mm/page_alloc: use existing highatomic reserves on the buddy fastpath
Subsystems (2)
PATCH v3 1/1 mm/page_alloc: use existing highatomic reserves on the buddy fastpath
ALLOC_HIGHATOMIC currently provides both access to MIGRATE_HIGHATOMIC free pages and permission to create new highatomic pageblock reserves. This makes it unsuitable for the fastpath.
However, the fastpath can reach rmqueue_buddy() while MIGRATE_HIGHATOMIC reserves have free pages available. In this situation, the allocation can fall back to other migratetypes without trying those reserves first.
Allow high-priority non-blocking allocations to use existing MIGRATE_HIGHATOMIC reserves on the buddy fastpath without growing them. First tighten the criteria for reserving pageblocks so that growth may only occur in the slowpath. Then allow fastpath usage by enabling ALLOC_HIGHATOMIC when the GFP mask describes a non-blocking high-priority allocation. This logic has been factored out from gfp_to_alloc_flags() to a new function gfp_to_alloc_flags_nonblocking().
A UDP receive workload was run with free MIGRATE_HIGHATOMIC pageblocks available in the target zone. Before this patch, the workload did not consume these blocks. With this patch, eligible order-1 allocations reaching the buddy path consumed existing MIGRATE_HIGHATOMIC pageblocks, with no highatomic misses observed. The workload did not grow highatomic reserves and NAPI page-frag allocations remained healthy with no failures or order-0 fallbacks.
---
v3:
- remove ALLOC_HIGHATOMIC_RESERVE and let ALLOC_HIGHATOMIC keep original behavior
- use ALLOC_WMARK_MIN to identify slowpath before growing reserve
- factor out non-blocking logic from gfp_to_alloc_flags() into *_nonblocking() helper
- dropped reviewed-by tag
v2: https://lore.kernel.org/linux-mm/[email protected]/
- decouple use semantics from ALLOC_HIGHATOMIC_RESERVE
- update changelog to reflect above change and reword test paragraph
- adjust comment in PCP path
v1: https://lore.kernel.org/linux-mm/[email protected]/
mm/page_alloc.c | 44 ++++++++++++++++++++++++++++++-------------- 1 file changed, 30 insertions(+), 14 deletions(-)
mm/page_alloc.c +30 -14
@@ -3247,10 +3247,11 @@ struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, }while (check_new_pages (page, order));/* -* If this is a high-order atomic allocation then check -* if the pageblock should be reserved for the future +* Slowpath (precarious) high-atomic allocations may reserve +* a pageblock for future use. */ -if (unlikely (alloc_flags &ALLOC_HIGHATOMIC )) +if (unlikely ((alloc_flags &ALLOC_HIGHATOMIC ) && + ((alloc_flags &ALLOC_WMARK_MASK ) ==ALLOC_WMARK_MIN )))reserve_highatomic_pageblock (page, order, zone);__count_zid_vm_events (PGALLOC ,page_zonenum (page),1 << order);
@@ -4473,6 +4474,29 @@ static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, } } +static inline unsigned int +gfp_to_alloc_flags_nonblocking (gfp_t gfp_mask,unsigned int order) +{ +unsigned int alloc_flags =0 ; + +if (gfp_mask & __GFP_DIRECT_RECLAIM) +return 0 ; + +/* +* Not worth trying to allocate harder for __GFP_NOMEMALLOC even +* if it can't schedule. +*/ +if (gfp_mask & __GFP_NOMEMALLOC) +return 0 ; + + alloc_flags |=ALLOC_NON_BLOCK ; + +if (order >0 && (gfp_mask & __GFP_HIGH)) + alloc_flags |=ALLOC_HIGHATOMIC ; + +return alloc_flags; +} +static inline unsigned int gfp_to_alloc_flags (gfp_t gfp_mask,unsigned int order) {
@@ -4495,18 +4519,9 @@ gfp_to_alloc_flags(gfp_t gfp_mask, unsigned int order) alloc_flags |= (__force int ) (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); -if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { -/* -* Not worth trying to allocate harder for __GFP_NOMEMALLOC even -* if it can't schedule. -*/ -if (!(gfp_mask & __GFP_NOMEMALLOC)) { - alloc_flags |=ALLOC_NON_BLOCK ; - -if (order >0 && (alloc_flags &ALLOC_MIN_RESERVE )) - alloc_flags |=ALLOC_HIGHATOMIC ; - } + alloc_flags |=gfp_to_alloc_flags_nonblocking (gfp_mask, order); +if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {/* * Ignore cpuset mems for non-blocking __GFP_HIGH (probably * GFP_ATOMIC) rather than fail, see the comment for
@@ -5299,6 +5314,7 @@ struct page *__alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order,* memory until all local zones are considered. */ alloc_flags |=alloc_flags_nofragment (zonelist_zone (ac.preferred_zoneref), gfp); + alloc_flags |=gfp_to_alloc_flags_nonblocking (gfp, order) &ALLOC_HIGHATOMIC ;/* First allocation attempt */ page =get_page_from_freelist (alloc_gfp, order, alloc_flags, &ac);
Signature · 1 line
2.54.0
Quoted 24 lines · JP Kobryn
ALLOC_HIGHATOMICcurrently provides both access toMIGRATE_HIGHATOMICfree pages and permission to create new highatomic pageblock reserves. This makes it unsuitable for the fastpath.However, the fastpath can reach
rmqueue_buddy()whileMIGRATE_HIGHATOMICreserves have free pages available. In this situation, the allocation can fall back to other migratetypes without trying those reserves first.Allow high-priority non-blocking allocations to use existing
MIGRATE_HIGHATOMICreserves on the buddy fastpath without growing them. First tighten the criteria for reserving pageblocks so that growth may only occur in the slowpath. Then allow fastpath usage by enablingALLOC_HIGHATOMICwhen the GFP mask describes a non-blocking high-priority allocation. This logic has been factored out fromgfp_to_alloc_flags()to a new functiongfp_to_alloc_flags_nonblocking().A UDP receive workload was run with free
MIGRATE_HIGHATOMICpageblocks available in the target zone. Before this patch, the workload did not consume these blocks. With this patch, eligible order-1 allocations reaching the buddy path consumed existingMIGRATE_HIGHATOMICpageblocks, with no highatomic misses observed. The workload did not grow highatomic reserves and NAPI page-frag allocations remained healthy with no failures or order-0 fallbacks.
Quoted 1 line
Signed-off-by JP Kobryn <[email protected]>
Quoted 95 lines
---
v3:
- removeALLOC_HIGHATOMIC_RESERVEand letALLOC_HIGHATOMICkeep original behavior
- useALLOC_WMARK_MINto identify slowpath before growing reserve
- factor out non-blocking logic fromgfp_to_alloc_flags()into*_nonblocking()helper
- dropped reviewed-by tagv2: https://lore.kernel.org/linux-mm/[email protected]/
- decouple use semantics fromALLOC_HIGHATOMIC_RESERVE
- update changelog to reflect above change and reword test paragraph
- adjust comment in PCP pathv1: https://lore.kernel.org/linux-mm/[email protected]/
mm/page_alloc.c | 44 ++++++++++++++++++++++++++++++-------------- 1 file changed, 30 insertions(+), 14 deletions(-)mm/page_alloc.c +30
-14@@ -3247,10 +3247,11 @@ struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, }while (check_new_pages (page, order));/* -* If this is a high-order atomic allocation then check -* if the pageblock should be reserved for the future +* Slowpath (precarious) high-atomic allocations may reserve +* a pageblock for future use. */ -if (unlikely (alloc_flags &ALLOC_HIGHATOMIC )) +if (unlikely ((alloc_flags &ALLOC_HIGHATOMIC ) && + ((alloc_flags &ALLOC_WMARK_MASK ) ==ALLOC_WMARK_MIN )))reserve_highatomic_pageblock (page, order, zone);__count_zid_vm_events (PGALLOC ,page_zonenum (page),1 << order);@@ -4473,6 +4474,29 @@ static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, } } +static inline unsigned int +gfp_to_alloc_flags_nonblocking (gfp_t gfp_mask,unsigned int order) +{ +unsigned int alloc_flags =0 ; + +if (gfp_mask & __GFP_DIRECT_RECLAIM) +return 0 ; + +/* +* Not worth trying to allocate harder for __GFP_NOMEMALLOC even +* if it can't schedule. +*/ +if (gfp_mask & __GFP_NOMEMALLOC) +return 0 ; + + alloc_flags |=ALLOC_NON_BLOCK ; + +if (order >0 && (gfp_mask & __GFP_HIGH)) + alloc_flags |=ALLOC_HIGHATOMIC ; + +return alloc_flags; +} +static inline unsigned int gfp_to_alloc_flags (gfp_t gfp_mask,unsigned int order) {@@ -4495,18 +4519,9 @@ gfp_to_alloc_flags(gfp_t gfp_mask, unsigned int order) alloc_flags |= (__force int ) (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); -if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { -/* -* Not worth trying to allocate harder for __GFP_NOMEMALLOC even -* if it can't schedule. -*/ -if (!(gfp_mask & __GFP_NOMEMALLOC)) { - alloc_flags |=ALLOC_NON_BLOCK ; - -if (order >0 && (alloc_flags &ALLOC_MIN_RESERVE )) - alloc_flags |=ALLOC_HIGHATOMIC ; - } + alloc_flags |=gfp_to_alloc_flags_nonblocking (gfp_mask, order); +if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {/* * Ignore cpuset mems for non-blocking __GFP_HIGH (probably * GFP_ATOMIC) rather than fail, see the comment for @@ -5299,6 +5314,7 @@ struct page *__alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order,* memory until all local zones are considered. */ alloc_flags |=alloc_flags_nofragment (zonelist_zone (ac.preferred_zoneref), gfp); + alloc_flags |=gfp_to_alloc_flags_nonblocking (gfp, order) &ALLOC_HIGHATOMIC ;/* First allocation attempt */ page =get_page_from_freelist (alloc_gfp, order, alloc_flags, &ac);
Quoted 24 lines · JP Kobryn
ALLOC_HIGHATOMICcurrently provides both access toMIGRATE_HIGHATOMICfree pages and permission to create new highatomic pageblock reserves. This makes it unsuitable for the fastpath.However, the fastpath can reach
rmqueue_buddy()whileMIGRATE_HIGHATOMICreserves have free pages available. In this situation, the allocation can fall back to other migratetypes without trying those reserves first.Allow high-priority non-blocking allocations to use existing
MIGRATE_HIGHATOMICreserves on the buddy fastpath without growing them. First tighten the criteria for reserving pageblocks so that growth may only occur in the slowpath. Then allow fastpath usage by enablingALLOC_HIGHATOMICwhen the GFP mask describes a non-blocking high-priority allocation. This logic has been factored out fromgfp_to_alloc_flags()to a new functiongfp_to_alloc_flags_nonblocking().A UDP receive workload was run with free
MIGRATE_HIGHATOMICpageblocks available in the target zone. Before this patch, the workload did not consume these blocks. With this patch, eligible order-1 allocations reaching the buddy path consumed existingMIGRATE_HIGHATOMICpageblocks, with no highatomic misses observed. The workload did not grow highatomic reserves and NAPI page-frag allocations remained healthy with no failures or order-0 fallbacks.
Quoted 1 line
Signed-off-by JP Kobryn <[email protected]>
Awesome.
Quoted 24 lines · JP Kobryn
ALLOC_HIGHATOMICcurrently provides both access toMIGRATE_HIGHATOMICfree pages and permission to create new highatomic pageblock reserves. This makes it unsuitable for the fastpath.However, the fastpath can reach
rmqueue_buddy()whileMIGRATE_HIGHATOMICreserves have free pages available. In this situation, the allocation can fall back to other migratetypes without trying those reserves first.Allow high-priority non-blocking allocations to use existing
MIGRATE_HIGHATOMICreserves on the buddy fastpath without growing them. First tighten the criteria for reserving pageblocks so that growth may only occur in the slowpath. Then allow fastpath usage by enablingALLOC_HIGHATOMICwhen the GFP mask describes a non-blocking high-priority allocation. This logic has been factored out fromgfp_to_alloc_flags()to a new functiongfp_to_alloc_flags_nonblocking().A UDP receive workload was run with free
MIGRATE_HIGHATOMICpageblocks available in the target zone. Before this patch, the workload did not consume these blocks. With this patch, eligible order-1 allocations reaching the buddy path consumed existingMIGRATE_HIGHATOMICpageblocks, with no highatomic misses observed. The workload did not grow highatomic reserves and NAPI page-frag allocations remained healthy with no failures or order-0 fallbacks.
Quoted 1 line
Signed-off-by JP Kobryn <[email protected]>
This is awesome.