From: Eric Dumazet <dada1@cosmosbay.com>
To: dipankar@in.ibm.com
Cc: Lee Revell <rlrevell@joe-job.com>,
paulmck@us.ibm.com, Ingo Molnar <mingo@elte.hu>,
linux-kernel <linux-kernel@vger.kernel.org>,
Linus Torvalds <torvalds@osdl.org>
Subject: Re: RCU latency regression in 2.6.16-rc1
Date: Sat, 28 Jan 2006 20:52:02 +0100 [thread overview]
Message-ID: <43DBCB62.7030308@cosmosbay.com> (raw)
In-Reply-To: <20060128193412.GH5633@in.ibm.com>
Dipankar Sarma a écrit :
> On Sat, Jan 28, 2006 at 01:51:23PM -0500, Lee Revell wrote:
>> On Sat, 2006-01-28 at 13:00 -0500, Lee Revell wrote:
>>> OK, now we are making progress.
>> I spoke too soon, it's not fixed:
>>
>> preemption latency trace v1.1.5 on 2.6.16-rc1
>> --------------------------------------------------------------------
>> latency: 4183 us, #3676/3676, CPU#0 | (M:rt VP:0, KP:0, SP:0 HP:0)
>> -----------------
>> evolutio-2877 0d.s. 97us : local_bh_enable (rt_run_flush)
>> evolutio-2877 0d.s. 98us : local_bh_enable (rt_run_flush)
>> evolutio-2877 0d.s. 99us : local_bh_enable (rt_run_flush)
>> evolutio-2877 0d.s. 100us : local_bh_enable (rt_run_flush)
>> evolutio-2877 0d.s. 101us : local_bh_enable (rt_run_flush)
>>
>> [ etc ]
>>
>> evolutio-2877 0d.s. 4079us : local_bh_enable (rt_run_flush)
>> evolutio-2877 0d.s. 4080us : local_bh_enable (rt_run_flush)
>
> I am not sure if I am interpreting the latency trace right,
> but it seems that there is a difference between the problem
> you were seeing earlier and now.
>
> In one of your earlier traces, I saw -
>
> <idle>-0 0d.s. 182us : dst_destroy (dst_rcu_free)
> <idle>-0 0d.s. 183us : ipv4_dst_destroy (dst_destroy)
>
> [ etc - zillions of dst_rcu_free()s deleted ]
>
> <idle>-0 0d.s. 13403us : dst_rcu_free (__rcu_process_callbacks)
> <idle>-0 0d.s. 13403us : dst_destroy (dst_rcu_free)
>
> This points to latency increase caused by lots and lots of
> RCU callbacks doing dst_rcu_free(). Do you still see those ?
>
> Your new trace shows that we are held up in in rt_run_flush().
> I guess we need to investigate why we spend so much time in rt_run_flush(),
> because of a big route table or the lock acquisitions.
Some machines have millions of entries in their route cache.
I suspect we cannot queue all them (or only hash heads as your previous patch)
by RCU. Latencies and/or OOM can occur.
What can be done is :
in rt_run_flush(), allocate a new empty hash table, and exchange the hash tables.
Then wait a quiescent/grace RCU period (may be the exact term is not this one,
sorry, I'm not RCU expert)
Then free all the entries from the old hash table (direclty of course, no need
for RCU grace period), and free the hash table.
As the hash table can be huge, we might need allocate it at boot time, just in
case a flush is needed (it usually is :) ). If we choose dynamic allocation
and this allocation fails, then fallback to what is done today.
Eric
next prev parent reply other threads:[~2006-01-28 19:52 UTC|newest]
Thread overview: 35+ messages / expand[flat|nested] mbox.gz Atom feed top
2006-01-24 7:52 RCU latency regression in 2.6.16-rc1 Lee Revell
2006-01-24 7:56 ` Ingo Molnar
2006-01-24 7:58 ` Lee Revell
2006-01-24 8:01 ` Ingo Molnar
2006-01-24 8:03 ` Lee Revell
2006-01-24 8:11 ` Ingo Molnar
2006-01-24 8:07 ` Lee Revell
2006-01-24 8:13 ` Ingo Molnar
2006-01-24 8:15 ` Lee Revell
2006-01-24 9:17 ` Paul E. McKenney
2006-01-24 9:23 ` Ingo Molnar
2006-01-24 9:44 ` Lee Revell
2006-01-24 16:28 ` Dipankar Sarma
2006-01-24 21:38 ` Dipankar Sarma
2006-01-25 21:28 ` Lee Revell
2006-01-25 22:56 ` Ingo Molnar
2006-01-25 23:13 ` Lee Revell
2006-01-26 19:18 ` Paul E. McKenney
2006-01-27 18:55 ` Lee Revell
2006-01-28 17:03 ` Dipankar Sarma
2006-01-28 18:00 ` Lee Revell
2006-01-28 18:51 ` Lee Revell
2006-01-28 19:34 ` Dipankar Sarma
2006-01-28 19:46 ` Lee Revell
2006-01-28 19:52 ` Eric Dumazet [this message]
2006-01-29 7:38 ` Lee Revell
2006-01-29 7:51 ` Ingo Molnar
2006-01-29 8:21 ` Lee Revell
2006-01-30 4:36 ` Paul E. McKenney
2006-01-30 4:55 ` Eric Dumazet
2006-01-30 5:11 ` Paul E. McKenney
2006-01-30 5:52 ` David S. Miller
2006-01-30 10:00 ` Paul E. McKenney
2006-02-12 0:45 ` Lee Revell
2006-01-24 16:57 ` Dipankar Sarma
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=43DBCB62.7030308@cosmosbay.com \
--to=dada1@cosmosbay.com \
--cc=dipankar@in.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=paulmck@us.ibm.com \
--cc=rlrevell@joe-job.com \
--cc=torvalds@osdl.org \
/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;
as well as URLs for NNTP newsgroup(s).