Re: [patch sungem] improved locking

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Wed, 2006-12-13 at 14:12 +1100, Benjamin Herrenschmidt wrote:
> Been hitting a raw throughput in both directions plus a few other things
> on a dual G5 and the driver didn't crash :-)
> 
> I'm seeing a problem though but I'm not sure it's related to your patch,
> I'll have to test without it.
> 
> Basically, if I use a slightly modified versio of tridge's socklib (raw
> xput test, basically, a tcp connection with one side pushing as fast as
> it can a known pattern and the other one just receiving and verifying
> the data integrity), it works fine when running only one side (either rx
> or tx, doesn't matter).
> 
> But if I start it both ways (that is both a receiver and a sender on the
> GMAC box) and the other end is a tg3 (quad g5), then I'm getting a lot
> of eth0: RX MAC fifo overflow smac[02045822 but there are other numbers
> here every now and then] errors on the sungem side.
> 
> David, could that be the pause stuff not working properly ?
> 
> The link is gigabit and the tg3 side doesn't complain about anything.

I've verified that the problem isn't related to Eric's patch and is
there without it... I suppose I just never noticed it before :-)

So far, the patch looks solid. I'll let it run overnight and will holler
if things went wrong tomorrow.

Cheers,
Ben.