maybe there is another reason for this behaviour. When profiling the kernel
on our main gateway (10Mb/s <-> 10Mb/s) ip_rt_run_bh() was one of the top
most called functions. We tracked it down to the point, that it's the
inlined rt_kick_free_queue(), which is called every time ip_rt_run_bh()
is called as soon as RT_BH_FREE is set in ip_rt_bh_mask. And this flag
is never reset. Below is a fix, which solved the problems on our gateway.
If this patch is correct, it should be part of 2.0.31.
Thomas.
--- linux/net/ipv4/route.c.orig Wed May 7 14:28:56 1997
+++ linux/net/ipv4/route.c Wed May 7 14:30:22 1997
@@ -936,6 +936,7 @@
}
rtp = &rt->rt_next;
}
+ ip_rt_bh_mask &= ~RT_BH_FREE;
}
void ip_rt_run_bh()
--
See, you not only have to be a good coder to create a system like Linux,
you have to be a sneaky bastard too ;-)
[Linus Torvalds in <4rikft$7g5@linux.cs.Helsinki.FI>]