On Thu, 17 Mar 2011, Eric Dumazet wrote:
Le jeudi 17 mars 2011 à 12:45 +0100, Eric Dumazet a écrit :
Le jeudi 17 mars 2011 à 12:36 +0100, Jesper Dangaard Brouer a écrit :
Hi Eric,
How critial is this bug fix?
Should I apply this on my stable production kernels?
(I'm preparing a 2.6.38 kernel for prod usage, eventhougt its just been
released, because I want your SFQ fixes...)
I would say the race is there, but probability must be very small, and
might need malicious iptables rules (with RETURN targets)
Actually do have a large number of return targets.
In the filter forward chain.
Especially in routers, where OUTPUT path is taken from softirq handler
anyway ;)
So dont worry at all, consider this as a cleanup :)
Almost forgot to mention your kernels probably have :
CONFIG_PREEMPT_NONE=y
# CONFIG_PREEMPT is not set
So you can take it easy ;)
I have actually enabled PREEMPTion, am I in trouble then?
I have modified the .config I got from you, quite a lot, during my
attempts to find the CCISS/HPSA problem.
Which reminds me that I needed to enable, the follwing config options,
(which your config didn't have):
CONFIG_PCIEASPM=y
CONFIG_X86_X2APIC=y
In order to get rid of a PCI warning:
NMI: PCI system error (SERR) for reason b1 on CPU 0.
Dazed and confused, but trying to continue
The system worked fine, and you might not see this issue on your G6
servers, as its most likely BIOS related (my server is a HP DL370 G6).
FYI: Operations have deployed the kernel on prod server this morning.
Cheers,
Jesper Brouer
--
-------------------------------------------------------------------
MSc. Master of Computer Science
Dept. of Computer Science, University of Copenhagen
Author of http://www.adsl-optimizer.dk
-------------------------------------------------------------------
