From: Rik van Riel <riel@redhat.com>
To: paulmck@linux.vnet.ibm.com
Cc: Paolo Bonzini <pbonzini@redhat.com>,
Ingo Molnar <mingo@kernel.org>,
Andy Lutomirski <luto@amacapital.net>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
X86 ML <x86@kernel.org>,
williams@redhat.com, Andrew Lutomirski <luto@kernel.org>,
fweisbec@redhat.com, Peter Zijlstra <peterz@infradead.org>,
Heiko Carstens <heiko.carstens@de.ibm.com>,
Thomas Gleixner <tglx@linutronix.de>,
Ingo Molnar <mingo@redhat.com>,
Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: question about RCU dynticks_nesting
Date: Mon, 04 May 2015 15:59:02 -0400 [thread overview]
Message-ID: <5547CF86.9060201@redhat.com> (raw)
In-Reply-To: <20150504193923.GX5381@linux.vnet.ibm.com>
On 05/04/2015 03:39 PM, Paul E. McKenney wrote:
> On Mon, May 04, 2015 at 03:00:44PM -0400, Rik van Riel wrote:
>> In case of the non-preemptible RCU, we could easily also
>> increase current->rcu_read_lock_nesting at the same time
>> we increase the preempt counter, and use that as the
>> indicator to test whether the cpu is in an extended
>> rcu quiescent state. That way there would be no extra
>> overhead at syscall entry or exit at all. The trick
>> would be getting the preempt count and the rcu read
>> lock nesting count in the same cache line for each task.
>
> But in non-preemptible RCU, we have PREEMPT=n, so there is no preempt
> counter in production kernels. Even if there was, we have to sample this
> on other CPUs, so the overhead of preempt_disable() and preempt_enable()
> would be where kernel entry/exit is, so I expect that this would be a
> net loss in overall performance.
CONFIG_PREEMPT_RCU seems to be independent of CONFIG_PREEMPT.
Not sure why, but they are :)
>> In case of the preemptible RCU scheme, we would have to
>> examine the per-task state (under the runqueue lock)
>> to get the current task info of all CPUs, and in
>> addition wait for the blkd_tasks list to empty out
>> when doing a synchronize_rcu().
>>
>> That does not appear to require special per-cpu
>> counters; examining the per-cpu rdp and the lists
>> inside it, with the rnp->lock held if doing any
>> list manipulation, looks like it would be enough.
>>
>> However, the current code is a lot more complicated
>> than that. Am I overlooking something obvious, Paul?
>> Maybe something non-obvious? :)
>
> Ummm... The need to maintain memory ordering when sampling task
> state from remote CPUs?
>
> Or am I completely confused about what you are suggesting?
>
> That said, are you chasing a real system-visible performance issue
> that you tracked to RCU's dyntick-idle system?
The goal is to reduce the syscall overhead of nohz_full.
Part of the overhead is in the vtime updates, part of it is
in the way RCU extended quiescent state is tracked.
--
All rights reversed
next prev parent reply other threads:[~2015-05-04 19:59 UTC|newest]
Thread overview: 83+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-04-30 21:23 [PATCH 0/3] reduce nohz_full syscall overhead by 10% riel
2015-04-30 21:23 ` [PATCH 1/3] reduce indentation in __acct_update_integrals riel
2015-04-30 21:23 ` [PATCH 2/3] remove local_irq_save from __acct_update_integrals riel
2015-04-30 21:23 ` [PATCH 3/3] context_tracking,x86: remove extraneous irq disable & enable from context tracking on syscall entry riel
2015-04-30 21:56 ` Andy Lutomirski
2015-05-01 6:40 ` Ingo Molnar
2015-05-01 15:20 ` Rik van Riel
2015-05-01 15:59 ` Ingo Molnar
2015-05-01 16:03 ` Andy Lutomirski
2015-05-01 16:21 ` Ingo Molnar
2015-05-01 16:26 ` Rik van Riel
2015-05-01 16:34 ` Ingo Molnar
2015-05-01 18:05 ` Rik van Riel
2015-05-01 18:40 ` Ingo Molnar
2015-05-01 19:11 ` Rik van Riel
2015-05-01 19:37 ` Andy Lutomirski
2015-05-02 5:27 ` Ingo Molnar
2015-05-02 18:27 ` Rik van Riel
2015-05-03 18:41 ` Andy Lutomirski
2015-05-07 10:35 ` Ingo Molnar
2015-05-04 9:26 ` Paolo Bonzini
2015-05-04 13:30 ` Rik van Riel
2015-05-04 14:06 ` Rik van Riel
2015-05-04 14:19 ` Rik van Riel
2015-05-04 15:59 ` question about RCU dynticks_nesting Rik van Riel
2015-05-04 18:39 ` Paul E. McKenney
2015-05-04 19:39 ` Rik van Riel
2015-05-04 20:02 ` Paul E. McKenney
2015-05-04 20:13 ` Rik van Riel
2015-05-04 20:38 ` Paul E. McKenney
2015-05-04 20:53 ` Rik van Riel
2015-05-05 5:54 ` Paul E. McKenney
2015-05-06 1:49 ` Mike Galbraith
2015-05-06 3:44 ` Mike Galbraith
2015-05-06 6:06 ` Paul E. McKenney
2015-05-06 6:52 ` Mike Galbraith
2015-05-06 7:01 ` Mike Galbraith
2015-05-07 0:59 ` Frederic Weisbecker
2015-05-07 15:44 ` Rik van Riel
2015-05-04 19:00 ` Rik van Riel
2015-05-04 19:39 ` Paul E. McKenney
2015-05-04 19:59 ` Rik van Riel [this message]
2015-05-04 20:40 ` Paul E. McKenney
2015-05-05 10:53 ` Peter Zijlstra
2015-05-05 12:34 ` Paul E. McKenney
2015-05-05 13:00 ` Peter Zijlstra
2015-05-05 18:35 ` Paul E. McKenney
2015-05-05 21:09 ` Rik van Riel
2015-05-06 5:41 ` Paul E. McKenney
2015-05-05 10:48 ` Peter Zijlstra
2015-05-05 10:51 ` Peter Zijlstra
2015-05-05 12:30 ` Paul E. McKenney
2015-05-02 4:06 ` [PATCH 3/3] context_tracking,x86: remove extraneous irq disable & enable from context tracking on syscall entry Mike Galbraith
2015-05-01 16:37 ` Ingo Molnar
2015-05-01 16:40 ` Rik van Riel
2015-05-01 16:45 ` Ingo Molnar
2015-05-01 16:54 ` Rik van Riel
2015-05-01 17:12 ` Ingo Molnar
2015-05-01 17:22 ` Rik van Riel
2015-05-01 17:59 ` Ingo Molnar
2015-05-01 16:22 ` Rik van Riel
2015-05-01 16:27 ` Ingo Molnar
2015-05-03 13:23 ` Mike Galbraith
2015-05-03 17:30 ` Rik van Riel
2015-05-03 18:24 ` Andy Lutomirski
2015-05-03 18:52 ` Rik van Riel
2015-05-07 10:48 ` Ingo Molnar
2015-05-07 12:18 ` Frederic Weisbecker
2015-05-07 12:29 ` Ingo Molnar
2015-05-07 15:47 ` Rik van Riel
2015-05-08 7:58 ` Ingo Molnar
2015-05-07 12:22 ` Andy Lutomirski
2015-05-07 12:44 ` Ingo Molnar
2015-05-07 12:49 ` Ingo Molnar
2015-05-08 6:17 ` Paul E. McKenney
2015-05-07 12:52 ` Andy Lutomirski
2015-05-07 15:08 ` Ingo Molnar
2015-05-07 17:47 ` Andy Lutomirski
2015-05-08 6:37 ` Ingo Molnar
2015-05-08 10:59 ` Andy Lutomirski
2015-05-08 11:27 ` Ingo Molnar
2015-05-08 12:56 ` Andy Lutomirski
2015-05-08 13:27 ` Ingo Molnar
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=5547CF86.9060201@redhat.com \
--to=riel@redhat.com \
--cc=fweisbec@redhat.com \
--cc=heiko.carstens@de.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@amacapital.net \
--cc=luto@kernel.org \
--cc=mingo@kernel.org \
--cc=mingo@redhat.com \
--cc=paulmck@linux.vnet.ibm.com \
--cc=pbonzini@redhat.com \
--cc=peterz@infradead.org \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
--cc=williams@redhat.com \
--cc=x86@kernel.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).