| Message ID | 20201022151844.489337-1-hannes@cmpxchg.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm: memcontrol: add file_thp, shmem_thp to memory.stat | expand |
On Thu, 2020-10-22 at 11:18 -0400, Johannes Weiner wrote: > index e80aa9d2db68..334ce608735c 100644 > --- a/mm/filemap.c > +++ b/mm/filemap.c > @@ -204,9 +204,9 @@ static void unaccount_page_cache_page(struct > address_space *mapping, > if (PageSwapBacked(page)) { > __mod_lruvec_page_state(page, NR_SHMEM, -nr); > if (PageTransHuge(page)) > - __dec_node_page_state(page, NR_SHMEM_THPS); > + __dec_lruvec_page_state(page, NR_SHMEM_THPS); > } else if (PageTransHuge(page)) { > - __dec_node_page_state(page, NR_FILE_THPS); > + __dec_lruvec_page_state(page, NR_FILE_THPS); > filemap_nr_thps_dec(mapping); > } This may be a dumb question, but does that mean the NR_FILE_THPS number will no longer be visible in /proc/vmstat or is there some magic I overlooked in a cursory look of the code?
On Thu, Oct 22, 2020 at 8:20 AM Johannes Weiner <hannes@cmpxchg.org> wrote: > > As huge page usage in the page cache and for shmem files proliferates > in our production environment, the performance monitoring team has > asked for per-cgroup stats on those pages. > > We already track and export anon_thp per cgroup. We already track file > THP and shmem THP per node, so making them per-cgroup is only a matter > of switching from node to lruvec counters. All callsites are in places > where the pages are charged and locked, so page->memcg is stable. > > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Reviewed-by: Shakeel Butt <shakeelb@google.com>
On Thu, 2020-10-22 at 12:49 -0400, Rik van Riel wrote: > On Thu, 2020-10-22 at 11:18 -0400, Johannes Weiner wrote: > > > index e80aa9d2db68..334ce608735c 100644 > > --- a/mm/filemap.c > > +++ b/mm/filemap.c > > @@ -204,9 +204,9 @@ static void unaccount_page_cache_page(struct > > address_space *mapping, > > if (PageSwapBacked(page)) { > > __mod_lruvec_page_state(page, NR_SHMEM, -nr); > > if (PageTransHuge(page)) > > - __dec_node_page_state(page, NR_SHMEM_THPS); > > + __dec_lruvec_page_state(page, NR_SHMEM_THPS); > > } else if (PageTransHuge(page)) { > > - __dec_node_page_state(page, NR_FILE_THPS); > > + __dec_lruvec_page_state(page, NR_FILE_THPS); > > filemap_nr_thps_dec(mapping); > > } > > This may be a dumb question, but does that mean the > NR_FILE_THPS number will no longer be visible in > /proc/vmstat or is there some magic I overlooked in > a cursory look of the code? Never mind, I found it a few levels deep in __dec_lruvec_page_state. Reviewed-by: Rik van Riel <riel@surriel.com>
On Thu, 22 Oct 2020, Johannes Weiner wrote: > As huge page usage in the page cache and for shmem files proliferates > in our production environment, the performance monitoring team has > asked for per-cgroup stats on those pages. > > We already track and export anon_thp per cgroup. We already track file > THP and shmem THP per node, so making them per-cgroup is only a matter > of switching from node to lruvec counters. All callsites are in places > where the pages are charged and locked, so page->memcg is stable. > > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: David Rientjes <rientjes@google.com> Nice!
On Thu, Oct 22, 2020 at 12:57:55PM -0400, Rik van Riel wrote: > On Thu, 2020-10-22 at 12:49 -0400, Rik van Riel wrote: > > On Thu, 2020-10-22 at 11:18 -0400, Johannes Weiner wrote: > > > > > index e80aa9d2db68..334ce608735c 100644 > > > --- a/mm/filemap.c > > > +++ b/mm/filemap.c > > > @@ -204,9 +204,9 @@ static void unaccount_page_cache_page(struct > > > address_space *mapping, > > > if (PageSwapBacked(page)) { > > > __mod_lruvec_page_state(page, NR_SHMEM, -nr); > > > if (PageTransHuge(page)) > > > - __dec_node_page_state(page, NR_SHMEM_THPS); > > > + __dec_lruvec_page_state(page, NR_SHMEM_THPS); > > > } else if (PageTransHuge(page)) { > > > - __dec_node_page_state(page, NR_FILE_THPS); > > > + __dec_lruvec_page_state(page, NR_FILE_THPS); > > > filemap_nr_thps_dec(mapping); > > > } > > > > This may be a dumb question, but does that mean the > > NR_FILE_THPS number will no longer be visible in > > /proc/vmstat or is there some magic I overlooked in > > a cursory look of the code? > > Never mind, I found it a few levels deep in > __dec_lruvec_page_state. No worries, it's a legit question. lruvec is at the intersection of node and memcg, so I'm just moving the accounting to a higher-granularity function that updates all layers, including the node. > Reviewed-by: Rik van Riel <riel@surriel.com> Thanks!
On Thu 22-10-20 11:18:44, Johannes Weiner wrote: > As huge page usage in the page cache and for shmem files proliferates > in our production environment, the performance monitoring team has > asked for per-cgroup stats on those pages. > > We already track and export anon_thp per cgroup. We already track file > THP and shmem THP per node, so making them per-cgroup is only a matter > of switching from node to lruvec counters. All callsites are in places > where the pages are charged and locked, so page->memcg is stable. > > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Michal Hocko <mhocko@suse.com> > --- > mm/filemap.c | 4 ++-- > mm/huge_memory.c | 4 ++-- > mm/khugepaged.c | 4 ++-- > mm/memcontrol.c | 6 +++++- > mm/shmem.c | 2 +- > 5 files changed, 12 insertions(+), 8 deletions(-) > > diff --git a/mm/filemap.c b/mm/filemap.c > index e80aa9d2db68..334ce608735c 100644 > --- a/mm/filemap.c > +++ b/mm/filemap.c > @@ -204,9 +204,9 @@ static void unaccount_page_cache_page(struct address_space *mapping, > if (PageSwapBacked(page)) { > __mod_lruvec_page_state(page, NR_SHMEM, -nr); > if (PageTransHuge(page)) > - __dec_node_page_state(page, NR_SHMEM_THPS); > + __dec_lruvec_page_state(page, NR_SHMEM_THPS); > } else if (PageTransHuge(page)) { > - __dec_node_page_state(page, NR_FILE_THPS); > + __dec_lruvec_page_state(page, NR_FILE_THPS); > filemap_nr_thps_dec(mapping); > } > > diff --git a/mm/huge_memory.c b/mm/huge_memory.c > index cba3812a5c3e..5fe044e5dad5 100644 > --- a/mm/huge_memory.c > +++ b/mm/huge_memory.c > @@ -2707,9 +2707,9 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) > spin_unlock(&ds_queue->split_queue_lock); > if (mapping) { > if (PageSwapBacked(head)) > - __dec_node_page_state(head, NR_SHMEM_THPS); > + __dec_lruvec_page_state(head, NR_SHMEM_THPS); > else > - __dec_node_page_state(head, NR_FILE_THPS); > + __dec_lruvec_page_state(head, NR_FILE_THPS); > } > > __split_huge_page(page, list, end, flags); > diff --git a/mm/khugepaged.c b/mm/khugepaged.c > index f1d5f6dde47c..04828e21f434 100644 > --- a/mm/khugepaged.c > +++ b/mm/khugepaged.c > @@ -1833,9 +1833,9 @@ static void collapse_file(struct mm_struct *mm, > } > > if (is_shmem) > - __inc_node_page_state(new_page, NR_SHMEM_THPS); > + __inc_lruvec_page_state(new_page, NR_SHMEM_THPS); > else { > - __inc_node_page_state(new_page, NR_FILE_THPS); > + __inc_lruvec_page_state(new_page, NR_FILE_THPS); > filemap_nr_thps_inc(mapping); > } > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index 2636f8bad908..98177d5e8e03 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -1507,6 +1507,8 @@ static struct memory_stat memory_stats[] = { > * constant(e.g. powerpc). > */ > { "anon_thp", 0, NR_ANON_THPS }, > + { "file_thp", 0, NR_FILE_THPS }, > + { "shmem_thp", 0, NR_SHMEM_THPS }, > #endif > { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, > { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, > @@ -1537,7 +1539,9 @@ static int __init memory_stats_init(void) > > for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { > #ifdef CONFIG_TRANSPARENT_HUGEPAGE > - if (memory_stats[i].idx == NR_ANON_THPS) > + if (memory_stats[i].idx == NR_ANON_THPS || > + memory_stats[i].idx == NR_FILE_THPS || > + memory_stats[i].idx == NR_SHMEM_THPS) > memory_stats[i].ratio = HPAGE_PMD_SIZE; > #endif > VM_BUG_ON(!memory_stats[i].ratio); > diff --git a/mm/shmem.c b/mm/shmem.c > index 537c137698f8..5009d783d954 100644 > --- a/mm/shmem.c > +++ b/mm/shmem.c > @@ -713,7 +713,7 @@ static int shmem_add_to_page_cache(struct page *page, > } > if (PageTransHuge(page)) { > count_vm_event(THP_FILE_ALLOC); > - __inc_node_page_state(page, NR_SHMEM_THPS); > + __inc_lruvec_page_state(page, NR_SHMEM_THPS); > } > mapping->nrpages += nr; > __mod_lruvec_page_state(page, NR_FILE_PAGES, nr); > -- > 2.29.0
On Thu, 22 Oct 2020 11:18:44 -0400 Johannes Weiner <hannes@cmpxchg.org> wrote: > As huge page usage in the page cache and for shmem files proliferates > in our production environment, the performance monitoring team has > asked for per-cgroup stats on those pages. > > We already track and export anon_thp per cgroup. We already track file > THP and shmem THP per node, so making them per-cgroup is only a matter > of switching from node to lruvec counters. All callsites are in places > where the pages are charged and locked, so page->memcg is stable. > > ... > > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -1507,6 +1507,8 @@ static struct memory_stat memory_stats[] = { > * constant(e.g. powerpc). > */ > { "anon_thp", 0, NR_ANON_THPS }, > + { "file_thp", 0, NR_FILE_THPS }, > + { "shmem_thp", 0, NR_SHMEM_THPS }, Documentation/admin-guide/cgroup-v2.rst is owed an update?
On Sun, Oct 25, 2020 at 11:37:25AM -0700, Andrew Morton wrote: > On Thu, 22 Oct 2020 11:18:44 -0400 Johannes Weiner <hannes@cmpxchg.org> wrote: > > > As huge page usage in the page cache and for shmem files proliferates > > in our production environment, the performance monitoring team has > > asked for per-cgroup stats on those pages. > > > > We already track and export anon_thp per cgroup. We already track file > > THP and shmem THP per node, so making them per-cgroup is only a matter > > of switching from node to lruvec counters. All callsites are in places > > where the pages are charged and locked, so page->memcg is stable. > > > > ... > > > > --- a/mm/memcontrol.c > > +++ b/mm/memcontrol.c > > @@ -1507,6 +1507,8 @@ static struct memory_stat memory_stats[] = { > > * constant(e.g. powerpc). > > */ > > { "anon_thp", 0, NR_ANON_THPS }, > > + { "file_thp", 0, NR_FILE_THPS }, > > + { "shmem_thp", 0, NR_SHMEM_THPS }, > > Documentation/admin-guide/cgroup-v2.rst is owed an update? Ah yes. This? From 310c3e1714e1c093d4cd26dff38326fc348cdd31 Mon Sep 17 00:00:00 2001 From: Johannes Weiner <hannes@cmpxchg.org> Date: Mon, 26 Oct 2020 13:39:19 -0400 Subject: [PATCH] mm: memcontrol: add file_thp, shmem_thp to memory.stat fix Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> --- Documentation/admin-guide/cgroup-v2.rst | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst index 608d7c279396..515bb13084a0 100644 --- a/Documentation/admin-guide/cgroup-v2.rst +++ b/Documentation/admin-guide/cgroup-v2.rst @@ -1300,6 +1300,14 @@ PAGE_SIZE multiple when read back. Amount of memory used in anonymous mappings backed by transparent hugepages + file_thp + Amount of cached filesystem data backed by transparent + hugepages + + shmem_thp + Amount of shm, tmpfs, shared anonymous mmap()s backed by + transparent hugepages + inactive_anon, active_anon, inactive_file, active_file, unevictable Amount of memory, swap-backed and filesystem-backed, on the internal memory management lists used by the
> On Oct 22, 2020, at 8:18 AM, Johannes Weiner <hannes@cmpxchg.org> wrote: > > As huge page usage in the page cache and for shmem files proliferates > in our production environment, the performance monitoring team has > asked for per-cgroup stats on those pages. > > We already track and export anon_thp per cgroup. We already track file > THP and shmem THP per node, so making them per-cgroup is only a matter > of switching from node to lruvec counters. All callsites are in places > where the pages are charged and locked, so page->memcg is stable. > > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Song Liu <songliubraving@fb.com> Thanks!
diff --git a/mm/filemap.c b/mm/filemap.c index e80aa9d2db68..334ce608735c 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -204,9 +204,9 @@ static void unaccount_page_cache_page(struct address_space *mapping, if (PageSwapBacked(page)) { __mod_lruvec_page_state(page, NR_SHMEM, -nr); if (PageTransHuge(page)) - __dec_node_page_state(page, NR_SHMEM_THPS); + __dec_lruvec_page_state(page, NR_SHMEM_THPS); } else if (PageTransHuge(page)) { - __dec_node_page_state(page, NR_FILE_THPS); + __dec_lruvec_page_state(page, NR_FILE_THPS); filemap_nr_thps_dec(mapping); } diff --git a/mm/huge_memory.c b/mm/huge_memory.c index cba3812a5c3e..5fe044e5dad5 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -2707,9 +2707,9 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) spin_unlock(&ds_queue->split_queue_lock); if (mapping) { if (PageSwapBacked(head)) - __dec_node_page_state(head, NR_SHMEM_THPS); + __dec_lruvec_page_state(head, NR_SHMEM_THPS); else - __dec_node_page_state(head, NR_FILE_THPS); + __dec_lruvec_page_state(head, NR_FILE_THPS); } __split_huge_page(page, list, end, flags); diff --git a/mm/khugepaged.c b/mm/khugepaged.c index f1d5f6dde47c..04828e21f434 100644 --- a/mm/khugepaged.c +++ b/mm/khugepaged.c @@ -1833,9 +1833,9 @@ static void collapse_file(struct mm_struct *mm, } if (is_shmem) - __inc_node_page_state(new_page, NR_SHMEM_THPS); + __inc_lruvec_page_state(new_page, NR_SHMEM_THPS); else { - __inc_node_page_state(new_page, NR_FILE_THPS); + __inc_lruvec_page_state(new_page, NR_FILE_THPS); filemap_nr_thps_inc(mapping); } diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 2636f8bad908..98177d5e8e03 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -1507,6 +1507,8 @@ static struct memory_stat memory_stats[] = { * constant(e.g. powerpc). */ { "anon_thp", 0, NR_ANON_THPS }, + { "file_thp", 0, NR_FILE_THPS }, + { "shmem_thp", 0, NR_SHMEM_THPS }, #endif { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, @@ -1537,7 +1539,9 @@ static int __init memory_stats_init(void) for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { #ifdef CONFIG_TRANSPARENT_HUGEPAGE - if (memory_stats[i].idx == NR_ANON_THPS) + if (memory_stats[i].idx == NR_ANON_THPS || + memory_stats[i].idx == NR_FILE_THPS || + memory_stats[i].idx == NR_SHMEM_THPS) memory_stats[i].ratio = HPAGE_PMD_SIZE; #endif VM_BUG_ON(!memory_stats[i].ratio); diff --git a/mm/shmem.c b/mm/shmem.c index 537c137698f8..5009d783d954 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -713,7 +713,7 @@ static int shmem_add_to_page_cache(struct page *page, } if (PageTransHuge(page)) { count_vm_event(THP_FILE_ALLOC); - __inc_node_page_state(page, NR_SHMEM_THPS); + __inc_lruvec_page_state(page, NR_SHMEM_THPS); } mapping->nrpages += nr; __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
As huge page usage in the page cache and for shmem files proliferates in our production environment, the performance monitoring team has asked for per-cgroup stats on those pages. We already track and export anon_thp per cgroup. We already track file THP and shmem THP per node, so making them per-cgroup is only a matter of switching from node to lruvec counters. All callsites are in places where the pages are charged and locked, so page->memcg is stable. Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> --- mm/filemap.c | 4 ++-- mm/huge_memory.c | 4 ++-- mm/khugepaged.c | 4 ++-- mm/memcontrol.c | 6 +++++- mm/shmem.c | 2 +- 5 files changed, 12 insertions(+), 8 deletions(-)