macallan@NetBSD.org wrote:

> > With a quick glance your extintr.c changes seems okay,
> > but I'll check how re(4) works on sources before the change.
> 
> Likely the same - I suspect the interrupt re-enable code. Looks like 
> either it's missing interrupts that fired while unblocking, could be a 
> race there, or it didn't run the appropriate handlers for pending 
> interrupts when lowering spl.

Hmm, NetBSD-3.1_RC4 (well, I have its bootable CD) kernel
with the latest re(4) driver (with some modification to
make it compile on netbsd-3) still has the similar problem.

ping(8) against macppc host from other host shows:
---
64 bytes from 192.168.20.4: icmp_seq=60 ttl=255 time=18.354 ms
64 bytes from 192.168.20.4: icmp_seq=61 ttl=255 time=37.708 ms
64 bytes from 192.168.20.4: icmp_seq=62 ttl=255 time=19.671 ms
64 bytes from 192.168.20.4: icmp_seq=63 ttl=255 time=11443.316 ms
64 bytes from 192.168.20.4: icmp_seq=64 ttl=255 time=10443.413 ms
64 bytes from 192.168.20.4: icmp_seq=65 ttl=255 time=9443.560 ms
64 bytes from 192.168.20.4: icmp_seq=66 ttl=255 time=8444.578 ms
64 bytes from 192.168.20.4: icmp_seq=67 ttl=255 time=7444.597 ms
64 bytes from 192.168.20.4: icmp_seq=68 ttl=255 time=6444.612 ms
64 bytes from 192.168.20.4: icmp_seq=69 ttl=255 time=5444.629 ms
64 bytes from 192.168.20.4: icmp_seq=70 ttl=255 time=4444.644 ms
64 bytes from 192.168.20.4: icmp_seq=71 ttl=255 time=3444.660 ms
64 bytes from 192.168.20.4: icmp_seq=72 ttl=255 time=2444.676 ms
64 bytes from 192.168.20.4: icmp_seq=73 ttl=255 time=1444.692 ms
64 bytes from 192.168.20.4: icmp_seq=74 ttl=255 time=444.707 ms
64 bytes from 192.168.20.4: icmp_seq=75 ttl=255 time=41.430 ms
64 bytes from 192.168.20.4: icmp_seq=76 ttl=255 time=36.730 ms
---

I'll check how it goes on prep next.
---
Izumi Tsutsui