On Sat, Mar 24, 2001 at 06:22:27PM +1100, Andrew Morton wrote:
> I tried to reproduce this memory leak and failed. Added some
> code to netif_rx() to corrupt incoming packets:
...
> Can you suggest any other way of reproducing this? Have I missed
> something?
>
> Do you think your broken ethernet switch was corrupting
> data at layer2 (ethernet checksum will catch it) or at
> layer 3 (IP checksums)?
I have no idea, I didn't notice anything at first, except for a network
slowdown which the others on my floor were seeing as well and everyone
was assuming simple network congestion. My machine went OOM about twice
in that week, which triggered me to started looking around. At first I
was suspecting some leak in Coda (my neck of the woods), but noticed in
slabinfo that a lot of size-2048 slabs were allocated.
Patched ikd into my system and ran it for a day to find out when it
happened and where. I looked at the 3c59x driver and could not find
anything wrong with it. The next day everything was back to normal.
I do have a SMP system, so it could be an obscure race. The thing that
might have been triggering it were the backups that consisted of about
121 tcp connection with a total datatransfer of about 1GB. But it could
also be UDP fragments, Coda tends to send 4KB UDP packets when
revalidating it's caches.
Jan
|