From: Robert Olsson <Robert.Olsson@data.slu.se>
To: Jesper Dangaard Brouer <hawk@diku.dk>
Cc: "David S. Miller" <davem@davemloft.net>,
dipankar@in.ibm.com, Robert Olsson <Robert.Olsson@data.slu.se>,
jens.laas@data.slu.se, hans.liss@its.uu.se,
linux-net@vger.kernel.org, linux-kernel@vger.kernel.org,
Eric Dumazet <dada1@cosmosbay.com>,
mike.stroyan@hp.com, Suresh Bhogavilli <sbhogavilli@verisign.com>
Subject: Re: Kernel panic: Route cache, RCU, possibly FIB trie.
Date: Thu, 23 Mar 2006 22:32:54 +0100 [thread overview]
Message-ID: <17443.5126.813111.61346@robur.slu.se> (raw)
In-Reply-To: <Pine.LNX.4.61.0603231536180.29788@ask.diku.dk>
Jesper Dangaard Brouer writes:
> > It is almost certainly the cause of your crashes, that code
> > is still extremely raw and that's why it is listed as "EXPERIMENTAL".
>
> It seems your are right :-) (and I'll take more care of using experimental
> code on production again). The machine, has now been running for 34 hours
> without crashing. The strange thing is that I'm running the same kernel
> on 30 other (similar) machines, which have not crashed. (I do suspect the
> specific traffic load pattern to influence this)
Sounds good... seems like Dave did the the correct analysis.
> BUT, I do think I have noticed another problem in the garbage collection
> code (route.c), that causes the garbage collector (almost) never to
> garbage collect.
We're trying to avoid most/all periodic GC in hi-flow systems and just to
do GC as new entries are created. We use the patch below to create et another
(unfortunately) /proc entry to better control this. ip_rt_gc_max_chain_length
it also decreases the threshhold for this from 8 to 4 for this.
Cheers.
--ro
--- linux-2.6.14.5/net/ipv4/route.c.orig 2006-01-02 14:24:00.000000000 +0100
+++ linux-2.6.14.5/net/ipv4/route.c 2006-01-02 15:26:29.000000000 +0100
@@ -126,6 +126,7 @@
static int ip_rt_error_cost = HZ;
static int ip_rt_error_burst = 5 * HZ;
static int ip_rt_gc_elasticity = 8;
+static int ip_rt_gc_max_chain_length = 4;
static int ip_rt_mtu_expires = 10 * 60 * HZ;
static int ip_rt_min_pmtu = 512 + 20 + 20;
static int ip_rt_min_advmss = 256;
@@ -977,7 +978,7 @@
* The second limit is less certain. At the moment it allows
* only 2 entries per bucket. We will see.
*/
- if (chain_length > ip_rt_gc_elasticity) {
+ if (chain_length > ip_rt_gc_max_chain_length) {
*candp = cand->u.rt_next;
rt_free(cand);
}
@@ -3017,6 +3018,14 @@
.proc_handler = &proc_dointvec,
},
{
+ .ctl_name = NET_IPV4_ROUTE_GC_MAX_CHAIN_LENGTH,
+ .procname = "gc_max_chain_length",
+ .data = &ip_rt_gc_max_chain_length,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec,
+ },
+ {
.ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
.procname = "mtu_expires",
.data = &ip_rt_mtu_expires,
--- linux-2.6.14.5/include/linux/sysctl.h.orig 2006-01-02 14:46:55.000000000 +0100
+++ linux-2.6.14.5/include/linux/sysctl.h 2006-01-02 14:47:01.000000000 +0100
@@ -375,6 +375,7 @@
NET_IPV4_ROUTE_MIN_ADVMSS=17,
NET_IPV4_ROUTE_SECRET_INTERVAL=18,
NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS=19,
+ NET_IPV4_ROUTE_GC_MAX_CHAIN_LENGTH=20,
};
enum
next prev parent reply other threads:[~2006-03-23 21:33 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2006-03-20 21:44 Jesper Dangaard Brouer
2006-03-20 22:09 ` Dipankar Sarma
2006-03-21 10:29 ` Jesper Dangaard Brouer
2006-03-21 10:37 ` David S. Miller
2006-03-21 14:51 ` Jesper Dangaard Brouer
2006-03-21 21:25 ` David S. Miller
2006-03-23 15:35 ` Jesper Dangaard Brouer
2006-03-23 15:44 ` Jesper Dangaard Brouer
2006-03-23 16:15 ` Eric Dumazet
2006-03-23 21:37 ` Jesper Dangaard Brouer
2006-03-24 6:11 ` Eric Dumazet
2006-03-24 10:34 ` Jesper Dangaard Brouer
2006-03-23 21:32 ` Robert Olsson [this message]
2006-03-21 13:28 ` Robert Olsson
2006-03-21 15:27 ` Jesper Dangaard Brouer
2006-03-21 15:37 ` Eric Dumazet
2006-03-21 15:43 ` Dipankar Sarma
2006-03-21 16:30 ` Eric Dumazet
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=17443.5126.813111.61346@robur.slu.se \
--to=robert.olsson@data.slu.se \
--cc=dada1@cosmosbay.com \
--cc=davem@davemloft.net \
--cc=dipankar@in.ibm.com \
--cc=hans.liss@its.uu.se \
--cc=hawk@diku.dk \
--cc=jens.laas@data.slu.se \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-net@vger.kernel.org \
--cc=mike.stroyan@hp.com \
--cc=sbhogavilli@verisign.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®