linux-kernel - Re: [PATCH 1/2] rcu: Don't chase unnecessary quiescent states after extended grace periods

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 24 Nov 2010 21:45:08 +0100
From:	Frederic Weisbecker <fweisbec@...il.com>
To:	paulmck@...ux.vnet.ibm.com
Cc:	LKML <linux-kernel@...r.kernel.org>,
	Lai Jiangshan <laijs@...fujitsu.com>,
	Ingo Molnar <mingo@...e.hu>,
	Thomas Gleixner <tglx@...utronix.de>,
	Peter Zijlstra <a.p.zijlstra@...llo.nl>,
	Steven Rostedt <rostedt@...dmis.org>
Subject: Re: [PATCH 1/2] rcu: Don't chase unnecessary quiescent states after
 extended grace periods

2010/11/24 Paul E. McKenney <paulmck@...ux.vnet.ibm.com>:
> I take it back.  I queued the following -- your code, but updated
> comment and commit message.  Please let me know if I missed anything.
>
>                                                        Thanx, Paul
>
> ------------------------------------------------------------------------
>
> commit 1d9d947bb882371a0877ba05207a0b996dcb38ee
> Author: Frederic Weisbecker <fweisbec@...il.com>
> Date:   Wed Nov 24 01:31:12 2010 +0100
>
>    rcu: Don't chase unnecessary quiescent states after extended grace periods
>
>    When a CPU is in an extended quiescent state, including offline and
>    dyntick-idle mode, other CPUs will detect the extended quiescent state
>    and respond to the the current grace period on that CPU's behalf.
>    However, the locking design prevents those other CPUs from updating
>    the first CPU's rcu_data state.
>
>    Therefore, when this CPU exits its extended quiescent state, it must
>    update its rcu_data state.  Because such a CPU will usually check for
>    the completion of a prior grace period before checking for the start of a
>    new grace period, the rcu_data ->completed field will be updated before
>    the rcu_data ->gpnum field.  This means that if RCU is currently idle,
>    the CPU will usually enter __note_new_gpnum() with ->completed set to
>    the current grace-period number, but with ->gpnum set to some long-ago
>    grace period number.  Unfortunately, __note_new_gpnum() will then insist
>    that the current CPU needlessly check for a new quiescent state.  This
>    checking can result in this CPU needlessly taking several scheduling-clock
>    interrupts.


So I'm all ok for the commit and the comments updated. But just a doubt about
the about sentence.

The effect seems more that there will be one extra softirq. But not an
extra tick
because before sleeping, the CPU will check rcu_needs_cpu() which
doesn't check for
the need of noting a quiescent state, IIRC...

And I think the softirq will be only raised on the next tick.

Hm?

>
>    This bug is harmless in most cases, but is a problem for users concerned
>    with OS jitter for HPC applications or concerned with battery lifetime
>    for portable SMP embedded devices.  This commit therefore makes the
>    test in __note_new_gpnum() check for this situation and avoid the needless
>    quiescent-state checks.
>
>    Signed-off-by: Frederic Weisbecker <fweisbec@...il.com>
>    Cc: Paul E. McKenney <paulmck@...ux.vnet.ibm.com>
>    Cc: Lai Jiangshan <laijs@...fujitsu.com>
>    Cc: Ingo Molnar <mingo@...e.hu>
>    Cc: Thomas Gleixner <tglx@...utronix.de>
>    Cc: Peter Zijlstra <a.p.zijlstra@...llo.nl>
>    Cc: Steven Rostedt <rostedt@...dmis.org>
>    Signed-off-by: Paul E. McKenney <paulmck@...ux.vnet.ibm.com>
>
> diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> index 5df948f..76cd5d2 100644
> --- a/kernel/rcutree.c
> +++ b/kernel/rcutree.c
> @@ -616,8 +616,20 @@ static void __init check_cpu_stall_init(void)
>  static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
>  {
>        if (rdp->gpnum != rnp->gpnum) {
> -               rdp->qs_pending = 1;
> -               rdp->passed_quiesc = 0;
> +               /*
> +                * Because RCU checks for the prior grace period ending
> +                * before checking for a new grace period starting, it
> +                * is possible for rdp->gpnum to be set to the old grace
> +                * period and rdp->completed to be set to the new grace
> +                * period.  So don't bother checking for a quiescent state
> +                * for the rnp->gpnum grace period unless it really is
> +                * waiting for this CPU.
> +                */
> +               if (rdp->completed != rnp->gpnum) {
> +                       rdp->qs_pending = 1;
> +                       rdp->passed_quiesc = 0;
> +               }
> +
>                rdp->gpnum = rnp->gpnum;
>        }
>  }
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/