netdev
[Top] [All Lists]

Re: ksoftirqd uses 99% CPU triggered by network traffic (maybe RLT-8139

To: Stephen Hemminger <shemminger@xxxxxxxx>
Subject: Re: ksoftirqd uses 99% CPU triggered by network traffic (maybe RLT-8139 related)
From: Pasi Sjoholm <ptsjohol@xxxxxxxxx>
Date: Tue, 27 Jul 2004 00:28:04 +0300 (EEST)
Cc: Francois Romieu <romieu@xxxxxxxxxxxxx>, H?ctor Mart?n <hector@xxxxxxxxxxxxxx>, Linux-Kernel <linux-kernel@xxxxxxxxxxxxxxx>, <akpm@xxxxxxxx>, <netdev@xxxxxxxxxxx>, <brad@xxxxxxxxxx>, <Robert.Olsson@xxxxxxxxxxx>
In-reply-to: <20040726095505.2ddb3bec@xxxxxxxxxxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
On Mon, 26 Jul 2004, Stephen Hemminger wrote:

>>>> I haven't been able to reproduce this with normal www-browsing or 
>>>> ssh-connections but it's always reproducible when my eth0 is under heavy 
>>>> load.
>>> I guess it can be reproduced even if the binary (nvidia ?) module is never
>>> loaded after boot, right ?

>> After 24 hours of hard working I can answer yes to this question. 
>> Now I can reproduce this from 2 to 15 minutes with 2 cp-processes from 
>> samba->workstation->nfs, some software building with make -j 3, playing 
>> some mp3 via nfs to notice when kernel goes down.. =) and so on..

>> After that ksoftirqd takes almost all the cpu-time and the network is not 
>> working at all.

> Is network traffic still coming in? 

At least tcpdump doesn't say anything, I can only see only arp-packets 
which my computer (not other computer's arp-packets) has made but no 
response to those.

> or perhaps there is a network packet that causes some soft irq to go 
> into an infinite loop. The recent iptables bug with ip options would be 
> an example.

That would make sense but Robert Olsson had some knowledge of this and 
said that this issue is probably the same that they have discussed in OSL.

--
Pasi Sjöholm



<Prev in Thread] Current Thread [Next in Thread>