problem with debug.mpsafenet (was: Problems with ANA-6944A + 5.3-BETA + IRQ sharing?)

[John Baldwin]

On Monday 27 September 2004 03:23 pm, Oliver Lehmann wrote:
> Ok,
>
> I tried now 3 FreeBSD versions.
>
> Setup:
> 	PI-166MHz, 64MB RAM
> 	de0             irq9 @ANA-6944A
> 	de1 10.0.0.60	irq9 @ANA-6944A
> 	de2 10.0.1.60	irq9 @ANA-6944A
> 	de3             irq9 @ANA-6944A
>         xl0             irq12
>         xl1             irq10
> 	ep0             irq5
> 	ep1             irq7
> 	/mnt/files is a nfs mount from 10.0.0.21
> 	3 * ssh sessions from 10.0.1.51 to 10.0.1.60
> 	2 * dd if=/mnt/files/dill_dd of=/dev/null
> 	1 * ping 10.0.1.51 running
>
>
> 5.3-BETA6 (GENERIC Kernel):
>
> Everything looked fine (exept same slowness as 5.2.1) for a while (~9min).
> 400736256 bytes transferred in 546.111599 secs (733799 bytes/sec)
> 414572544 bytes transferred in 565.398805 secs (733239 bytes/sec)
>
> But as I started a ping -f from 10.0.1.51 to 10.0.0.60. The ping times
> from 10.0.1.60 to 10.0.1.51 went up, and didn't renormalized after I break
> the ping -f
>
> 64 bytes from 10.0.1.51: icmp_seq=355 ttl=64 time=4.039 ms
> 64 bytes from 10.0.1.51: icmp_seq=356 ttl=64 time=4.801 ms
> 64 bytes from 10.0.1.51: icmp_seq=357 ttl=64 time=6.702 ms
> 64 bytes from 10.0.1.51: icmp_seq=358 ttl=64 time=9.290 ms
> 64 bytes from 10.0.1.51: icmp_seq=359 ttl=64 time=10.543 ms
> 64 bytes from 10.0.1.51: icmp_seq=360 ttl=64 time=9.989 ms
> 64 bytes from 10.0.1.51: icmp_seq=361 ttl=64 time=25.966 ms
> 64 bytes from 10.0.1.51: icmp_seq=362 ttl=64 time=25.262 ms
> 64 bytes from 10.0.1.51: icmp_seq=363 ttl=64 time=27.746 ms
> 64 bytes from 10.0.1.51: icmp_seq=364 ttl=64 time=30.859 ms
> 64 bytes from 10.0.1.51: icmp_seq=365 ttl=64 time=31.507 ms
> 64 bytes from 10.0.1.51: icmp_seq=366 ttl=64 time=35.783 ms
> 64 bytes from 10.0.1.51: icmp_seq=367 ttl=64 time=29295.367 ms
> 64 bytes from 10.0.1.51: icmp_seq=368 ttl=64 time=28288.204 ms
> 64 bytes from 10.0.1.51: icmp_seq=369 ttl=64 time=27279.969 ms
> 64 bytes from 10.0.1.51: icmp_seq=370 ttl=64 time=26274.269 ms
> 64 bytes from 10.0.1.51: icmp_seq=371 ttl=64 time=25267.441 ms
> 64 bytes from 10.0.1.51: icmp_seq=372 ttl=64 time=24260.161 ms
> ....
>
>
> After that, I added debug.mpsafenet=0 to /boot/loader.conf and rebooted
> the system. I did the same as above. (2dd's, one ping from 10.0.1.60 to
> 10.0.1.51, and one ping -f from 10.0.1.51 to 10.0.1.60.)
> The ping times from 10.0.1.60 -> 10.0.1.51 went a bit up (~0.3ms -> 0.6 -
> 1ms), but nothing else happend. dd finished w/o errors, no strange things
> happend. (running now for ~20 mins w/o problems)
>
>
> For me, It looks like there is a problem with mpsafenet!
> Any help/patches to test/information requests/.... appreciated.

The problem appears to be in the de(4) driver and there are numerous reports 
of that specific driver having problems.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org