrakeSoft | -- Picasso - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Nov 15 2000 - 21:00:24 EST