Re: Softnet/eepro100 driver conversion bug in 2.3.43pre8?

[Jeff Garzik]

[e-mail] wrote:
> 
> Hello!
> 
> > That _sounds_ reasonably, I agree, but eepro100 gets tx timeouts very
> > frequently if the access to "status" is not atomic.
> 
> Could you try to answer another, simpler, question?
> Namely, why is the question raised now?
> 
> Softnet did not introduce _anything_ new to this area.
> Even 2.3.15 did not this. If eepro100 fails now,
> it always (including 2.2) failed _exactly_ with the same frequency.

(off-list)

Some bug reports to linux-kernel, and my own experience testing 8139too,
seem to say otherwise...

I got _constant_ timeouts on UP until I updated the 8139too interrupt
handler from

	if (tx buffer not full) {
		tx_full=0;
		netif_wake_queue(dev);
	}

to

	if (tx buffer not full) {
		tx_full = 0;
	}
	if (tx_full)
		netif_wake_queue(dev);
	else
		netif_stop_queue(dev);

Regards,

	Jeff

-- 
Jeff Garzik         | "Vegetarian" is the Indian word
Building 1024       | for 'lousy hunter.'
MandrakeSoft, Inc.  |