numa node stats alternative fix
Subsystems (6)
cover numa node stats alternative fix
Hi,
this is an alternative fix for [1] which is currently sitting in the mm tree. I believe that the patch 1 is better because it allows to get rid of __GFP_OTHER_NODE (patch 2) and it uses less branches as well. Vlastimil has also shown [2] that the patch from Jia He is not fully compatible with the code before the patch it tries to fix. I do not think that the issue is serious enough to warrant stable tree inclusion.
Can we have these patches merged instead?
[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH 1/2 mm: fix remote numa hits statistics
From: Michal Hocko <[email protected]>
Jia He has noticed that b9f00e147f27 ("mm, page_alloc: reduce branches in zone_statistics") has an unintentional side effect that remote node allocation requests are accounted as NUMA_MISS rathat than NUMA_HIT and NUMA_OTHER if such a request doesn't use __GFP_OTHER_NODE. There are many of these potentially because the flag is used very rarely while we have many users of __alloc_pages_node.
Fix this by simply ignoring __GFP_OTHER_NODE (it can be removed in a follow up patch) and treat all allocations that were satisfied from the preferred zone's node as NUMA_HITS because this is the same node we requested the allocation from in most cases. If this is not the local node then we just account it as NUMA_OTHER rather than NUMA_LOCAL.
One downsize would be that an allocation request for a node which is outside of the mempolicy nodemask would be reported as a hit which is a bit weird but that was the case before b9f00e147f27 already.
[1] http://paulmck.livejournal.com/38997.html
--- mm/page_alloc.c | 15 ++++----------- 1 file changed, 4 insertions(+), 11 deletions(-)
mm/page_alloc.c +4 -11
@@ -2583,30 +2583,23 @@ int __isolate_free_page(struct page *page, unsigned int order)* Update NUMA hit/miss statistics * * Must be called with interrupts disabled. -* -* When __GFP_OTHER_NODE is set assume the node of the preferred -* zone is the local node. This is useful for daemons who allocate -* memory on behalf of other processes. */ static inline void zone_statistics (struct zone *preferred_zone,struct zone *z,gfp_t flags) {#ifdef CONFIG_NUMA -int local_nid =numa_node_id ();enum zone_stat_item local_stat =NUMA_LOCAL ; -if (unlikely (flags & __GFP_OTHER_NODE)) { +if (z->node !=numa_node_id ()) local_stat =NUMA_OTHER ; - local_nid = preferred_zone->node; - } -if (z->node == local_nid) { +if (z->node == preferred_zone->node)__inc_zone_state (z,NUMA_HIT ); -__inc_zone_state (z, local_stat); - }else { +else {__inc_zone_state (z,NUMA_MISS );__inc_zone_state (preferred_zone,NUMA_FOREIGN ); } +__inc_zone_state (z, local_stat);#endif }
Signature · 7 lines
2.11.0 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
PATCH 2/2 mm: get rid of __GFP_OTHER_NODE
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
The flag has been introduced by 78afd5612deb ("mm: add __GFP_OTHER_NODE flag") to allow proper accounting of remote node allocations done by kernel daemons on behalf of a process - e.g. khugepaged.
After "mm: fix remote numa hits statistics" we do not need and actually use the flag so we can safely remove it because all allocations which are satisfied from their "home" node are accounted properly.
--- include/linux/gfp.h | 13 +++---------- include/trace/events/mmflags.h | 1 - mm/huge_memory.c | 3 +-- mm/khugepaged.c | 5 ++--- mm/page_alloc.c | 5 ++--- tools/perf/builtin-kmem.c | 1 - 6 files changed, 8 insertions(+), 20 deletions(-)
include/linux/gfp.h +3 -10
@@ -38,9 +38,8 @@ struct vm_area_struct;#define ___GFP_ACCOUNT 0x100000u#define ___GFP_NOTRACK 0x200000u#define ___GFP_DIRECT_RECLAIM 0x400000u -#define ___GFP_OTHER_NODE 0x800000u -#define ___GFP_WRITE 0x1000000u -#define ___GFP_KSWAPD_RECLAIM 0x2000000u +#define ___GFP_WRITE 0x800000u +#define ___GFP_KSWAPD_RECLAIM 0x1000000u/* If the above are modified, __GFP_BITS_SHIFT may need updating */ /*
@@ -172,11 +171,6 @@ struct vm_area_struct;* __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of * distinguishing in the source between false positives and allocations that * cannot be supported (e.g. page tables). -* -* __GFP_OTHER_NODE is for allocations that are on a remote node but that -* should not be accounted for as a remote allocation in vmstat. A -* typical user would be khugepaged collapsing a huge page on a remote -* node. */ #define __GFP_COLD ((__force gfp_t)___GFP_COLD)#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
@@ -184,10 +178,9 @@ struct vm_area_struct;#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK) -#define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE)/* Room for N __GFP_FOO bits */ -#define __GFP_BITS_SHIFT 26 +#define __GFP_BITS_SHIFT 25#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))/*
include/trace/events/mmflags.h +0 -1
@@ -48,7 +48,6 @@ {(unsigned long)__GFP_RECLAIM, "__GFP_RECLAIM"}, \ {(unsigned long)__GFP_DIRECT_RECLAIM, "__GFP_DIRECT_RECLAIM"},\ {(unsigned long)__GFP_KSWAPD_RECLAIM, "__GFP_KSWAPD_RECLAIM"},\ - {(unsigned long)__GFP_OTHER_NODE, "__GFP_OTHER_NODE"} \ #define show_gfp_flags(flags) \ (flags) ? __print_flags(flags, "|", \
mm/huge_memory.c +1 -2
@@ -918,8 +918,7 @@ static int do_huge_pmd_wp_page_fallback(struct vm_fault *vmf, pmd_t orig_pmd, } for (i = 0; i < HPAGE_PMD_NR; i++) { - pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE | - __GFP_OTHER_NODE, vma, + pages[i] = alloc_page_vma_node(GFP_HIGHUSER_MOVABLE, vma, vmf->address, page_to_nid(page)); if (unlikely(!pages[i] || mem_cgroup_try_charge(pages[i], vma->vm_mm,
mm/khugepaged.c +2 -3
@@ -943,7 +943,7 @@ static void collapse_huge_page(struct mm_struct *mm,VM_BUG_ON (address & ~HPAGE_PMD_MASK );/* Only allocate from the target node */ - gfp =alloc_hugepage_khugepaged_gfpmask () | __GFP_OTHER_NODE | __GFP_THISNODE; + gfp =alloc_hugepage_khugepaged_gfpmask () | __GFP_THISNODE;/* * Before allocating the hugepage, release the mmap_sem read lock.
@@ -1326,8 +1326,7 @@ static void collapse_shmem(struct mm_struct *mm,VM_BUG_ON (start & (HPAGE_PMD_NR -1 ));/* Only allocate from the target node */ - gfp =alloc_hugepage_khugepaged_gfpmask () | - __GFP_OTHER_NODE | __GFP_THISNODE; + gfp =alloc_hugepage_khugepaged_gfpmask () | __GFP_THISNODE; new_page =khugepaged_alloc_page (hpage, gfp, node);if (!new_page) {
mm/page_alloc.c +2 -3
@@ -2584,8 +2584,7 @@ int __isolate_free_page(struct page *page, unsigned int order)* * Must be called with interrupts disabled. */ -static inline void zone_statistics (struct zone *preferred_zone,struct zone *z, -gfp_t flags) +static inline void zone_statistics (struct zone *preferred_zone,struct zone *z) {#ifdef CONFIG_NUMA enum zone_stat_item local_stat =NUMA_LOCAL ;
@@ -2667,7 +2666,7 @@ struct page *buffered_rmqueue(struct zone *preferred_zone, }__count_zid_vm_events (PGALLOC ,page_zonenum (page),1 << order); -zone_statistics (preferred_zone, zone, gfp_flags); +zone_statistics (preferred_zone, zone);local_irq_restore (flags);VM_BUG_ON_PAGE (bad_range (zone, page), page);
tools/perf/builtin-kmem.c +0 -1
@@ -645,7 +645,6 @@ static const struct { { "__GFP_RECLAIM", "R" }, { "__GFP_DIRECT_RECLAIM", "DR" }, { "__GFP_KSWAPD_RECLAIM", "KR" }, - { "__GFP_OTHER_NODE", "ON" }, };static size_t max_gfp_len;
Signature · 7 lines
2.11.0 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>