From 53853e2d2bfb748a8b5aa2fd1de15699266865e0 Mon Sep 17 00:00:00 2001 From: Vlastimil Babka Date: Thu, 9 Oct 2014 15:27:02 -0700 Subject: mm, compaction: defer each zone individually instead of preferred zone When direct sync compaction is often unsuccessful, it may become deferred for some time to avoid further useless attempts, both sync and async. Successful high-order allocations un-defer compaction, while further unsuccessful compaction attempts prolong the compaction deferred period. Currently the checking and setting deferred status is performed only on the preferred zone of the allocation that invoked direct compaction. But compaction itself is attempted on all eligible zones in the zonelist, so the behavior is suboptimal and may lead both to scenarios where 1) compaction is attempted uselessly, or 2) where it's not attempted despite good chances of succeeding, as shown on the examples below: 1) A direct compaction with Normal preferred zone failed and set deferred compaction for the Normal zone. Another unrelated direct compaction with DMA32 as preferred zone will attempt to compact DMA32 zone even though the first compaction attempt also included DMA32 zone. In another scenario, compaction with Normal preferred zone failed to compact Normal zone, but succeeded in the DMA32 zone, so it will not defer compaction. In the next attempt, it will try Normal zone which will fail again, instead of skipping Normal zone and trying DMA32 directly. 2) Kswapd will balance DMA32 zone and reset defer status based on watermarks looking good. A direct compaction with preferred Normal zone will skip compaction of all zones including DMA32 because Normal was still deferred. The allocation might have succeeded in DMA32, but won't. This patch makes compaction deferring work on individual zone basis instead of preferred zone. For each zone, it checks compaction_deferred() to decide if the zone should be skipped. If watermarks fail after compacting the zone, defer_compaction() is called. The zone where watermarks passed can still be deferred when the allocation attempt is unsuccessful. When allocation is successful, compaction_defer_reset() is called for the zone containing the allocated page. This approach should approximate calling defer_compaction() only on zones where compaction was attempted and did not yield allocated page. There might be corner cases but that is inevitable as long as the decision to stop compacting dues not guarantee that a page will be allocated. Due to a new COMPACT_DEFERRED return value, some functions relying implicitly on COMPACT_SKIPPED = 0 had to be updated, with comments made more accurate. The did_some_progress output parameter of __alloc_pages_direct_compact() is removed completely, as the caller actually does not use it after compaction sets it - it is only considered when direct reclaim sets it. During testing on a two-node machine with a single very small Normal zone on node 1, this patch has improved success rates in stress-highalloc mmtests benchmark. The success here were previously made worse by commit 3a025760fc15 ("mm: page_alloc: spill to remote nodes before waking kswapd") as kswapd was no longer resetting often enough the deferred compaction for the Normal zone, and DMA32 zones on both nodes were thus not considered for compaction. On different machine, success rates were improved with __GFP_NO_KSWAPD allocations. [akpm@linux-foundation.org: fix CONFIG_COMPACTION=n build] Signed-off-by: Vlastimil Babka Acked-by: Minchan Kim Reviewed-by: Zhang Yanfei Acked-by: Mel Gorman Cc: Joonsoo Kim Cc: Michal Nazarewicz Cc: Naoya Horiguchi Cc: Christoph Lameter Cc: Rik van Riel Cc: David Rientjes Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- include/linux/compaction.h | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) (limited to 'include/linux/compaction.h') diff --git a/include/linux/compaction.h b/include/linux/compaction.h index 01e3132820da..b2e4c92d0445 100644 --- a/include/linux/compaction.h +++ b/include/linux/compaction.h @@ -2,14 +2,16 @@ #define _LINUX_COMPACTION_H /* Return values for compact_zone() and try_to_compact_pages() */ +/* compaction didn't start as it was deferred due to past failures */ +#define COMPACT_DEFERRED 0 /* compaction didn't start as it was not possible or direct reclaim was more suitable */ -#define COMPACT_SKIPPED 0 +#define COMPACT_SKIPPED 1 /* compaction should continue to another pageblock */ -#define COMPACT_CONTINUE 1 +#define COMPACT_CONTINUE 2 /* direct compaction partially compacted a zone and there are suitable pages */ -#define COMPACT_PARTIAL 2 +#define COMPACT_PARTIAL 3 /* The full zone was compacted */ -#define COMPACT_COMPLETE 3 +#define COMPACT_COMPLETE 4 #ifdef CONFIG_COMPACTION extern int sysctl_compact_memory; @@ -22,7 +24,8 @@ extern int sysctl_extfrag_handler(struct ctl_table *table, int write, extern int fragmentation_index(struct zone *zone, unsigned int order); extern unsigned long try_to_compact_pages(struct zonelist *zonelist, int order, gfp_t gfp_mask, nodemask_t *mask, - enum migrate_mode mode, bool *contended); + enum migrate_mode mode, bool *contended, + struct zone **candidate_zone); extern void compact_pgdat(pg_data_t *pgdat, int order); extern void reset_isolation_suitable(pg_data_t *pgdat); extern unsigned long compaction_suitable(struct zone *zone, int order); @@ -91,7 +94,8 @@ static inline bool compaction_restarting(struct zone *zone, int order) #else static inline unsigned long try_to_compact_pages(struct zonelist *zonelist, int order, gfp_t gfp_mask, nodemask_t *nodemask, - enum migrate_mode mode, bool *contended) + enum migrate_mode mode, bool *contended, + struct zone **candidate_zone) { return COMPACT_CONTINUE; } -- cgit v1.2.3 From 1f9efdef4f3f1d2a073e524113fd0038af636f2b Mon Sep 17 00:00:00 2001 From: Vlastimil Babka Date: Thu, 9 Oct 2014 15:27:14 -0700 Subject: mm, compaction: khugepaged should not give up due to need_resched() Async compaction aborts when it detects zone lock contention or need_resched() is true. David Rientjes has reported that in practice, most direct async compactions for THP allocation abort due to need_resched(). This means that a second direct compaction is never attempted, which might be OK for a page fault, but khugepaged is intended to attempt a sync compaction in such case and in these cases it won't. This patch replaces "bool contended" in compact_control with an int that distinguishes between aborting due to need_resched() and aborting due to lock contention. This allows propagating the abort through all compaction functions as before, but passing the abort reason up to __alloc_pages_slowpath() which decides when to continue with direct reclaim and another compaction attempt. Another problem is that try_to_compact_pages() did not act upon the reported contention (both need_resched() or lock contention) immediately and would proceed with another zone from the zonelist. When need_resched() is true, that means initializing another zone compaction, only to check again need_resched() in isolate_migratepages() and aborting. For zone lock contention, the unintended consequence is that the lock contended status reported back to the allocator is detrmined from the last zone where compaction was attempted, which is rather arbitrary. This patch fixes the problem in the following way: - async compaction of a zone aborting due to need_resched() or fatal signal pending means that further zones should not be tried. We report COMPACT_CONTENDED_SCHED to the allocator. - aborting zone compaction due to lock contention means we can still try another zone, since it has different set of locks. We report back COMPACT_CONTENDED_LOCK only if *all* zones where compaction was attempted, it was aborted due to lock contention. As a result of these fixes, khugepaged will proceed with second sync compaction as intended, when the preceding async compaction aborted due to need_resched(). Page fault compactions aborting due to need_resched() will spare some cycles previously wasted by initializing another zone compaction only to abort again. Lock contention will be reported only when compaction in all zones aborted due to lock contention, and therefore it's not a good idea to try again after reclaim. In stress-highalloc from mmtests configured to use __GFP_NO_KSWAPD, this has improved number of THP collapse allocations by 10%, which shows positive effect on khugepaged. The benchmark's success rates are unchanged as it is not recognized as khugepaged. Numbers of compact_stall and compact_fail events have however decreased by 20%, with compact_success still a bit improved, which is good. With benchmark configured not to use __GFP_NO_KSWAPD, there is 6% improvement in THP collapse allocations, and only slight improvement in stalls and failures. [akpm@linux-foundation.org: fix warnings] Reported-by: David Rientjes Signed-off-by: Vlastimil Babka Cc: Minchan Kim Acked-by: Mel Gorman Cc: Joonsoo Kim Cc: Michal Nazarewicz Cc: Naoya Horiguchi Cc: Christoph Lameter Cc: Rik van Riel Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- include/linux/compaction.h | 12 +++++-- mm/compaction.c | 87 ++++++++++++++++++++++++++++++++++++++++------ mm/internal.h | 4 +-- mm/page_alloc.c | 45 +++++++++++++++++------- 4 files changed, 121 insertions(+), 27 deletions(-) (limited to 'include/linux/compaction.h') diff --git a/include/linux/compaction.h b/include/linux/compaction.h index b2e4c92d0445..60bdf8dc02a3 100644 --- a/include/linux/compaction.h +++ b/include/linux/compaction.h @@ -13,6 +13,14 @@ /* The full zone was compacted */ #define COMPACT_COMPLETE 4 +/* Used to signal whether compaction detected need_sched() or lock contention */ +/* No contention detected */ +#define COMPACT_CONTENDED_NONE 0 +/* Either need_sched() was true or fatal signal pending */ +#define COMPACT_CONTENDED_SCHED 1 +/* Zone lock or lru_lock was contended in async compaction */ +#define COMPACT_CONTENDED_LOCK 2 + #ifdef CONFIG_COMPACTION extern int sysctl_compact_memory; extern int sysctl_compaction_handler(struct ctl_table *table, int write, @@ -24,7 +32,7 @@ extern int sysctl_extfrag_handler(struct ctl_table *table, int write, extern int fragmentation_index(struct zone *zone, unsigned int order); extern unsigned long try_to_compact_pages(struct zonelist *zonelist, int order, gfp_t gfp_mask, nodemask_t *mask, - enum migrate_mode mode, bool *contended, + enum migrate_mode mode, int *contended, struct zone **candidate_zone); extern void compact_pgdat(pg_data_t *pgdat, int order); extern void reset_isolation_suitable(pg_data_t *pgdat); @@ -94,7 +102,7 @@ static inline bool compaction_restarting(struct zone *zone, int order) #else static inline unsigned long try_to_compact_pages(struct zonelist *zonelist, int order, gfp_t gfp_mask, nodemask_t *nodemask, - enum migrate_mode mode, bool *contended, + enum migrate_mode mode, int *contended, struct zone **candidate_zone) { return COMPACT_CONTINUE; diff --git a/mm/compaction.c b/mm/compaction.c index 1067c07cb33d..26bb20ef853d 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -223,9 +223,21 @@ static void update_pageblock_skip(struct compact_control *cc, } #endif /* CONFIG_COMPACTION */ -static inline bool should_release_lock(spinlock_t *lock) +static int should_release_lock(spinlock_t *lock) { - return need_resched() || spin_is_contended(lock); + /* + * Sched contention has higher priority here as we may potentially + * have to abort whole compaction ASAP. Returning with lock contention + * means we will try another zone, and further decisions are + * influenced only when all zones are lock contended. That means + * potentially missing a lock contention is less critical. + */ + if (need_resched()) + return COMPACT_CONTENDED_SCHED; + else if (spin_is_contended(lock)) + return COMPACT_CONTENDED_LOCK; + + return COMPACT_CONTENDED_NONE; } /* @@ -240,7 +252,9 @@ static inline bool should_release_lock(spinlock_t *lock) static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, bool locked, struct compact_control *cc) { - if (should_release_lock(lock)) { + int contended = should_release_lock(lock); + + if (contended) { if (locked) { spin_unlock_irqrestore(lock, *flags); locked = false; @@ -248,7 +262,7 @@ static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, /* async aborts if taking too long or contended */ if (cc->mode == MIGRATE_ASYNC) { - cc->contended = true; + cc->contended = contended; return false; } @@ -274,7 +288,7 @@ static inline bool compact_should_abort(struct compact_control *cc) /* async compaction aborts if contended */ if (need_resched()) { if (cc->mode == MIGRATE_ASYNC) { - cc->contended = true; + cc->contended = COMPACT_CONTENDED_SCHED; return true; } @@ -1140,7 +1154,7 @@ out: } static unsigned long compact_zone_order(struct zone *zone, int order, - gfp_t gfp_mask, enum migrate_mode mode, bool *contended) + gfp_t gfp_mask, enum migrate_mode mode, int *contended) { unsigned long ret; struct compact_control cc = { @@ -1172,14 +1186,15 @@ int sysctl_extfrag_threshold = 500; * @gfp_mask: The GFP mask of the current allocation * @nodemask: The allowed nodes to allocate from * @mode: The migration mode for async, sync light, or sync migration - * @contended: Return value that is true if compaction was aborted due to lock contention + * @contended: Return value that determines if compaction was aborted due to + * need_resched() or lock contention * @candidate_zone: Return the zone where we think allocation should succeed * * This is the main entry point for direct page compaction. */ unsigned long try_to_compact_pages(struct zonelist *zonelist, int order, gfp_t gfp_mask, nodemask_t *nodemask, - enum migrate_mode mode, bool *contended, + enum migrate_mode mode, int *contended, struct zone **candidate_zone) { enum zone_type high_zoneidx = gfp_zone(gfp_mask); @@ -1189,6 +1204,9 @@ unsigned long try_to_compact_pages(struct zonelist *zonelist, struct zone *zone; int rc = COMPACT_DEFERRED; int alloc_flags = 0; + int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ + + *contended = COMPACT_CONTENDED_NONE; /* Check if the GFP flags allow compaction */ if (!order || !may_enter_fs || !may_perform_io) @@ -1202,13 +1220,19 @@ unsigned long try_to_compact_pages(struct zonelist *zonelist, for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, nodemask) { int status; + int zone_contended; if (compaction_deferred(zone, order)) continue; status = compact_zone_order(zone, order, gfp_mask, mode, - contended); + &zone_contended); rc = max(status, rc); + /* + * It takes at least one zone that wasn't lock contended + * to clear all_zones_contended. + */ + all_zones_contended &= zone_contended; /* If a normal allocation would succeed, stop compacting */ if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, @@ -1221,8 +1245,21 @@ unsigned long try_to_compact_pages(struct zonelist *zonelist, * succeeds in this zone. */ compaction_defer_reset(zone, order, false); - break; - } else if (mode != MIGRATE_ASYNC) { + /* + * It is possible that async compaction aborted due to + * need_resched() and the watermarks were ok thanks to + * somebody else freeing memory. The allocation can + * however still fail so we better signal the + * need_resched() contention anyway (this will not + * prevent the allocation attempt). + */ + if (zone_contended == COMPACT_CONTENDED_SCHED) + *contended = COMPACT_CONTENDED_SCHED; + + goto break_loop; + } + + if (mode != MIGRATE_ASYNC) { /* * We think that allocation won't succeed in this zone * so we defer compaction there. If it ends up @@ -1230,8 +1267,36 @@ unsigned long try_to_compact_pages(struct zonelist *zonelist, */ defer_compaction(zone, order); } + + /* + * We might have stopped compacting due to need_resched() in + * async compaction, or due to a fatal signal detected. In that + * case do not try further zones and signal need_resched() + * contention. + */ + if ((zone_contended == COMPACT_CONTENDED_SCHED) + || fatal_signal_pending(current)) { + *contended = COMPACT_CONTENDED_SCHED; + goto break_loop; + } + + continue; +break_loop: + /* + * We might not have tried all the zones, so be conservative + * and assume they are not all lock contended. + */ + all_zones_contended = 0; + break; } + /* + * If at least one zone wasn't deferred or skipped, we report if all + * zones that were tried were lock contended. + */ + if (rc > COMPACT_SKIPPED && all_zones_contended) + *contended = COMPACT_CONTENDED_LOCK; + return rc; } diff --git a/mm/internal.h b/mm/internal.h index 5a0738fa649c..4c1d604c396c 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -144,8 +144,8 @@ struct compact_control { int order; /* order a direct compactor needs */ int migratetype; /* MOVABLE, RECLAIMABLE etc */ struct zone *zone; - bool contended; /* True if a lock was contended, or - * need_resched() true during async + int contended; /* Signal need_sched() or lock + * contention detected during * compaction */ }; diff --git a/mm/page_alloc.c b/mm/page_alloc.c index dfbf54b51649..313338d74095 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -2297,7 +2297,7 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, enum zone_type high_zoneidx, nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, int classzone_idx, int migratetype, enum migrate_mode mode, - bool *contended_compaction, bool *deferred_compaction) + int *contended_compaction, bool *deferred_compaction) { struct zone *last_compact_zone = NULL; unsigned long compact_result; @@ -2371,7 +2371,7 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, enum zone_type high_zoneidx, nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, int classzone_idx, int migratetype, enum migrate_mode mode, - bool *contended_compaction, bool *deferred_compaction) + int *contended_compaction, bool *deferred_compaction) { return NULL; } @@ -2547,7 +2547,7 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, unsigned long did_some_progress; enum migrate_mode migration_mode = MIGRATE_ASYNC; bool deferred_compaction = false; - bool contended_compaction = false; + int contended_compaction = COMPACT_CONTENDED_NONE; /* * In the slowpath, we sanity check order to avoid ever trying to @@ -2651,15 +2651,36 @@ rebalance: if (page) goto got_pg; - /* - * If compaction is deferred for high-order allocations, it is because - * sync compaction recently failed. In this is the case and the caller - * requested a movable allocation that does not heavily disrupt the - * system then fail the allocation instead of entering direct reclaim. - */ - if ((deferred_compaction || contended_compaction) && - (gfp_mask & __GFP_NO_KSWAPD)) - goto nopage; + /* Checks for THP-specific high-order allocations */ + if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) { + /* + * If compaction is deferred for high-order allocations, it is + * because sync compaction recently failed. If this is the case + * and the caller requested a THP allocation, we do not want + * to heavily disrupt the system, so we fail the allocation + * instead of entering direct reclaim. + */ + if (deferred_compaction) + goto nopage; + + /* + * In all zones where compaction was attempted (and not + * deferred or skipped), lock contention has been detected. + * For THP allocation we do not want to disrupt the others + * so we fallback to base pages instead. + */ + if (contended_compaction == COMPACT_CONTENDED_LOCK) + goto nopage; + + /* + * If compaction was aborted due to need_resched(), we do not + * want to further increase allocation latency, unless it is + * khugepaged trying to collapse. + */ + if (contended_compaction == COMPACT_CONTENDED_SCHED + && !(current->flags & PF_KTHREAD)) + goto nopage; + } /* * It can become very expensive to allocate transparent hugepages at -- cgit v1.2.3