As already updated in bugzilla, I think the problem will show only when
a socket of type PF_PACKET receives small messages and in the failing
machines they use DHCP daemon which uses such sockets. We need a test to
prove that. Does anyone have such a test that he can share?
On Sun, 2008-04-27 at 0
Shirley Ma wrote:
We have seen skb_under_panic() in our test enviornment as well. It's
easy reproduce this with tcpdump on and off.
I have just tried iperf with tcpdump running, no crash, tried it with
both udp and tcp, both connected and datagram modes, please provide the
details on how to rep
Hello Or,
We have seen skb_under_panic() in our test enviornment as well. It's
easy reproduce this with tcpdump on and off.
thanks
Shirley
On Thu, 2008-04-24 at 14:11 +0300, Or Gerlitz wrote:
> > https://bugs.openfabrics.org/show_bug.cgi?id=989
> > --- Comment #13 from [EMAIL PROTECTED] 200
https://bugs.openfabrics.org/show_bug.cgi?id=989
--- Comment #13 from [EMAIL PROTECTED] 2008-04-23 13:23 ---
I think I found the problem and have a fix.
In ipoib_cm_handle_rx_wc(), if the byte_len is < SKB_TSHOLD, a new sk_buff
is allocated. The sk_buff's mac.raw is not initialized.