On Fri, 10 Sep 2004 13:30:55 +1000
Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, Sep 09, 2004 at 11:12:33AM +0000, Alex Riesen wrote:
> > The box froze after being left for some time (some 10 hours) unattended.
> > The only thing in I could find in logs was:
> >
> > Sep 8 22:30:18 steel kernel: KERNEL: assertion
> > ((int)tcp_get_pcount(&tp->lost_out) >= 0) failed at net/ipv4/tcp_input.c
> > (2422)
> > Sep 8 22:30:18 steel kernel: Leak l=4294967295 4
>
> Looks like the factor isn't set early enough. Can you please check
> that you had the changeset titled
>
> [TCP]: Make sure SKB tso factor is setup early enough.
>
> from davem?
>
> If you did, then please apply the following patch and tell us what
> the resulting messages.
Herbert did you see my division fix I made today for
tso_factor calculation? I was dividing by the TSO mss
instead of the normal one :-)
I think that is the cause of these problems. It was
definitely the cause of a BUG() trap hit in tcp_transmit_skb()
that someone else reported in the past day.
|