mm: collect the number of anon mTHP
Subsystems (7)
cover mm: collect the number of anon mTHP
From: Barry Song <[email protected]>
Knowing the number of anon mTHPs in the system is crucial for performance analysis. It helps in understanding the ratio and distribution of mTHPs versus small folios throughout the system.
Additionally, partial unmapping by userspace can lead to significant waste of mTHPs over time and increase memory reclamation pressure. We need this information for comprehensive system tuning.
Barry Song (2):
mm: collect the number of anon large folios
mm: collect the number of anon large folios partially unmapped
Documentation/admin-guide/mm/transhuge.rst | 10 ++++++++++ include/linux/huge_mm.h | 16 ++++++++++++++-- mm/huge_memory.c | 8 ++++++++ mm/rmap.c | 3 +++ 4 files changed, 35 insertions(+), 2 deletions(-)
Signature · 1 line
2.34.1
RFC PATCH 1/2 mm: collect the number of anon large folios
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)
Documentation/admin-guide/mm/transhuge.rst +5 -0
@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to help
include/linux/huge_mm.h +13 -2
@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };
@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }
mm/huge_memory.c +2 -0
@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,
@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };
mm/rmap.c +3 -0
@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,
@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped)) +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/* * It would be tidy to reset folio_test_anon mapping when fully
Signature · 1 line
2.34.1
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 102 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +5
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 4 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))
could have a risk here two processes unmap at the same time, so they both get zero on atomic_read(mapped)? should read the value of atomic_dec_return() instead to confirm we are the last one doing unmap?
mm/rmap.c +6 -1
@@ -1530,6 +1530,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,atomic_t *mapped = &folio->_nr_pages_mapped;int last, nr =0 , nr_pmdmapped =0 ;bool partially_mapped = false; +bool last_unmapped = false;__folio_rmap_sanity_checks (folio, page, nr_pages, level);
@@ -1545,6 +1546,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, last =atomic_add_negative (-1 , &page->_mapcount);if (last) { last =atomic_dec_return_relaxed (mapped); +if (!last) + last_unmapped = true;if (last <ENTIRELY_MAPPED ) nr++; }
@@ -1557,6 +1560,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, last =atomic_add_negative (-1 , &folio->_entire_mapcount);if (last) { nr =atomic_sub_return_relaxed (ENTIRELY_MAPPED , mapped); +if (nr ==0 ) + last_unmapped = true;if (likely (nr <ENTIRELY_MAPPED )) { nr_pmdmapped =folio_nr_pages (folio); nr = nr_pmdmapped - (nr &FOLIO_PAGES_MAPPED );
@@ -1583,7 +1588,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); -if (folio_test_anon (folio) && !atomic_read (mapped)) +if (folio_test_anon (folio) && last_unmapped)mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/*
Quoted 7 lines
mm/rmap.c +1
-1+mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/* * It would be tidy to reset folio_test_anon mapping when fully -- 2.34.1
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 101 lines · Barry Song
Quoted 100 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +5
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); }
Quoted 11 lines
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 6 lines
mm/rmap.c +1
-0static __always_inline void __folio_add_file_rmap(struct folio *folio,@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))could have a risk here two processes unmap at the same time, so they both get zero on
atomic_read(mapped)? should read the value ofatomic_dec_return()instead to confirm we are the last one doing unmap?
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio? folio_test_anon() is sticky until freed.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 114 lines · David Hildenbrand
Quoted 101 lines · Barry Song
Quoted 100 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +5
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); }Quoted 11 lines
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 6 lines
mm/rmap.c +1
-0static __always_inline void __folio_add_file_rmap(struct folio *folio,@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))could have a risk here two processes unmap at the same time, so they both get zero on
atomic_read(mapped)? should read the value ofatomic_dec_return()instead to confirm we are the last one doing unmap?
Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.
To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 120 lines · David Hildenbrand
Quoted 114 lines · David Hildenbrand
Quoted 101 lines · Barry Song
Quoted 100 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +5
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); }Quoted 11 lines
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 6 lines
mm/rmap.c +1
-0static __always_inline void __folio_add_file_rmap(struct folio *folio,@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))could have a risk here two processes unmap at the same time, so they both get zero on
atomic_read(mapped)? should read the value ofatomic_dec_return()instead to confirm we are the last one doing unmap?Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.
Quoted 8 lines
To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 126 lines · David Hildenbrand
Quoted 124 lines · David Hildenbrand
Quoted 114 lines · David Hildenbrand
Quoted 101 lines · Barry Song
Quoted 100 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +5
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); }Quoted 11 lines
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 6 lines
mm/rmap.c +1
-0static __always_inline void __folio_add_file_rmap(struct folio *folio,@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))could have a risk here two processes unmap at the same time, so they both get zero on
atomic_read(mapped)? should read the value ofatomic_dec_return()instead to confirm we are the last one doing unmap?Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().
Quoted 8 lines
Quoted 4 lines
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
yes, if we don't use remove_rmap. because we could allocate them as mTHP but free them as nr_pages small folios.
Quoted 6 lines
--
Cheers,David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to David HildenbrandQuoted 9 lines
Quoted 9 lines
Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().
Quoted 8 lines
Quoted 4 lines
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
Hi David,
The conceptual code is shown below. Does this make more sense to you? we have a line "mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));"
@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
struct deferred_split *ds_queue = get_deferred_split_queue(folio);
/* reset xarray order to new order after split */
XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
- struct anon_vma *anon_vma = NULL;
+ bool is_anon = folio_test_anon(folio);
struct address_space *mapping = NULL;
+ struct anon_vma *anon_vma = NULL;
int order = folio_order(folio);
int extra_pins, ret;
pgoff_t end;
@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
if (new_order >= folio_order(folio))
return -EINVAL;
- if (folio_test_anon(folio)) {
+ if (is_anon) {
/* order-1 is not supported for anonymous THP. */
if (new_order == 1) {
VM_WARN_ONCE(1, "Cannot split to order-1 folio");
@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
if (folio_test_writeback(folio))
return -EBUSY;
- if (folio_test_anon(folio)) {
+ if (is_anon) {
/*
* The caller does not necessarily hold an mmap_lock that would
* prevent the anon_vma disappearing so we first we take a
@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
}
}
+ if (is_anon) {
+ mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
+ mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
+ }
__split_huge_page(page, list, end, new_order); ret = 0;
} else {
mm/page_alloc.c +4 -0
@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,bool skip_kasan_poison =should_skip_kasan_poison (page);bool init =want_init_on_free ();bool compound =PageCompound (page); +bool anon =PageAnon (page);VM_BUG_ON_PAGE (PageTail (page), page);
@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,debug_pagealloc_unmap_pages (page,1 << order); +if (anon && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); +return true; }
mm/rmap.c +1 -0
@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 77 lines · Barry Song
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().Quoted 8 lines
Quoted 4 lines
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
Hi David,
The conceptual code is shown below. Does this make more sense to you? we have a line "
mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));"(quoted diff) +8
-3@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, struct deferred_split *ds_queue = get_deferred_split_queue(folio); /* reset xarray order to new order after split */ XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order); - struct anon_vma *anon_vma = NULL; + bool is_anon = folio_test_anon(folio); struct address_space *mapping = NULL; + struct anon_vma *anon_vma = NULL; int order = folio_order(folio); int extra_pins, ret; pgoff_t end;@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (new_order >= folio_order(folio)) return -EINVAL; - if (folio_test_anon(folio)) { + if (is_anon) { /* order-1 is not supported for anonymous THP. */ if (new_order == 1) { VM_WARN_ONCE(1, "Cannot split to order-1 folio");@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (folio_test_writeback(folio)) return -EBUSY; - if (folio_test_anon(folio)) { + if (is_anon) { /* * The caller does not necessarily hold an mmap_lock that would * prevent the anon_vma disappearing so we first we take a@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, } } + if (is_anon) { + mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1); + mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order)); + } __split_huge_page(page, list, end, new_order); ret = 0; } else {mm/page_alloc.c +1
-0@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,bool skip_kasan_poison =should_skip_kasan_poison (page);bool init =want_init_on_free ();bool compound =PageCompound (page); +bool anon =PageAnon (page);VM_BUG_ON_PAGE (PageTail (page), page);
mm/page_alloc.c __always_inline bool free_pages_prepare(struct page *page, · line 1137
Quoted 7 lines
mm/page_alloc.c +3
-0@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,debug_pagealloc_unmap_pages (page,1 << order); +if (anon && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); +return true;
I'd have placed it here, when we are already passed the "PageMappingFlags" check and shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should run into that path).
mm/page_alloc.c +4 -1
@@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page, (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP ; } } -if (PageMappingFlags (page)) +if (PageMappingFlags (page)) { +if (PageAnon (page) && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); page->mapping =NULL ; + }if (is_check_pages_enabled ()) {if (free_page_is_bad (page)) bad++;
Conceptually LGTM. We account an anon folio as long as it's anon, even when still GUP-pinned after unmapping it or when temporarily unmapping+remapping it during migration.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 99 lines · David Hildenbrand
Quoted 77 lines · Barry Song
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().Quoted 8 lines
Quoted 4 lines
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
Hi David,
The conceptual code is shown below. Does this make more sense to you? we have a line "
mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));"(quoted diff) +8
-3@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, struct deferred_split *ds_queue = get_deferred_split_queue(folio); /* reset xarray order to new order after split */ XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order); - struct anon_vma *anon_vma = NULL; + bool is_anon = folio_test_anon(folio); struct address_space *mapping = NULL; + struct anon_vma *anon_vma = NULL; int order = folio_order(folio); int extra_pins, ret; pgoff_t end;@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (new_order >= folio_order(folio)) return -EINVAL; - if (folio_test_anon(folio)) { + if (is_anon) { /* order-1 is not supported for anonymous THP. */ if (new_order == 1) { VM_WARN_ONCE(1, "Cannot split to order-1 folio");@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (folio_test_writeback(folio)) return -EBUSY; - if (folio_test_anon(folio)) { + if (is_anon) { /* * The caller does not necessarily hold an mmap_lock that would * prevent the anon_vma disappearing so we first we take a@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, } } + if (is_anon) { + mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1); + mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order)); + } __split_huge_page(page, list, end, new_order); ret = 0; } else {mm/page_alloc.c +1
-0@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,bool skip_kasan_poison =should_skip_kasan_poison (page);bool init =want_init_on_free ();bool compound =PageCompound (page); +bool anon =PageAnon (page);VM_BUG_ON_PAGE (PageTail (page), page);mm/page_alloc.c __always_inline bool free_pages_prepare(struct page *page, · line 1137
Quoted 7 lines
mm/page_alloc.c +3
-0@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,debug_pagealloc_unmap_pages (page,1 << order); +if (anon && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); +return true;I'd have placed it here, when we are already passed the "PageMappingFlags" check and shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should run into that path).
mm/page_alloc.c +0
-0@@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page, (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP ; }
mm/page_alloc.c __always_inline bool free_pages_prepare(struct page *page, · line 1089
Quoted 10 lines
mm/page_alloc.c +4
-1} -if (PageMappingFlags (page)) +if (PageMappingFlags (page)) { +if (PageAnon (page) && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); page->mapping =NULL ; + }if (is_check_pages_enabled ()) {if (free_page_is_bad (page)) bad++;
This looks better, but we're not concerned about the bad pages, correct? For bad pages, we're reducing the count by 1, but they aren't actually freed. We don't need to worry about this since it's already considered a bug, right?
Quoted 3 lines
Conceptually LGTM. We account an anon folio as long as it's anon, even when still GUP-pinned after unmapping it or when temporarily unmapping+remapping it during migration.
right. but migration might be a problem? as the dst folio doesn't call folio_add_new_anon_rmap(), dst->mapping is copied from src. So I suspect we need the below (otherwise, src has been put and got -1, but dst won't get +1),
mm/migrate.c +2 -1
@@ -1102,6 +1102,8 @@ static void migrate_folio_done(struct folio *src,mod_node_page_state (folio_pgdat (src),NR_ISOLATED_ANON +folio_is_file_lru (src), -folio_nr_pages (src)); + mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1); +if (reason !=MR_MEMORY_FAILURE )/* We release the page in page_handle_poison. */
folio_put(src);
Quoted 6 lines
--
Cheers,David / dhildenb
Thanks
Barry
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 132 lines · Barry Song
Quoted 99 lines · David Hildenbrand
Quoted 77 lines · Barry Song
Quoted 9 lines
Quoted 9 lines
Quoted 4 lines
I would appreciate if we leave the rmap out here.
Can't we handle that when actually freeing the folio?
folio_test_anon()is sticky until freed.To be clearer: we increment the counter when we set a folio anon, which should indeed only happen in
folio_add_new_anon_rmap(). We'll have to ignore hugetlb here where we do it inhugetlb_add_new_anon_rmap().Quoted 8 lines
Quoted 4 lines
Then, when we free an anon folio we decrement the counter. (hugetlb should clear the anon flag when an anon folio gets freed back to its allocator -- likely that is already done).
Sorry that I am talking to myself: I'm wondering if we also have to adjust the counter when splitting a large folio to multiple smaller-but-still-large folios.
Hi David,
The conceptual code is shown below. Does this make more sense to you? we have a line "
mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));"(quoted diff) +8
-3@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, struct deferred_split *ds_queue = get_deferred_split_queue(folio); /* reset xarray order to new order after split */ XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order); - struct anon_vma *anon_vma = NULL; + bool is_anon = folio_test_anon(folio); struct address_space *mapping = NULL; + struct anon_vma *anon_vma = NULL; int order = folio_order(folio); int extra_pins, ret; pgoff_t end;@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (new_order >= folio_order(folio)) return -EINVAL; - if (folio_test_anon(folio)) { + if (is_anon) { /* order-1 is not supported for anonymous THP. */ if (new_order == 1) { VM_WARN_ONCE(1, "Cannot split to order-1 folio");@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, if (folio_test_writeback(folio)) return -EBUSY; - if (folio_test_anon(folio)) { + if (is_anon) { /* * The caller does not necessarily hold an mmap_lock that would * prevent the anon_vma disappearing so we first we take a@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list, } } + if (is_anon) { + mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1); + mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order)); + } __split_huge_page(page, list, end, new_order); ret = 0; } else {mm/page_alloc.c +1
-0@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,bool skip_kasan_poison =should_skip_kasan_poison (page);bool init =want_init_on_free ();bool compound =PageCompound (page); +bool anon =PageAnon (page);VM_BUG_ON_PAGE (PageTail (page), page);mm/page_alloc.c __always_inline bool free_pages_prepare(struct page *page, · line 1137
Quoted 7 lines
mm/page_alloc.c +3
-0@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,debug_pagealloc_unmap_pages (page,1 << order); +if (anon && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); +return true;I'd have placed it here, when we are already passed the "PageMappingFlags" check and shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should run into that path).
mm/page_alloc.c +0
-0@@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page, (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP ; }mm/page_alloc.c __always_inline bool free_pages_prepare(struct page *page, · line 1089
Quoted 10 lines
mm/page_alloc.c +4
-1} -if (PageMappingFlags (page)) +if (PageMappingFlags (page)) { +if (PageAnon (page) && compound) +mod_mthp_stat (order,MTHP_STAT_NR_ANON ,-1 ); page->mapping =NULL ; + }if (is_check_pages_enabled ()) {if (free_page_is_bad (page)) bad++;This looks better, but we're not concerned about the bad pages, correct? For bad pages, we're reducing the count by 1, but they aren't actually freed. We don't need to worry about this since it's already considered a bug, right?
Quoted 3 lines
Conceptually LGTM. We account an anon folio as long as it's anon, even when still GUP-pinned after unmapping it or when temporarily unmapping+remapping it during migration.
right. but migration might be a problem? as the dst folio doesn't call
folio_add_new_anon_rmap(),dst->mappingis copied from src. So I suspect we need the below (otherwise, src has been put and got -1, but dst won't get +1),mm/migrate.c +0
-0
mm/migrate.c static void migrate_folio_done(struct folio *src,
Quoted 9 lines
mm/migrate.c +2
-1@@ -1102,6 +1102,8 @@ static void migrate_folio_done(struct folio *src,mod_node_page_state (folio_pgdat (src),NR_ISOLATED_ANON +folio_is_file_lru (src), -folio_nr_pages (src)); + mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1); +if (reason !=MR_MEMORY_FAILURE )/* We release the page in page_handle_poison. */ folio_put (src);
sorry. wrong place as migrate_folio_done() doesn't necessarily mean migration is done. it could be "Folio was freed from under us"
if (folio_ref_count(src) == 1) {
/* Folio was freed from under us. So we are done. */
folio_clear_active(src);
folio_clear_unevictable(src);
/* free_pages_prepare() will clear PG_isolated. */
list_del(&src->lru);
migrate_folio_done(src, reason);
return MIGRATEPAGE_SUCCESS;
}the correct place should be:
@@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t put_new_folio, unsigned long private,
if (anon_vma)
put_anon_vma(anon_vma);
folio_unlock(src);+
+ if (folio_test_anon(src))
+ mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
+
migrate_folio_done(src, reason);
return rc;Without this modification in migration code, my tests fail, anon_num can become negative.
Quoted 7 lines
Quoted 7 lines
--
Cheers,David / dhildenb
Thanks
Barry
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 15 lines
the correct place should be:
(quoted diff) +4
-0@@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t put_new_folio, unsigned long private, if (anon_vma) put_anon_vma(anon_vma); folio_unlock(src); + + if (folio_test_anon(src)) + mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1); + migrate_folio_done(src, reason);return rc;
Quoted 2 lines
Without this modification in migration code, my tests fail,
anon_numcan become negative.
I was wondering if we should do it in __folio_migrate_mapping().
There, we set newfolio->mapping.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 14 lines · David Hildenbrand
Quoted 14 lines
the correct place should be:
(quoted diff) +4
-0@@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t put_new_folio, unsigned long private, if (anon_vma) put_anon_vma(anon_vma); folio_unlock(src); + + if (folio_test_anon(src)) + mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1); + migrate_folio_done(src, reason);
Quoted 8 lines
Quoted 4 lines
return rc;Without this modification in migration code, my tests fail,
anon_numcan become negative.I was wondering if we should do it in
__folio_migrate_mapping().There, we set
newfolio->mapping.
Correct! To handle cases where the destination might not be migrated to but could be put or freed, migrate_folio_undo_dst() will be called to release the destination earlier. It's better to simply place the count where the anon mapping
is set and cleared.
mm/migrate.c +0 -0
@@ -423,6 +423,8 @@ static int __folio_migrate_mapping(struct
address_space *mapping,
/* No turning back from here */
newfolio->index = folio->index;
newfolio->mapping = folio->mapping;
+ if (folio_test_anon(folio) && folio_test_large(folio))
+ mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
if (folio_test_swapbacked(folio))
__folio_set_swapbacked(newfolio);@@ -446,6 +448,9 @@ static int __folio_migrate_mapping(struct address_space *mapping,
* no turning back from here.
*/
newfolio->index = folio->index;
+
+ if (folio_test_anon(folio) && folio_test_large(folio))
+ mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
newfolio->mapping = folio->mapping;
folio_ref_add(newfolio, nr); /* add cache reference */
if (folio_test_swapbacked(folio)) {Quoted 6 lines
--
Cheers,David / dhildenb
kernel test robot
build botRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to Barry Songkernel test robot noticed the following build errors:
Hi Barry,
[This is a private test report for your RFC patch.] kernel test robot noticed the following build errors:
[auto build test ERROR on next-20240807]
[also build test ERROR on v6.11-rc2]
[cannot apply to akpm-mm/mm-everything linus/master v6.11-rc2 v6.11-rc1 v6.10] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Barry-Song/mm-collect-the-number-of-anon-large-folios/20240808-090627 base: next-20240807 patch link: https://lore.kernel.org/r/20240808010457.228753-2-21cnbao%40gmail.com patch subject: [PATCH RFC 1/2] mm: collect the number of anon large folios config: openrisc-allnoconfig (https://download.01.org/0day-ci/archive/20240809/[email protected]/config) compiler: or1k-linux-gcc (GCC) 14.1.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240809/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
mm/rmap.c: In function 'folio_add_new_anon_rmap':
>> mm/rmap.c:1470:9: error: implicit declaration of function 'mod_mthp_stat'; did you mean 'mod_memcg_state'? [-Wimplicit-function-declaration]
1470 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
| ^~~~~~~~~~~~~
| mod_memcg_state
>> mm/rmap.c:1470:43: error: 'MTHP_STAT_NR_ANON' undeclared (first use in this function)
1470 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
| ^~~~~~~~~~~~~~~~~
mm/rmap.c:1470:43: note: each undeclared identifier is reported only once for each function it appears in
mm/rmap.c: In function '__folio_remove_rmap':
mm/rmap.c:1587:51: error: 'MTHP_STAT_NR_ANON' undeclared (first use in this function)
1587 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
| ^~~~~~~~~~~~~~~~~vim +1470 mm/rmap.c
1402
1403 /**
1404 * folio_add_new_anon_rmap - Add mapping to a new anonymous folio.
1405 * @folio: The folio to add the mapping to.
1406 * @vma: the vm area in which the mapping is added
1407 * @address: the user virtual address mapped
1408 * @flags: The rmap flags
1409 *
1410 * Like folio_add_anon_rmap_*() but must only be called on *new* folios.
1411 * This means the inc-and-test can be bypassed.
1412 * The folio doesn't necessarily need to be locked while it's exclusive
1413 * unless two threads map it concurrently. However, the folio must be
1414 * locked if it's shared.
1415 *
1416 * If the folio is pmd-mappable, it is accounted as a THP.
1417 */
1418 void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
1419 unsigned long address, rmap_t flags)
1420 {
1421 const int nr = folio_nr_pages(folio);
1422 const bool exclusive = flags & RMAP_EXCLUSIVE;
1423 int nr_pmdmapped = 0;
1424
1425 VM_WARN_ON_FOLIO(folio_test_hugetlb(folio), folio);
1426 VM_WARN_ON_FOLIO(!exclusive && !folio_test_locked(folio), folio);
1427 VM_BUG_ON_VMA(address < vma->vm_start ||
1428 address + (nr << PAGE_SHIFT) > vma->vm_end, vma);
1429
1430 /*
1431 * VM_DROPPABLE mappings don't swap; instead they're just dropped when
1432 * under memory pressure.
1433 */
1434 if (!folio_test_swapbacked(folio) && !(vma->vm_flags & VM_DROPPABLE))
1435 __folio_set_swapbacked(folio);
1436 __folio_set_anon(folio, vma, address, exclusive);
1437
1438 if (likely(!folio_test_large(folio))) {
1439 /* increment count (starts at -1) */
1440 atomic_set(&folio->_mapcount, 0);
1441 if (exclusive)
1442 SetPageAnonExclusive(&folio->page);
1443 } else if (!folio_test_pmd_mappable(folio)) {
1444 int i;
1445
1446 for (i = 0; i < nr; i++) {
1447 struct page *page = folio_page(folio, i);
1448
1449 /* increment count (starts at -1) */
1450 atomic_set(&page->_mapcount, 0);
1451 if (exclusive)
1452 SetPageAnonExclusive(page);
1453 }
1454
1455 /* increment count (starts at -1) */
1456 atomic_set(&folio->_large_mapcount, nr - 1);
1457 atomic_set(&folio->_nr_pages_mapped, nr);
1458 } else {
1459 /* increment count (starts at -1) */
1460 atomic_set(&folio->_entire_mapcount, 0);
1461 /* increment count (starts at -1) */
1462 atomic_set(&folio->_large_mapcount, 0);
1463 atomic_set(&folio->_nr_pages_mapped, ENTIRELY_MAPPED);
1464 if (exclusive)
1465 SetPageAnonExclusive(&folio->page);
1466 nr_pmdmapped = nr;
1467 }
1468
1469 __folio_mod_stat(folio, nr, nr_pmdmapped);
> 1470 mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
1471 }
1472
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
kernel test robot
build botRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to Barry Songkernel test robot noticed the following build errors:
Hi Barry,
[This is a private test report for your RFC patch.] kernel test robot noticed the following build errors:
[auto build test ERROR on next-20240807]
[also build test ERROR on v6.11-rc2]
[cannot apply to akpm-mm/mm-everything linus/master v6.11-rc2 v6.11-rc1 v6.10] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Barry-Song/mm-collect-the-number-of-anon-large-folios/20240808-090627 base: next-20240807 patch link: https://lore.kernel.org/r/20240808010457.228753-2-21cnbao%40gmail.com patch subject: [PATCH RFC 1/2] mm: collect the number of anon large folios config: s390-allnoconfig (https://download.01.org/0day-ci/archive/20240809/[email protected]/config) compiler: clang version 20.0.0git (https://github.com/llvm/llvm-project f86594788ce93b696675c94f54016d27a6c21d18) reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240809/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
In file included from mm/rmap.c:56:
In file included from include/linux/mm.h:2199:
include/linux/vmstat.h:514:36: warning: arithmetic between different enumeration types ('enum node_stat_item' and 'enum lru_list') [-Wenum-enum-conversion]
514 | return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_"
| ~~~~~~~~~~~ ^ ~~~
In file included from mm/rmap.c:77:
include/linux/mm_inline.h:47:41: warning: arithmetic between different enumeration types ('enum node_stat_item' and 'enum lru_list') [-Wenum-enum-conversion]
47 | __mod_lruvec_state(lruvec, NR_LRU_BASE + lru, nr_pages);
| ~~~~~~~~~~~ ^ ~~~
include/linux/mm_inline.h:49:22: warning: arithmetic between different enumeration types ('enum zone_stat_item' and 'enum lru_list') [-Wenum-enum-conversion]
49 | NR_ZONE_LRU_BASE + lru, nr_pages);
| ~~~~~~~~~~~~~~~~ ^ ~~~
>> mm/rmap.c:1470:2: error: call to undeclared function 'mod_mthp_stat'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
1470 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
| ^
>> mm/rmap.c:1470:36: error: use of undeclared identifier 'MTHP_STAT_NR_ANON'
1470 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
| ^
mm/rmap.c:1587:3: error: call to undeclared function 'mod_mthp_stat'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
1587 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
| ^
mm/rmap.c:1587:37: error: use of undeclared identifier 'MTHP_STAT_NR_ANON'
1587 | mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
| ^3 warnings and 4 errors generated.
vim +/mod_mthp_stat +1470 mm/rmap.c
1402
1403 /**
1404 * folio_add_new_anon_rmap - Add mapping to a new anonymous folio.
1405 * @folio: The folio to add the mapping to.
1406 * @vma: the vm area in which the mapping is added
1407 * @address: the user virtual address mapped
1408 * @flags: The rmap flags
1409 *
1410 * Like folio_add_anon_rmap_*() but must only be called on *new* folios.
1411 * This means the inc-and-test can be bypassed.
1412 * The folio doesn't necessarily need to be locked while it's exclusive
1413 * unless two threads map it concurrently. However, the folio must be
1414 * locked if it's shared.
1415 *
1416 * If the folio is pmd-mappable, it is accounted as a THP.
1417 */
1418 void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
1419 unsigned long address, rmap_t flags)
1420 {
1421 const int nr = folio_nr_pages(folio);
1422 const bool exclusive = flags & RMAP_EXCLUSIVE;
1423 int nr_pmdmapped = 0;
1424
1425 VM_WARN_ON_FOLIO(folio_test_hugetlb(folio), folio);
1426 VM_WARN_ON_FOLIO(!exclusive && !folio_test_locked(folio), folio);
1427 VM_BUG_ON_VMA(address < vma->vm_start ||
1428 address + (nr << PAGE_SHIFT) > vma->vm_end, vma);
1429
1430 /*
1431 * VM_DROPPABLE mappings don't swap; instead they're just dropped when
1432 * under memory pressure.
1433 */
1434 if (!folio_test_swapbacked(folio) && !(vma->vm_flags & VM_DROPPABLE))
1435 __folio_set_swapbacked(folio);
1436 __folio_set_anon(folio, vma, address, exclusive);
1437
1438 if (likely(!folio_test_large(folio))) {
1439 /* increment count (starts at -1) */
1440 atomic_set(&folio->_mapcount, 0);
1441 if (exclusive)
1442 SetPageAnonExclusive(&folio->page);
1443 } else if (!folio_test_pmd_mappable(folio)) {
1444 int i;
1445
1446 for (i = 0; i < nr; i++) {
1447 struct page *page = folio_page(folio, i);
1448
1449 /* increment count (starts at -1) */
1450 atomic_set(&page->_mapcount, 0);
1451 if (exclusive)
1452 SetPageAnonExclusive(page);
1453 }
1454
1455 /* increment count (starts at -1) */
1456 atomic_set(&folio->_large_mapcount, nr - 1);
1457 atomic_set(&folio->_nr_pages_mapped, nr);
1458 } else {
1459 /* increment count (starts at -1) */
1460 atomic_set(&folio->_entire_mapcount, 0);
1461 /* increment count (starts at -1) */
1462 atomic_set(&folio->_large_mapcount, 0);
1463 atomic_set(&folio->_nr_pages_mapped, ENTIRELY_MAPPED);
1464 if (exclusive)
1465 SetPageAnonExclusive(&folio->page);
1466 nr_pmdmapped = nr;
1467 }
1468
1469 __folio_mod_stat(folio, nr, nr_pmdmapped);
> 1470 mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
1471 }
1472
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
Ryan Roberts
reviewerRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to Barry SongQuoted 18 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0
Documentation/admin-guide/mm/transhuge.rst split_deferred · line 533
Quoted 7 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages.
nit: "entirely mapped and have partially unmapped and unused subpages" -> "entirely mapped or have partially unmapped/unused subpages"
Quoted 76 lines
Documentation/admin-guide/mm/transhuge.rst +1
-0+ As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 4 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))
Agree that atomic_read() is dodgy here.
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Quoted 4 lines
mm/rmap.c +1
-0+mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/* * It would be tidy to reset folio_test_anon mapping when fully
Ryan Roberts
reviewerRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 74 lines · Ryan Roberts
Quoted 18 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0Documentation/admin-guide/mm/transhuge.rst split_deferred · line 533
Quoted 7 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages.nit: "entirely mapped and have partially unmapped and unused subpages" -> "entirely mapped or have partially unmapped/unused subpages"
Quoted 44 lines
Documentation/admin-guide/mm/transhuge.rst +1
-0+ As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +0
-0
Quoted 8 lines
mm/huge_memory.c DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE); · line 581
Quoted 8 lines
index
697fcf89f975..b6bc2a3791e3100644mm/huge_memory.c +1
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );
Why are the user-facing and internal names different? Perhaps it would be clearer to call this nr_anon in sysfs?
Quoted 31 lines
Quoted 24 lines
mm/huge_memory.c +1
-0static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 4 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))Agree that
atomic_read()is dodgy here.
Quoted 7 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in rmap_unmap to be consistent for all?
Quoted 6 lines
Quoted 5 lines
mm/rmap.c +1
-0+mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/* * It would be tidy to reset folio_test_anon mapping when fully
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · Ryan Roberts
Quoted 18 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0Documentation/admin-guide/mm/transhuge.rst split_deferred · line 533
Quoted 7 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages.nit: "entirely mapped and have partially unmapped and unused subpages" -> "entirely mapped or have partially unmapped/unused subpages"
Quoted 47 lines
Documentation/admin-guide/mm/transhuge.rst +1
-0+ As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +0
-0
Quoted 44 lines
Quoted 5 lines
mm/huge_memory.c DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE); · line 581
Quoted 5 lines
mm/huge_memory.c +1
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );Why are the user-facing and internal names different? Perhaps it would be clearer to call this
nr_anonin sysfs?Quoted 35 lines
Quoted 24 lines
mm/huge_memory.c +1
-0static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 4 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))Agree that
atomic_read()is dodgy here.Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
As long as we can account for mTHP when clearing the Anon flag for the folio, we should be safe. It’s challenging to add +1 when allocating a large folio because we don’t know its intended use—it could be for file, anon, or shmem.
Quoted 8 lines
Quoted 3 lines
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?
I've been struggling quite a bit with rmap. Despite trying various approaches, I’m
still occasionally seeing a negative mTHP counter after running it some hours on phones. It seems that rmap is really tricky to handle. I admit that I have failed on rmap :-)
On the other hand, for anon folios, we have cases where they are split from order M to order N. So, we add +1 when a new anon folio is added to rmap and subtract -1 when we either split it or free it. This approach seems clearer to me. When we split from order M to order N, we can add 1 << (M - N) for order N.
Quoted 7 lines
Quoted 7 lines
mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1590
Quoted 7 lines
mm/rmap.c +1
-0+mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,-1 );/* * It would be tidy to reset folio_test_anon mapping when fully
Thanks
Barry
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to Ryan RobertsQuoted 5 lines
Quoted 5 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
Quoted 7 lines
Quoted 2 lines
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?
Again, see NR_FILE_THPS handling. No rmap over-complication please.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 8 lines · David Hildenbrand
Quoted 8 lines
Quoted 7 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Quoted 6 lines
Quoted 4 lines
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?Again, see
NR_FILE_THPShandling. No rmap over-complication please.
... not to mention that it is non-sensical to only count pageache folios that are mapped to user space ;)
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 14 lines · David Hildenbrand
Quoted 13 lines · David Hildenbrand
Quoted 12 lines
Quoted 7 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?
Quoted 4 lines
Quoted 1 line
Again, see
NR_FILE_THPShandling. No rmap over-complication please.... not to mention that it is non-sensical to only count pageache folios that are mapped to user space ;)
Yes, good point. I'll get back in my box. :)
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 15 lines · Ryan Roberts
Quoted 14 lines · David Hildenbrand
Quoted 13 lines · David Hildenbrand
Quoted 12 lines
Quoted 7 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Again, see
NR_FILE_THPShandling. No rmap over-complication please.... not to mention that it is non-sensical to only count pageache folios that are mapped to user space ;)
Yes, good point. I'll get back in my box. :)
Well, it was a valuable discussion!
anon folios in the swapcache are interesting: they are only "anon" after we first mapped them (harder to change, but would be possible by using a NULL mapping maybe, if really worth it; with memdesc that might turn out interesting). But once they are anon, they will stay anon until actually reclaimed -> freed.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 24 lines · David Hildenbrand
Quoted 15 lines · Ryan Roberts
Quoted 14 lines · David Hildenbrand
Quoted 13 lines · David Hildenbrand
Quoted 12 lines
Quoted 7 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Additionally, if we wanted to extend (eventually) to track the number of shmem and file mthps in additional counters, could we also account using similar folio free-time hooks? If not, it might be an argument to account in
rmap_unmapto be consistent for all?Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Again, see
NR_FILE_THPShandling. No rmap over-complication please.... not to mention that it is non-sensical to only count pageache folios that are mapped to user space ;)
Yes, good point. I'll get back in my box. :)
Quoted 7 lines
Well, it was a valuable discussion!
anon folios in the swapcache are interesting: they are only "anon" after we first mapped them (harder to change, but would be possible by using a NULL mapping maybe, if really worth it; with memdesc that might turn out interesting). But once they are anon, they will stay anon until actually reclaimed -> freed.
I assume we don’t need to worry about this, as even AnonPages (NR_ANON_MAPPED) in /proc/meminfo also entirely depends on anon pages becoming anon mappings.
Quoted 6 lines
--
Cheers,David / dhildenb
Thanks
Barry
Re: [PATCH RFC 1/2] mm: collect the number of anon large folios
In reply to Ryan RobertsQuoted 113 lines · Ryan Roberts
Quoted 18 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++ 4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0Documentation/admin-guide/mm/transhuge.rst split_deferred · line 533
Quoted 7 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0@@ -527,6 +527,11 @@ split_deferred it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num + the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages.nit: "entirely mapped and have partially unmapped and unused subpages" -> "entirely mapped or have partially unmapped/unused subpages"
Quoted 76 lines
Documentation/admin-guide/mm/transhuge.rst +1
-0+ As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +13
-2@@ -281,6 +281,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED , +MTHP_STAT_NR_ANON , __MTHP_STAT_COUNT };@@ -291,14 +292,24 @@ struct mthp_stat {#ifdef CONFIG_SYSFS DECLARE_PER_CPU(struct mthp_stat, mthp_stats); -static inline void count_mthp_stat (int order,enum mthp_stat_item item) +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) {if (order <=0 || order >PMD_ORDER )return ; -this_cpu_inc (mthp_stats.stats[order][item]); +this_cpu_add (mthp_stats.stats[order][item], delta); +} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ +mod_mthp_stat (order, item,1 ); } +#else +static inline void mod_mthp_stat (int order,enum mthp_stat_item item,int delta) +{ +} +static inline void count_mthp_stat (int order,enum mthp_stat_item item) { }mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED ); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma, }__folio_mod_stat (folio, nr, nr_pmdmapped); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON ,1 ); } static __always_inline void __folio_add_file_rmap(struct folio *folio,mm/rmap.c static __always_inline void __folio_remove_rmap(struct folio *folio, · line 1586
Quoted 4 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, list_empty(&folio->_deferred_list))deferred_split_folio (folio);__folio_mod_stat (folio, -nr, -nr_pmdmapped); +if (folio_test_anon (folio) && !atomic_read (mapped))Agree that
atomic_read()is dodgy here.
Quoted 4 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
Doing it when we set/clear folio->mapping is straight forward.
Anon folios currently come to live when we first map them, and they stay that way until we free them.
In the future, we'll have to move that anon handling further out, when if have to allocate anon-specific memdesc ahead of time, then, it will be clued to that lifetime.
Quoted 3 lines
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Likely the deconstructor could work as well. Not sure if that is any better than the freeing path where folio->mapping currently gets cleared.
The generic constructor certainly won't work right now. That's not where the "anon" part comes to live.
Let's take a look how NR_FILE_THPS is handled:
__filemap_add_folio() increments it -- when we set folio->mapping__filemap_remove_folio() (->filemap_unaccount_folio) decrements it -- after which we usually call page_cache_delete() to set folio->mapping = NULL;
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerRe: [PATCH RFC 1/2] mm: collect the number of anon large folios
Quoted 133 lines · David Hildenbrand
Quoted 111 lines · Ryan Roberts
Quoted 19 lines · Barry Song
From: Barry Song <[email protected]>
When a new anonymous mTHP is added to the rmap, we increase the count. We reduce the count whenever an mTHP is completely unmapped.
Signed-off-by Barry Song <[email protected]>---
Documentation/admin-guide/mm/transhuge.rst| 5 +++++ include/linux/huge_mm.h | 15 +++++++++++++-- mm/huge_memory.c | 2 ++ mm/rmap.c | 3 +++4 files changed, 23 insertions(+), 2 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0@@ -527,6 +527,11 @@ split_deferredQuoted 6 lines
it would free up some memory. Pages on split queue are going to be split under memory pressure, if splitting is possible. +anon_num
+ the number of anon huge pages we have in the whole system. + These huge pages could be still entirely mapped and have partially + unmapped and unused subpages.nit: "entirely mapped and have partially unmapped and unused subpages" -> "entirely mapped or have partially unmapped/unused subpages"
Quoted 73 lines
+
As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +1
-0@@ -281,6 +281,7 @@ enum mthp_stat_item { MTHP_STAT_SPLIT, MTHP_STAT_SPLIT_FAILED, MTHP_STAT_SPLIT_DEFERRED, + MTHP_STAT_NR_ANON,
__MTHP_STAT_COUNT
};
@@ -291,14 +292,24 @@ structmthp_stat{
#ifdefCONFIG_SYSFS
DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
-static inline voidcount_mthp_stat(int order, enum mthp_stat_item item)+static inline voidmod_mthp_stat(int order, enum mthp_stat_item item, int delta)
{
if (order <= 0 || order >PMD_ORDER)
return;
-this_cpu_inc(mthp_stats.stats[order][item]);
+this_cpu_add(mthp_stats.stats[order][item], delta);include/linux/huge_mm.h +5
-0+} + +static inline void count_mthp_stat (int order,enum mthp_stat_item item) +{ + mod_mthp_stat(order, item, 1);}
+
#else
+static inline voidmod_mthp_stat(int order, enum mthp_stat_item item, int delta)include/linux/huge_mm.h +3
-0+{ +} +static inline void
count_mthp_stat(int order, enum mthp_stat_item item){
}mm/huge_memory.c +2
-0@@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE); DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT); DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED); DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED); +DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON ); static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = { &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr, + &anon_num_attr.attr, NULL, };diff --git
a/mm/rmap.cb/mm/rmap.c
index 901950200957..2b722f26224c 100644mm/rmap.c +1
-0@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio,struct vm_area_struct *vma, } __folio_mod_stat(folio, nr, nr_pmdmapped); + mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1); } static __always_inline void __folio_add_file_rmap(struct folio *folio,mm/rmap.c static __always_inline void __folio_remove_rmap(struct · line 1587
Quoted 5 lines
mm/rmap.c +1
-0@@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(structfolio *folio, list_empty(&folio->_deferred_list)) deferred_split_folio(folio); __folio_mod_stat(folio, -nr, -nr_pmdmapped); + if (folio_test_anon(folio) && !atomic_read(mapped))Agree that
atomic_read()is dodgy here.Quoted 4 lines
Not sure I fully understand why David prefers to do the unaccounting at free-time though? It feels unbalanced to me to increment when first mapped but decrement when freed. Surely its safer to either use alloc/free or use first map/last map?
Doing it when we set/clear
folio->mappingis straight forward.Anon folios currently come to live when we first map them, and they stay that way until we free them.
In the future, we'll have to move that anon handling further out, when if have to allocate anon-specific memdesc ahead of time, then, it will be clued to that lifetime.
Quoted 3 lines
If using alloc/free isn't there a THP constructor/destructor that prepares the deferred list? (My memory may be failing me). Could we use that?
Likely the deconstructor could work as well. Not sure if that is any better than the freeing path where
folio->mappingcurrently gets cleared.
Quoted 9 lines
The generic constructor certainly won't work right now. That's not where the "anon" part comes to live.
Let's take a look how
NR_FILE_THPSis handled:
__filemap_add_folio()increments it -- when we setfolio->mapping__filemap_remove_folio()(->filemap_unaccount_folio) decrements it -- after which we usually callpage_cache_delete()to setfolio->mapping= NULL;
OK got it, thanks!
RFC PATCH 2/2 mm: collect the number of anon large folios partially unmapped
In reply to Barry SongFrom: Barry Song <[email protected]>
When an mTHP is added to the deferred_list, its partial pages are unused, leading to wasted memory and potentially increasing memory reclamation pressure. Tracking this number indicates the extent to which userspace is partially unmapping mTHPs.
Detailing the specifics of how unmapping occurs is quite difficult and not that useful, so we adopt a simple approach: each time an mTHP enters the deferred_list, we increment the count by 1; whenever it leaves for any reason, we decrement the count by 1.
--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 1 + mm/huge_memory.c | 6 ++++++ 3 files changed, 12 insertions(+)
Documentation/admin-guide/mm/transhuge.rst +5 -0
@@ -532,6 +532,11 @@ anon_num These huge pages could be still entirely mapped and have partially unmapped and unused subpages. +anon_num_partial_unused + the number of anon huge pages which have been partially unmapped + we have in the whole system. These unmapped subpages are also + unused and temporarily wasting memory. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to help
include/linux/huge_mm.h +1 -0
@@ -282,6 +282,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED ,MTHP_STAT_NR_ANON , +MTHP_STAT_NR_ANON_SPLIT_DEFERRED , __MTHP_STAT_COUNT };
mm/huge_memory.c +6 -0
@@ -579,6 +579,7 @@ DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON ); +DEFINE_MTHP_STAT_ATTR (anon_num_partial_unused,MTHP_STAT_NR_ANON_SPLIT_DEFERRED );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,
@@ -593,6 +594,7 @@ static struct attribute *stats_attrs[] = { &split_failed_attr.attr, &split_deferred_attr.attr, &anon_num_attr.attr, + &anon_num_partial_unused_attr.attr, NULL, };
@@ -3229,6 +3231,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,if (folio_order (folio) >1 && !list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );/* * Reinitialize page_deferred_list after removing the * page from the split_queue, otherwise a subsequent
@@ -3291,6 +3294,7 @@ void __folio_undo_large_rmappable(struct folio *folio)spin_lock_irqsave (&ds_queue->split_queue_lock, flags);if (!list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); }spin_unlock_irqrestore (&ds_queue->split_queue_lock, flags);
@@ -3332,6 +3336,7 @@ void deferred_split_folio(struct folio *folio)if (folio_test_pmd_mappable (folio))count_vm_event (THP_DEFERRED_SPLIT_PAGE );count_mthp_stat (folio_order (folio),MTHP_STAT_SPLIT_DEFERRED ); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,1 );list_add_tail (&folio->_deferred_list, &ds_queue->split_queue); ds_queue->split_queue_len++;#ifdef CONFIG_MEMCG
@@ -3379,6 +3384,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,list_move (&folio->_deferred_list, &list); }else {/* We lost race with folio_put() */ +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); ds_queue->split_queue_len--; }
Signature · 1 line
2.34.1
Ryan Roberts
reviewerRe: [PATCH RFC 2/2] mm: collect the number of anon large folios partially unmapped
Quoted 24 lines · Barry Song
From: Barry Song <[email protected]>
When an mTHP is added to the
deferred_list, its partial pages are unused, leading to wasted memory and potentially increasing memory reclamation pressure. Tracking this number indicates the extent to which userspace is partially unmapping mTHPs.Detailing the specifics of how unmapping occurs is quite difficult and not that useful, so we adopt a simple approach: each time an mTHP enters the
deferred_list, we increment the count by 1; whenever it leaves for any reason, we decrement the count by 1.Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 1 + mm/huge_memory.c | 6 ++++++ 3 files changed, 12 insertions(+)Documentation/admin-guide/mm/transhuge.rst +0
-0
Documentation/admin-guide/mm/transhuge.rst anon_num · line 535
Quoted 4 lines
Documentation/admin-guide/mm/transhuge.rst +1
-0@@ -532,6 +532,11 @@ anon_num These huge pages could be still entirely mapped and have partially unmapped and unused subpages. +anon_num_partial_unused
Why is the user-exposed name completely different to the internal (MTHP_STAT_NR_ANON_SPLIT_DEFERRED) name?
Quoted 12 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0+ the number of anon huge pages which have been partially unmapped + we have in the whole system. These unmapped subpages are also + unused and temporarily wasting memory. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +0
-0
include/linux/huge_mm.h enum mthp_stat_item { · line 285
Quoted 4 lines
include/linux/huge_mm.h +1
-0@@ -282,6 +282,7 @@ enum mthp_stat_item {MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED ,MTHP_STAT_NR_ANON , + MTHP_STAT_NR_ANON_SPLIT_DEFERRED,
So the existing MTHP_STAT_SPLIT_DEFERRED is counting all folios that were ever put on the list, and the new MTHP_STAT_NR_ANON_SPLIT_DEFERRED is counting the number of folios that are currently on the list?
In which case, do we need the "ANON" in the name? It's implicit for the existing split counters that they are anon-only. That would relate it more clearly to the existing MTHP_STAT_SPLIT_DEFERRED too?
Quoted 55 lines
include/linux/huge_mm.h +0
-0__MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -579,6 +579,7 @@ DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON ); +DEFINE_MTHP_STAT_ATTR (anon_num_partial_unused,MTHP_STAT_NR_ANON_SPLIT_DEFERRED );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -593,6 +594,7 @@ static struct attribute *stats_attrs[] = { &split_failed_attr.attr, &split_deferred_attr.attr, &anon_num_attr.attr, + &anon_num_partial_unused_attr.attr, NULL, };@@ -3229,6 +3231,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,if (folio_order (folio) >1 && !list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );/* * Reinitialize page_deferred_list after removing the * page from the split_queue, otherwise a subsequent @@ -3291,6 +3294,7 @@ void __folio_undo_large_rmappable(struct folio *folio)spin_lock_irqsave (&ds_queue->split_queue_lock, flags);if (!list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); }spin_unlock_irqrestore (&ds_queue->split_queue_lock, flags);@@ -3332,6 +3336,7 @@ void deferred_split_folio(struct folio *folio)if (folio_test_pmd_mappable (folio))count_vm_event (THP_DEFERRED_SPLIT_PAGE );count_mthp_stat (folio_order (folio),MTHP_STAT_SPLIT_DEFERRED ); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,1 );list_add_tail (&folio->_deferred_list, &ds_queue->split_queue); ds_queue->split_queue_len++;#ifdef CONFIG_MEMCG @@ -3379,6 +3384,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,list_move (&folio->_deferred_list, &list); }else {/* We lost race with folio_put() */ +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); ds_queue->split_queue_len--; }
Re: [PATCH RFC 2/2] mm: collect the number of anon large folios partially unmapped
Quoted 29 lines · Ryan Roberts
Quoted 27 lines · Barry Song
From: Barry Song <[email protected]>
When an mTHP is added to the
deferred_list, its partial pages are unused, leading to wasted memory and potentially increasing memory reclamation pressure. Tracking this number indicates the extent to which userspace is partially unmapping mTHPs.Detailing the specifics of how unmapping occurs is quite difficult and not that useful, so we adopt a simple approach: each time an mTHP enters the
deferred_list, we increment the count by 1; whenever it leaves for any reason, we decrement the count by 1.Signed-off-by Barry Song <[email protected]>--- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/huge_mm.h | 1 + mm/huge_memory.c | 6 ++++++ 3 files changed, 12 insertions(+)Documentation/admin-guide/mm/transhuge.rst +0
-0@@ -532,6 +532,11 @@ anon_num These huge pages could be still entirely mapped and have partially unmapped and unused subpages.
Quoted 4 lines
Documentation/admin-guide/mm/transhuge.rst anon_num · line 535
Quoted 1 line
Documentation/admin-guide/mm/transhuge.rst +1
-0+anon_num_partial_unusedWhy is the user-exposed name completely different to the internal (
MTHP_STAT_NR_ANON_SPLIT_DEFERRED) name?
My point is that the user might not even know what a deferred split is; they are more concerned with whether there's any temporary memory waste or what the deferred list means from a user perspective.
However, since we've referred to it as SPLIT_DEFERRED in other sys ABI, I agree with you that we should continue using that term.
Quoted 13 lines
Quoted 13 lines
Documentation/admin-guide/mm/transhuge.rst +4
-0+ the number of anon huge pages which have been partially unmapped + we have in the whole system. These unmapped subpages are also + unused and temporarily wasting memory. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to helpinclude/linux/huge_mm.h +0
-0@@ -282,6 +282,7 @@ enum mthp_stat_item {
Quoted 8 lines
include/linux/huge_mm.h enum mthp_stat_item { · line 285
Quoted 4 lines
include/linux/huge_mm.h +1
-0MTHP_STAT_SPLIT_FAILED ,MTHP_STAT_SPLIT_DEFERRED ,MTHP_STAT_NR_ANON , + MTHP_STAT_NR_ANON_SPLIT_DEFERRED,So the existing
MTHP_STAT_SPLIT_DEFERREDis counting all folios that were ever put on the list, and the newMTHP_STAT_NR_ANON_SPLIT_DEFERREDis counting the number of folios that are currently on the list?
Yep.
Quoted 4 lines
In which case, do we need the "ANON" in the name? It's implicit for the existing split counters that they are anon-only. That would relate it more clearly to the existing
MTHP_STAT_SPLIT_DEFERREDtoo?
ack.
Quoted 48 lines
Quoted 48 lines
include/linux/huge_mm.h +0
-0__MTHP_STAT_COUNT };mm/huge_memory.c +5
-0@@ -579,6 +579,7 @@ DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );DEFINE_MTHP_STAT_ATTR (anon_num,MTHP_STAT_NR_ANON ); +DEFINE_MTHP_STAT_ATTR (anon_num_partial_unused,MTHP_STAT_NR_ANON_SPLIT_DEFERRED );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -593,6 +594,7 @@ static struct attribute *stats_attrs[] = { &split_failed_attr.attr, &split_deferred_attr.attr, &anon_num_attr.attr, + &anon_num_partial_unused_attr.attr, NULL, };@@ -3229,6 +3231,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,if (folio_order (folio) >1 && !list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );/* * Reinitialize page_deferred_list after removing the * page from the split_queue, otherwise a subsequent @@ -3291,6 +3294,7 @@ void __folio_undo_large_rmappable(struct folio *folio)spin_lock_irqsave (&ds_queue->split_queue_lock, flags);if (!list_empty (&folio->_deferred_list)) { ds_queue->split_queue_len--; +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); }spin_unlock_irqrestore (&ds_queue->split_queue_lock, flags);@@ -3332,6 +3336,7 @@ void deferred_split_folio(struct folio *folio)if (folio_test_pmd_mappable (folio))count_vm_event (THP_DEFERRED_SPLIT_PAGE );count_mthp_stat (folio_order (folio),MTHP_STAT_SPLIT_DEFERRED ); +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,1 );list_add_tail (&folio->_deferred_list, &ds_queue->split_queue); ds_queue->split_queue_len++;#ifdef CONFIG_MEMCG
Quoted 9 lines
mm/huge_memory.c static unsigned long deferred_split_scan(struct shrinker *shrink, · line 3390
Quoted 8 lines
mm/huge_memory.c +1
-0@@ -3379,6 +3384,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,list_move (&folio->_deferred_list, &list); }else {/* We lost race with folio_put() */ +mod_mthp_stat (folio_order (folio),MTHP_STAT_NR_ANON_SPLIT_DEFERRED ,-1 );list_del_init (&folio->_deferred_list); ds_queue->split_queue_len--; }
Thanks
Barry