Revision 72853e2991a2702ae93aaf889ac7db743a415dd3 authored by Mel Gorman on 09 September 2010, 23:38:16 UTC, committed by Linus Torvalds on 10 September 2010, 01:57:25 UTC
When allocating a page, the system uses NR_FREE_PAGES counters to
determine if watermarks would remain intact after the allocation was made.
This check is made without interrupts disabled or the zone lock held and
so is race-prone by nature.  Unfortunately, when pages are being freed in
batch, the counters are updated before the pages are added on the list.
During this window, the counters are misleading as the pages do not exist
yet.  When under significant pressure on systems with large numbers of
CPUs, it's possible for processes to make progress even though they should
have been stalled.  This is particularly problematic if a number of the
processes are using GFP_ATOMIC as the min watermark can be accidentally
breached and in extreme cases, the system can livelock.

This patch updates the counters after the pages have been added to the
list.  This makes the allocator more cautious with respect to preserving
the watermarks and mitigates livelock possibilities.

[akpm@linux-foundation.org: avoid modifying incoming args]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent 5ee28a4
Raw File
proportions.c
/*
 * Floating proportions
 *
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
 *
 * Description:
 *
 * The floating proportion is a time derivative with an exponentially decaying
 * history:
 *
 *   p_{j} = \Sum_{i=0} (dx_{j}/dt_{-i}) / 2^(1+i)
 *
 * Where j is an element from {prop_local}, x_{j} is j's number of events,
 * and i the time period over which the differential is taken. So d/dt_{-i} is
 * the differential over the i-th last period.
 *
 * The decaying history gives smooth transitions. The time differential carries
 * the notion of speed.
 *
 * The denominator is 2^(1+i) because we want the series to be normalised, ie.
 *
 *   \Sum_{i=0} 1/2^(1+i) = 1
 *
 * Further more, if we measure time (t) in the same events as x; so that:
 *
 *   t = \Sum_{j} x_{j}
 *
 * we get that:
 *
 *   \Sum_{j} p_{j} = 1
 *
 * Writing this in an iterative fashion we get (dropping the 'd's):
 *
 *   if (++x_{j}, ++t > period)
 *     t /= 2;
 *     for_each (j)
 *       x_{j} /= 2;
 *
 * so that:
 *
 *   p_{j} = x_{j} / t;
 *
 * We optimize away the '/= 2' for the global time delta by noting that:
 *
 *   if (++t > period) t /= 2:
 *
 * Can be approximated by:
 *
 *   period/2 + (++t % period/2)
 *
 * [ Furthermore, when we choose period to be 2^n it can be written in terms of
 *   binary operations and wraparound artefacts disappear. ]
 *
 * Also note that this yields a natural counter of the elapsed periods:
 *
 *   c = t / (period/2)
 *
 * [ Its monotonic increasing property can be applied to mitigate the wrap-
 *   around issue. ]
 *
 * This allows us to do away with the loop over all prop_locals on each period
 * expiration. By remembering the period count under which it was last accessed
 * as c_{j}, we can obtain the number of 'missed' cycles from:
 *
 *   c - c_{j}
 *
 * We can then lazily catch up to the global period count every time we are
 * going to use x_{j}, by doing:
 *
 *   x_{j} /= 2^(c - c_{j}), c_{j} = c
 */

#include <linux/proportions.h>
#include <linux/rcupdate.h>

int prop_descriptor_init(struct prop_descriptor *pd, int shift)
{
	int err;

	if (shift > PROP_MAX_SHIFT)
		shift = PROP_MAX_SHIFT;

	pd->index = 0;
	pd->pg[0].shift = shift;
	mutex_init(&pd->mutex);
	err = percpu_counter_init(&pd->pg[0].events, 0);
	if (err)
		goto out;

	err = percpu_counter_init(&pd->pg[1].events, 0);
	if (err)
		percpu_counter_destroy(&pd->pg[0].events);

out:
	return err;
}

/*
 * We have two copies, and flip between them to make it seem like an atomic
 * update. The update is not really atomic wrt the events counter, but
 * it is internally consistent with the bit layout depending on shift.
 *
 * We copy the events count, move the bits around and flip the index.
 */
void prop_change_shift(struct prop_descriptor *pd, int shift)
{
	int index;
	int offset;
	u64 events;
	unsigned long flags;

	if (shift > PROP_MAX_SHIFT)
		shift = PROP_MAX_SHIFT;

	mutex_lock(&pd->mutex);

	index = pd->index ^ 1;
	offset = pd->pg[pd->index].shift - shift;
	if (!offset)
		goto out;

	pd->pg[index].shift = shift;

	local_irq_save(flags);
	events = percpu_counter_sum(&pd->pg[pd->index].events);
	if (offset < 0)
		events <<= -offset;
	else
		events >>= offset;
	percpu_counter_set(&pd->pg[index].events, events);

	/*
	 * ensure the new pg is fully written before the switch
	 */
	smp_wmb();
	pd->index = index;
	local_irq_restore(flags);

	synchronize_rcu();

out:
	mutex_unlock(&pd->mutex);
}

/*
 * wrap the access to the data in an rcu_read_lock() section;
 * this is used to track the active references.
 */
static struct prop_global *prop_get_global(struct prop_descriptor *pd)
__acquires(RCU)
{
	int index;

	rcu_read_lock();
	index = pd->index;
	/*
	 * match the wmb from vcd_flip()
	 */
	smp_rmb();
	return &pd->pg[index];
}

static void prop_put_global(struct prop_descriptor *pd, struct prop_global *pg)
__releases(RCU)
{
	rcu_read_unlock();
}

static void
prop_adjust_shift(int *pl_shift, unsigned long *pl_period, int new_shift)
{
	int offset = *pl_shift - new_shift;

	if (!offset)
		return;

	if (offset < 0)
		*pl_period <<= -offset;
	else
		*pl_period >>= offset;

	*pl_shift = new_shift;
}

/*
 * PERCPU
 */

#define PROP_BATCH (8*(1+ilog2(nr_cpu_ids)))

int prop_local_init_percpu(struct prop_local_percpu *pl)
{
	spin_lock_init(&pl->lock);
	pl->shift = 0;
	pl->period = 0;
	return percpu_counter_init(&pl->events, 0);
}

void prop_local_destroy_percpu(struct prop_local_percpu *pl)
{
	percpu_counter_destroy(&pl->events);
}

/*
 * Catch up with missed period expirations.
 *
 *   until (c_{j} == c)
 *     x_{j} -= x_{j}/2;
 *     c_{j}++;
 */
static
void prop_norm_percpu(struct prop_global *pg, struct prop_local_percpu *pl)
{
	unsigned long period = 1UL << (pg->shift - 1);
	unsigned long period_mask = ~(period - 1);
	unsigned long global_period;
	unsigned long flags;

	global_period = percpu_counter_read(&pg->events);
	global_period &= period_mask;

	/*
	 * Fast path - check if the local and global period count still match
	 * outside of the lock.
	 */
	if (pl->period == global_period)
		return;

	spin_lock_irqsave(&pl->lock, flags);
	prop_adjust_shift(&pl->shift, &pl->period, pg->shift);

	/*
	 * For each missed period, we half the local counter.
	 * basically:
	 *   pl->events >> (global_period - pl->period);
	 */
	period = (global_period - pl->period) >> (pg->shift - 1);
	if (period < BITS_PER_LONG) {
		s64 val = percpu_counter_read(&pl->events);

		if (val < (nr_cpu_ids * PROP_BATCH))
			val = percpu_counter_sum(&pl->events);

		__percpu_counter_add(&pl->events, -val + (val >> period),
					PROP_BATCH);
	} else
		percpu_counter_set(&pl->events, 0);

	pl->period = global_period;
	spin_unlock_irqrestore(&pl->lock, flags);
}

/*
 *   ++x_{j}, ++t
 */
void __prop_inc_percpu(struct prop_descriptor *pd, struct prop_local_percpu *pl)
{
	struct prop_global *pg = prop_get_global(pd);

	prop_norm_percpu(pg, pl);
	__percpu_counter_add(&pl->events, 1, PROP_BATCH);
	percpu_counter_add(&pg->events, 1);
	prop_put_global(pd, pg);
}

/*
 * identical to __prop_inc_percpu, except that it limits this pl's fraction to
 * @frac/PROP_FRAC_BASE by ignoring events when this limit has been exceeded.
 */
void __prop_inc_percpu_max(struct prop_descriptor *pd,
			   struct prop_local_percpu *pl, long frac)
{
	struct prop_global *pg = prop_get_global(pd);

	prop_norm_percpu(pg, pl);

	if (unlikely(frac != PROP_FRAC_BASE)) {
		unsigned long period_2 = 1UL << (pg->shift - 1);
		unsigned long counter_mask = period_2 - 1;
		unsigned long global_count;
		long numerator, denominator;

		numerator = percpu_counter_read_positive(&pl->events);
		global_count = percpu_counter_read(&pg->events);
		denominator = period_2 + (global_count & counter_mask);

		if (numerator > ((denominator * frac) >> PROP_FRAC_SHIFT))
			goto out_put;
	}

	percpu_counter_add(&pl->events, 1);
	percpu_counter_add(&pg->events, 1);

out_put:
	prop_put_global(pd, pg);
}

/*
 * Obtain a fraction of this proportion
 *
 *   p_{j} = x_{j} / (period/2 + t % period/2)
 */
void prop_fraction_percpu(struct prop_descriptor *pd,
		struct prop_local_percpu *pl,
		long *numerator, long *denominator)
{
	struct prop_global *pg = prop_get_global(pd);
	unsigned long period_2 = 1UL << (pg->shift - 1);
	unsigned long counter_mask = period_2 - 1;
	unsigned long global_count;

	prop_norm_percpu(pg, pl);
	*numerator = percpu_counter_read_positive(&pl->events);

	global_count = percpu_counter_read(&pg->events);
	*denominator = period_2 + (global_count & counter_mask);

	prop_put_global(pd, pg);
}

/*
 * SINGLE
 */

int prop_local_init_single(struct prop_local_single *pl)
{
	spin_lock_init(&pl->lock);
	pl->shift = 0;
	pl->period = 0;
	pl->events = 0;
	return 0;
}

void prop_local_destroy_single(struct prop_local_single *pl)
{
}

/*
 * Catch up with missed period expirations.
 */
static
void prop_norm_single(struct prop_global *pg, struct prop_local_single *pl)
{
	unsigned long period = 1UL << (pg->shift - 1);
	unsigned long period_mask = ~(period - 1);
	unsigned long global_period;
	unsigned long flags;

	global_period = percpu_counter_read(&pg->events);
	global_period &= period_mask;

	/*
	 * Fast path - check if the local and global period count still match
	 * outside of the lock.
	 */
	if (pl->period == global_period)
		return;

	spin_lock_irqsave(&pl->lock, flags);
	prop_adjust_shift(&pl->shift, &pl->period, pg->shift);
	/*
	 * For each missed period, we half the local counter.
	 */
	period = (global_period - pl->period) >> (pg->shift - 1);
	if (likely(period < BITS_PER_LONG))
		pl->events >>= period;
	else
		pl->events = 0;
	pl->period = global_period;
	spin_unlock_irqrestore(&pl->lock, flags);
}

/*
 *   ++x_{j}, ++t
 */
void __prop_inc_single(struct prop_descriptor *pd, struct prop_local_single *pl)
{
	struct prop_global *pg = prop_get_global(pd);

	prop_norm_single(pg, pl);
	pl->events++;
	percpu_counter_add(&pg->events, 1);
	prop_put_global(pd, pg);
}

/*
 * Obtain a fraction of this proportion
 *
 *   p_{j} = x_{j} / (period/2 + t % period/2)
 */
void prop_fraction_single(struct prop_descriptor *pd,
	       	struct prop_local_single *pl,
		long *numerator, long *denominator)
{
	struct prop_global *pg = prop_get_global(pd);
	unsigned long period_2 = 1UL << (pg->shift - 1);
	unsigned long counter_mask = period_2 - 1;
	unsigned long global_count;

	prop_norm_single(pg, pl);
	*numerator = pl->events;

	global_count = percpu_counter_read(&pg->events);
	*denominator = period_2 + (global_count & counter_mask);

	prop_put_global(pd, pg);
}
back to top