Re: poor network performace of mec on R10000 O2

[Michael <macallan%netbsd.org@localhost>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Sep 25, 2009, at 2:45 PM, Mark Kirby wrote:

> I have an O2 running netbsd-5 branch with the following cpu
>
> cpu0 at mainbus0: MIPS R10000 CPU (0x934) Rev. 3.4 with built-in FPU  
> Rev. 0.0
>
> When using the mec driver it frequently stalls, or at least the kb/s  
> counter drops to 0, when doing any heavy lifting.
>
> It just took me 1hr 50 mins to copy a gziped src archive from the  
> machine next to it. That was just over 700mb, the little kb/s  
> counter would shoot up to over 100kb/s then drop to 0 repeat until  
> done. Its the same with ftp and building with sources on an nfs share.

I had problems like this with 4.x, but tsutsui-san fixed it a year ago  
or so. My O2 has an R5k though.

> does anyone have any ideas? i spoke to jmcneill about it and he  
> thinks its the fact i have a R10K cpu.

Probably, I'm not sure we have the hardware R10k workaround enabled  
and I can't test for lack of R10k hardware.

> I also see crime cpu error 4 messages occasionally
>
> Sep 25 15:29:21 dribble /netbsd: crime: cpu error 4 at address  
> 67548016
> Sep 25 16:46:19 dribble /netbsd: crime: cpu error 4 at address  
> 82680212
> Sep 25 17:02:54 dribble /netbsd: crime: cpu error 4 at address  
> 81361944

The addresses look bogus to me, might be the R10k's speculative  
accesses which we would need the above mentioned workaround for.

have fun
Michael

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)

iQEVAwUBSr0S4MpnzkX8Yg2nAQIAcAf/bn2HmEos1y9+VdZrPwWfRoi0nYdM65Kh
POtSrLGlNi3zGgKTav5kj6jdhAsvGb1GkHMITFnj3jFDQAwlXIs8/S1oOumsIXjL
YQmbM6pG5aXUQGCx3PPxf5uumAMhvUB3vTpi7Bs919JnQIhswH3UuLiLZeNDvgaI
kUolXqOYH8xoUt5CGr837rOjQI3KaAO3iAAZk8KI60dS/jKSt+UoUama8HIPThvg
+EtRKFbhOVWISL7uvJawh/FpvVo2RH64h6JWOwqYIMcd31qFtZRlW+ArMW77LzuV
iDOZc1HZdxXcALmQSmMq3B8LH6eldp7nLNktM16Kwh2qhz56X/mn2A==
=YuyA
-----END PGP SIGNATURE-----