linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Rafael Aquini <aquini@redhat.com>
To: Rik van Riel <riel@redhat.com>
Cc: linux-kernel@vger.kernel.org, walken@google.com,
	eric.dumazet@gmail.com, lwoodman@redhat.com, jeremy@goop.org,
	Jan Beulich <JBeulich@novell.com>,
	knoel@redhat.com, chegu_vinod@hp.com,
	raghavendra.kt@linux.vnet.ibm.com, mingo@redhat.com
Subject: Re: [PATCH 3/5] x86,smp: auto tune spinlock backoff delay factor
Date: Thu, 10 Jan 2013 01:13:39 -0200	[thread overview]
Message-ID: <20130110031339.GC1636@x61.redhat.com> (raw)
In-Reply-To: <20130108173029.305d99c0@annuminas.surriel.com>

On Tue, Jan 08, 2013 at 05:30:29PM -0500, Rik van Riel wrote:
> Many spinlocks are embedded in data structures; having many CPUs
> pounce on the cache line the lock is in will slow down the lock
> holder, and can cause system performance to fall off a cliff.
> 
> The paper "Non-scalable locks are dangerous" is a good reference:
> 
> 	http://pdos.csail.mit.edu/papers/linux:lock.pdf
> 
> In the Linux kernel, spinlocks are optimized for the case of
> there not being contention. After all, if there is contention,
> the data structure can be improved to reduce or eliminate
> lock contention.
> 
> Likewise, the spinlock API should remain simple, and the
> common case of the lock not being contended should remain
> as fast as ever.
> 
> However, since spinlock contention should be fairly uncommon,
> we can add functionality into the spinlock slow path that keeps
> system performance from falling off a cliff when there is lock
> contention.
> 
> Proportional delay in ticket locks is delaying the time between
> checking the ticket based on a delay factor, and the number of
> CPUs ahead of us in the queue for this lock. Checking the lock
> less often allows the lock holder to continue running, resulting
> in better throughput and preventing performance from dropping
> off a cliff.
> 
> Proportional spinlock delay with a high delay factor works well
> when there is lots contention on a lock. Likewise, a smaller
> delay factor works well when a lock is lightly contended.
> 
> Making the code auto-tune the delay factor results in a system
> that performs well with both light and heavy lock contention.
> 
> Signed-off-by: Rik van Riel <riel@redhat.com>
> ---
> v3: use fixed-point math for the delay calculations, suggested by Michel Lespinasse
>

Acked-by: Rafael Aquini <aquini@redhat.com>

 
>  arch/x86/kernel/smp.c |   43 +++++++++++++++++++++++++++++++++++++++----
>  1 files changed, 39 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/x86/kernel/smp.c b/arch/x86/kernel/smp.c
> index aa743e9..05f828b 100644
> --- a/arch/x86/kernel/smp.c
> +++ b/arch/x86/kernel/smp.c
> @@ -113,13 +113,34 @@ static atomic_t stopping_cpu = ATOMIC_INIT(-1);
>  static bool smp_no_nmi_ipi = false;
>  
>  /*
> - * Wait on a congested ticket spinlock.
> + * Wait on a congested ticket spinlock. Many spinlocks are embedded in
> + * data structures; having many CPUs pounce on the cache line with the
> + * spinlock simultaneously can slow down the lock holder, and the system
> + * as a whole.
> + *
> + * To prevent total performance collapse in case of bad spinlock contention,
> + * perform proportional backoff. The per-cpu value of delay is automatically
> + * tuned to limit the number of times spinning CPUs poll the lock before
> + * obtaining it. This limits the amount of cross-CPU traffic required to obtain
> + * a spinlock, and keeps system performance from dropping off a cliff.
> + *
> + * There is a tradeoff. If we poll too often, the whole system is slowed
> + * down. If we sleep too long, the lock will go unused for a period of
> + * time. The solution is to go for a fast spin if we are at the head of
> + * the queue, to slowly increase the delay if we sleep for too short a
> + * time, and to decrease the delay if we slept for too long.
>   */
> +#define DELAY_SHIFT 8
> +#define DELAY_FIXED_1 (1<<DELAY_SHIFT)
> +#define MIN_SPINLOCK_DELAY (1 * DELAY_FIXED_1)
> +#define MAX_SPINLOCK_DELAY (16000 * DELAY_FIXED_1)
> +DEFINE_PER_CPU(unsigned, spinlock_delay) = { MIN_SPINLOCK_DELAY };
>  void ticket_spin_lock_wait(arch_spinlock_t *lock, struct __raw_tickets inc)
>  {
>  	__ticket_t head = inc.head, ticket = inc.tail;
>  	__ticket_t waiters_ahead;
> -	unsigned loops;
> +	unsigned delay = __this_cpu_read(spinlock_delay);
> +	unsigned loops = 1;
>  
>  	for (;;) {
>  		waiters_ahead = ticket - head - 1;
> @@ -133,14 +154,28 @@ void ticket_spin_lock_wait(arch_spinlock_t *lock, struct __raw_tickets inc)
>  			} while (ACCESS_ONCE(lock->tickets.head) != ticket);
>  			break;
>  		}
> -		loops = 50 * waiters_ahead;
> +
> +		/* Aggressively increase delay, to minimize lock accesses. */
> +		if (delay < MAX_SPINLOCK_DELAY)
> +			delay += DELAY_FIXED_1 / 7;
> +
> +		loops = (delay * waiters_ahead) >> DELAY_SHIFT;
>  		while (loops--)
>  			cpu_relax();
>  
>  		head = ACCESS_ONCE(lock->tickets.head);
> -		if (head == ticket)
> +		if (head == ticket) {
> +			/*
> +			 * We overslept, and do not know by how.
> +			 * Exponentially decay the value of delay,
> +			 * to get it back to a good value quickly.
> +			 */
> +			if (delay >= 2 * DELAY_FIXED_1)
> +				delay -= max(delay/32, DELAY_FIXED_1);
>  			break;
> +		}
>  	}
> +	__this_cpu_write(spinlock_delay, delay);
>  }
>  
>  /*
> 

  reply	other threads:[~2013-01-10  3:13 UTC|newest]

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-01-08 22:26 [PATCH 0/5] x86,smp: make ticket spinlock proportional backoff w/ auto tuning Rik van Riel
2013-01-08 22:30 ` [PATCH 3/5] x86,smp: auto tune spinlock backoff delay factor Rik van Riel
2013-01-10  3:13   ` Rafael Aquini [this message]
2013-01-10 12:49   ` Michel Lespinasse
2013-01-08 22:31 ` [PATCH 4/5] x86,smp: keep spinlock delay values per hashed spinlock address Rik van Riel
2013-01-10  3:14   ` Rafael Aquini
2013-01-10 13:01   ` Michel Lespinasse
2013-01-10 13:05     ` Rik van Riel
2013-01-10 13:15       ` Michel Lespinasse
2013-01-08 22:32 ` [DEBUG PATCH 5/5] x86,smp: add debugging code to track spinlock delay value Rik van Riel
2013-01-08 22:32 ` [PATCH 2/5] x86,smp: proportional backoff for ticket spinlocks Rik van Riel
2013-01-08 22:50   ` Eric Dumazet
2013-01-08 22:54     ` Rik van Riel
2013-01-10  2:30   ` Rafael Aquini
2013-01-08 22:32 ` [PATCH 1/5] x86,smp: move waiting on contended ticket lock out of line Rik van Riel
2013-01-08 22:43   ` Eric Dumazet
2013-01-10 17:38   ` Raghavendra K T
2013-01-09 12:50 ` [PATCH 0/5] x86,smp: make ticket spinlock proportional backoff w/ auto tuning Raghavendra K T
2013-01-10  2:27   ` Rafael Aquini
2013-01-10 17:36     ` Raghavendra K T
2013-01-11 20:11       ` Rik van Riel
2013-01-13 18:07         ` Raghavendra K T
2013-01-10 15:19 ` Mike Galbraith
2013-01-10 15:31   ` Rik van Riel
2013-01-10 19:30     ` Mike Galbraith
2013-01-24 13:28       ` Ingo Molnar
2013-01-10 22:24 ` Chegu Vinod

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=20130110031339.GC1636@x61.redhat.com \
    --to=aquini@redhat.com \
    --cc=JBeulich@novell.com \
    --cc=chegu_vinod@hp.com \
    --cc=eric.dumazet@gmail.com \
    --cc=jeremy@goop.org \
    --cc=knoel@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=lwoodman@redhat.com \
    --cc=mingo@redhat.com \
    --cc=raghavendra.kt@linux.vnet.ibm.com \
    --cc=riel@redhat.com \
    --cc=walken@google.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;
as well as URLs for NNTP newsgroup(s).