| Message ID | 20240220214558.3377482-2-souravpanda@google.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm: report per-page metadata information | expand |
On Tue, Feb 20, 2024 at 1:46 PM Sourav Panda <souravpanda@google.com> wrote: > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > to /proc/meminfo. This information can be used by users to see how > much memory is being used by per-page metadata, which can vary > depending on build configuration, machine architecture, and system > use. /proc/vmstat also has the system-wide nr_memmap and nr_memmap_boot. Given that nr_memmap in /proc/vmstat provides the same info (in different units) as Memmap in /proc/meminfo, it would be better to remove Memmap from /proc/meminfo to avoid duplication and confusion. > Per-page metadata is the amount of memory that Linux needs in order to > manage memory at the page granularity. The majority of such memory is > used by "struct page" and "page_ext" data structures. In contrast to > most other memory consumption statistics, per-page metadata might not > be included in MemTotal. For example, MemTotal does not include memblock > allocations but includes buddy allocations. In this patch, exported > field nr_memmap in /sys/devices/system/node/nodeN/vmstat would > exclusively track buddy allocations while nr_memmap_boot would > exclusively track memblock allocations. Furthermore, Memmap in > /proc/meminfo would exclusively track buddy allocations allowing it to > be compared against MemTotal. > > This memory depends on build configurations, machine architectures, and > the way system is used: > > Build configuration may include extra fields into "struct page", > and enable / disable "page_ext" > Machine architecture defines base page sizes. For example 4K x86, > 8K SPARC, 64K ARM64 (optionally), etc. The per-page metadata > overhead is smaller on machines with larger page sizes. > System use can change per-page overhead by using vmemmap > optimizations with hugetlb pages, and emulated pmem devdax pages. > Also, boot parameters can determine whether page_ext is needed > to be allocated. This memory can be part of MemTotal or be outside > MemTotal depending on whether the memory was hot-plugged, booted with, > or hugetlb memory was returned back to the system. > > Utility for userspace: > > Application Optimization: Depending on the kernel version and command > line options, the kernel would relinquish a different number of pages > (that contain struct pages) when a hugetlb page is reserved (e.g., 0, 6 > or 7 for a 2MB hugepage). The userspace application would want to know > the exact savings achieved through page metadata deallocation without > dealing with the intricacies of the kernel. > > Observability: Struct page overhead can only be calculated on-paper at > boot time (e.g., 1.5% machine capacity). Beyond boot once hugepages are > reserved or memory is hotplugged, the computation becomes complex. > Per-page metrics will help explain part of the system memory overhead, > which shall help guide memory optimizations and memory cgroup sizing. > > Debugging: Tracking the changes or absolute value in struct pages can > help detect anomalies as they can be correlated with other metrics in > the machine (e.g., memtotal, number of huge pages, etc). > > page_ext overheads: Some kernel features such as page_owner > page_table_check that use page_ext can be optionally enabled via kernel > parameters. Having the total per-page metadata information helps users > precisely measure impact. > > Suggested-by: Pasha Tatashin <pasha.tatashin@soleen.com> > Signed-off-by: Sourav Panda <souravpanda@google.com> > --- > Documentation/filesystems/proc.rst | 3 +++ > fs/proc/meminfo.c | 4 ++++ > include/linux/mmzone.h | 4 ++++ > include/linux/vmstat.h | 4 ++++ > mm/hugetlb_vmemmap.c | 17 ++++++++++++---- > mm/mm_init.c | 3 +++ > mm/page_alloc.c | 1 + > mm/page_ext.c | 32 +++++++++++++++++++++--------- > mm/sparse-vmemmap.c | 8 ++++++++ > mm/sparse.c | 7 ++++++- > mm/vmstat.c | 26 +++++++++++++++++++++++- > 11 files changed, 94 insertions(+), 15 deletions(-) > > diff --git a/Documentation/filesystems/proc.rst b/Documentation/filesystems/proc.rst > index 104c6d047d9b..c9b4de65f162 100644 > --- a/Documentation/filesystems/proc.rst > +++ b/Documentation/filesystems/proc.rst > @@ -993,6 +993,7 @@ Example output. You may not have all of these fields. > AnonPages: 4654780 kB > Mapped: 266244 kB > Shmem: 9976 kB > + Memmap: 513419 kB > KReclaimable: 517708 kB > Slab: 660044 kB > SReclaimable: 517708 kB > @@ -1095,6 +1096,8 @@ Mapped > files which have been mmapped, such as libraries > Shmem > Total memory used by shared memory (shmem) and tmpfs > +Memmap > + Memory used for per-page metadata > KReclaimable > Kernel allocations that the kernel will attempt to reclaim > under memory pressure. Includes SReclaimable (below), and other > diff --git a/fs/proc/meminfo.c b/fs/proc/meminfo.c > index 45af9a989d40..3d3db55cfeab 100644 > --- a/fs/proc/meminfo.c > +++ b/fs/proc/meminfo.c > @@ -39,6 +39,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v) > long available; > unsigned long pages[NR_LRU_LISTS]; > unsigned long sreclaimable, sunreclaim; > + unsigned long nr_memmap; > int lru; > > si_meminfo(&i); > @@ -57,6 +58,8 @@ static int meminfo_proc_show(struct seq_file *m, void *v) > sreclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B); > sunreclaim = global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B); > > + nr_memmap = global_node_page_state_pages(NR_MEMMAP); > + > show_val_kb(m, "MemTotal: ", i.totalram); > show_val_kb(m, "MemFree: ", i.freeram); > show_val_kb(m, "MemAvailable: ", available); > @@ -104,6 +107,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v) > show_val_kb(m, "Mapped: ", > global_node_page_state(NR_FILE_MAPPED)); > show_val_kb(m, "Shmem: ", i.sharedram); > + show_val_kb(m, "Memmap: ", nr_memmap); > show_val_kb(m, "KReclaimable: ", sreclaimable + > global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE)); > show_val_kb(m, "Slab: ", sreclaimable + sunreclaim); > diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h > index a497f189d988..59b244092325 100644 > --- a/include/linux/mmzone.h > +++ b/include/linux/mmzone.h > @@ -214,6 +214,10 @@ enum node_stat_item { > PGDEMOTE_KSWAPD, > PGDEMOTE_DIRECT, > PGDEMOTE_KHUGEPAGED, > + NR_MEMMAP, /* Page metadata size (struct page and page_ext) > + * in pages > + */ > + NR_MEMMAP_BOOT, /* NR_MEMMAP for bootmem */ > NR_VM_NODE_STAT_ITEMS > }; > > diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h > index 343906a98d6e..c3785fdd3668 100644 > --- a/include/linux/vmstat.h > +++ b/include/linux/vmstat.h > @@ -632,4 +632,8 @@ static inline void lruvec_stat_sub_folio(struct folio *folio, > { > lruvec_stat_mod_folio(folio, idx, -folio_nr_pages(folio)); > } > + > +void __meminit mod_node_early_perpage_metadata(int nid, long delta); > +void __meminit store_early_perpage_metadata(void); > + > #endif /* _LINUX_VMSTAT_H */ > diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c > index da177e49d956..2da8689aeb93 100644 > --- a/mm/hugetlb_vmemmap.c > +++ b/mm/hugetlb_vmemmap.c > @@ -184,10 +184,13 @@ static int vmemmap_remap_range(unsigned long start, unsigned long end, > */ > static inline void free_vmemmap_page(struct page *page) > { > - if (PageReserved(page)) > + if (PageReserved(page)) { > free_bootmem_page(page); > - else > + mod_node_page_state(page_pgdat(page), NR_MEMMAP_BOOT, -1); > + } else { > __free_page(page); > + mod_node_page_state(page_pgdat(page), NR_MEMMAP, -1); > + } > } > > /* Free a list of the vmemmap pages */ > @@ -338,6 +341,7 @@ static int vmemmap_remap_free(unsigned long start, unsigned long end, > copy_page(page_to_virt(walk.reuse_page), > (void *)walk.reuse_addr); > list_add(&walk.reuse_page->lru, vmemmap_pages); > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, 1); > } > > /* > @@ -384,14 +388,19 @@ static int alloc_vmemmap_page_list(unsigned long start, unsigned long end, > unsigned long nr_pages = (end - start) >> PAGE_SHIFT; > int nid = page_to_nid((struct page *)start); > struct page *page, *next; > + int i; > > - while (nr_pages--) { > + for (i = 0; i < nr_pages; i++) { > page = alloc_pages_node(nid, gfp_mask, 0); > - if (!page) > + if (!page) { > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, i); > goto out; > + } > list_add(&page->lru, list); > } > > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, nr_pages); > + > return 0; > out: > list_for_each_entry_safe(page, next, list, lru) > diff --git a/mm/mm_init.c b/mm/mm_init.c > index 2c19f5515e36..b61372431b7d 100644 > --- a/mm/mm_init.c > +++ b/mm/mm_init.c > @@ -27,6 +27,7 @@ > #include <linux/swap.h> > #include <linux/cma.h> > #include <linux/crash_dump.h> > +#include <linux/vmstat.h> > #include "internal.h" > #include "slab.h" > #include "shuffle.h" > @@ -1656,6 +1657,8 @@ static void __init alloc_node_mem_map(struct pglist_data *pgdat) > panic("Failed to allocate %ld bytes for node %d memory map\n", > size, pgdat->node_id); > pgdat->node_mem_map = map + offset; > + mod_node_early_perpage_metadata(pgdat->node_id, > + DIV_ROUND_UP(size, PAGE_SIZE)); > pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", > __func__, pgdat->node_id, (unsigned long)pgdat, > (unsigned long)pgdat->node_mem_map); > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 150d4f23b010..236cfdf5a8fa 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -5635,6 +5635,7 @@ void __init setup_per_cpu_pageset(void) > for_each_online_pgdat(pgdat) > pgdat->per_cpu_nodestats = > alloc_percpu(struct per_cpu_nodestat); > + store_early_perpage_metadata(); > } > > __meminit void zone_pcp_init(struct zone *zone) > diff --git a/mm/page_ext.c b/mm/page_ext.c > index 4548fcc66d74..c1e324a1427e 100644 > --- a/mm/page_ext.c > +++ b/mm/page_ext.c > @@ -201,6 +201,8 @@ static int __init alloc_node_page_ext(int nid) > return -ENOMEM; > NODE_DATA(nid)->node_page_ext = base; > total_usage += table_size; > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP_BOOT, > + DIV_ROUND_UP(table_size, PAGE_SIZE)); > return 0; > } > > @@ -255,12 +257,15 @@ static void *__meminit alloc_page_ext(size_t size, int nid) > void *addr = NULL; > > addr = alloc_pages_exact_nid(nid, size, flags); > - if (addr) { > + if (addr) > kmemleak_alloc(addr, size, 1, flags); > - return addr; > - } > + else > + addr = vzalloc_node(size, nid); > > - addr = vzalloc_node(size, nid); > + if (addr) { > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, > + DIV_ROUND_UP(size, PAGE_SIZE)); > + } > > return addr; > } > @@ -303,18 +308,27 @@ static int __meminit init_section_page_ext(unsigned long pfn, int nid) > > static void free_page_ext(void *addr) > { > + size_t table_size; > + struct page *page; > + struct pglist_data *pgdat; > + > + table_size = page_ext_size * PAGES_PER_SECTION; > + > if (is_vmalloc_addr(addr)) { > + page = vmalloc_to_page(addr); > + pgdat = page_pgdat(page); > vfree(addr); > } else { > - struct page *page = virt_to_page(addr); > - size_t table_size; > - > - table_size = page_ext_size * PAGES_PER_SECTION; > - > + page = virt_to_page(addr); > + pgdat = page_pgdat(page); > BUG_ON(PageReserved(page)); > kmemleak_free(addr); > free_pages_exact(addr, table_size); > } > + > + mod_node_page_state(pgdat, NR_MEMMAP, > + -1L * (DIV_ROUND_UP(table_size, PAGE_SIZE))); > + > } > > static void __free_page_ext(unsigned long pfn) > diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c > index a2cbe44c48e1..1dda6c53370b 100644 > --- a/mm/sparse-vmemmap.c > +++ b/mm/sparse-vmemmap.c > @@ -469,5 +469,13 @@ struct page * __meminit __populate_section_memmap(unsigned long pfn, > if (r < 0) > return NULL; > > + if (system_state == SYSTEM_BOOTING) { > + mod_node_early_perpage_metadata(nid, DIV_ROUND_UP(end - start, > + PAGE_SIZE)); > + } else { > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, > + DIV_ROUND_UP(end - start, PAGE_SIZE)); > + } > + > return pfn_to_page(pfn); > } > diff --git a/mm/sparse.c b/mm/sparse.c > index 338cf946dee8..eb2aeb4e226b 100644 > --- a/mm/sparse.c > +++ b/mm/sparse.c > @@ -14,7 +14,7 @@ > #include <linux/swap.h> > #include <linux/swapops.h> > #include <linux/bootmem_info.h> > - > +#include <linux/vmstat.h> > #include "internal.h" > #include <asm/dma.h> > > @@ -465,6 +465,9 @@ static void __init sparse_buffer_init(unsigned long size, int nid) > */ > sparsemap_buf = memmap_alloc(size, section_map_size(), addr, nid, true); > sparsemap_buf_end = sparsemap_buf + size; > +#ifndef CONFIG_SPARSEMEM_VMEMMAP > + mod_node_early_perpage_metadata(nid, DIV_ROUND_UP(size, PAGE_SIZE)); > +#endif > } > > static void __init sparse_buffer_fini(void) > @@ -641,6 +644,8 @@ static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages, > unsigned long start = (unsigned long) pfn_to_page(pfn); > unsigned long end = start + nr_pages * sizeof(struct page); > > + mod_node_page_state(page_pgdat(pfn_to_page(pfn)), NR_MEMMAP, > + -1L * (DIV_ROUND_UP(end - start, PAGE_SIZE))); > vmemmap_free(start, end, altmap); > } > static void free_map_bootmem(struct page *memmap) > diff --git a/mm/vmstat.c b/mm/vmstat.c > index db79935e4a54..79466450040e 100644 > --- a/mm/vmstat.c > +++ b/mm/vmstat.c > @@ -1252,7 +1252,8 @@ const char * const vmstat_text[] = { > "pgdemote_kswapd", > "pgdemote_direct", > "pgdemote_khugepaged", > - > + "nr_memmap", > + "nr_memmap_boot", > /* enum writeback_stat_item counters */ > "nr_dirty_threshold", > "nr_dirty_background_threshold", > @@ -2279,4 +2280,27 @@ static int __init extfrag_debug_init(void) > } > > module_init(extfrag_debug_init); > + > #endif > + > +/* > + * Page metadata size (struct page and page_ext) in pages > + */ > +static unsigned long early_perpage_metadata[MAX_NUMNODES] __meminitdata; > + > +void __meminit mod_node_early_perpage_metadata(int nid, long delta) > +{ > + early_perpage_metadata[nid] += delta; > +} > + > +void __meminit store_early_perpage_metadata(void) > +{ > + int nid; > + struct pglist_data *pgdat; > + > + for_each_online_pgdat(pgdat) { > + nid = pgdat->node_id; > + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP_BOOT, > + early_perpage_metadata[nid]); > + } > +} > -- > 2.44.0.rc0.258.g7320e95886-goog >
On Mon, Feb 26, 2024 at 1:12 PM Wei Xu <weixugc@google.com> wrote: > > On Tue, Feb 20, 2024 at 1:46 PM Sourav Panda <souravpanda@google.com> wrote: > > > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > > to /proc/meminfo. This information can be used by users to see how > > much memory is being used by per-page metadata, which can vary > > depending on build configuration, machine architecture, and system > > use. > > /proc/vmstat also has the system-wide nr_memmap and nr_memmap_boot. > Given that nr_memmap in /proc/vmstat provides the same info (in > different units) as Memmap in /proc/meminfo, it would be better to > remove Memmap from /proc/meminfo to avoid duplication and confusion. There are many items both in meminfo and in vmstat. Given that /proc/meminfo covers all kmem memory, it is beneficial to keep the kmem part of memmap in meminfo as another classification item. Pasha
On Tue, Feb 20, 2024 at 4:46 PM Sourav Panda <souravpanda@google.com> wrote: > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > to /proc/meminfo. This information can be used by users to see how > much memory is being used by per-page metadata, which can vary > depending on build configuration, machine architecture, and system > use. > > Per-page metadata is the amount of memory that Linux needs in order to > manage memory at the page granularity. The majority of such memory is > used by "struct page" and "page_ext" data structures. In contrast to > most other memory consumption statistics, per-page metadata might not > be included in MemTotal. For example, MemTotal does not include memblock > allocations but includes buddy allocations. In this patch, exported > field nr_memmap in /sys/devices/system/node/nodeN/vmstat would > exclusively track buddy allocations while nr_memmap_boot would > exclusively track memblock allocations. Furthermore, Memmap in > /proc/meminfo would exclusively track buddy allocations allowing it to > be compared against MemTotal. > > This memory depends on build configurations, machine architectures, and > the way system is used: > > Build configuration may include extra fields into "struct page", > and enable / disable "page_ext" > Machine architecture defines base page sizes. For example 4K x86, > 8K SPARC, 64K ARM64 (optionally), etc. The per-page metadata > overhead is smaller on machines with larger page sizes. > System use can change per-page overhead by using vmemmap > optimizations with hugetlb pages, and emulated pmem devdax pages. > Also, boot parameters can determine whether page_ext is needed > to be allocated. This memory can be part of MemTotal or be outside > MemTotal depending on whether the memory was hot-plugged, booted with, > or hugetlb memory was returned back to the system. > > Utility for userspace: > > Application Optimization: Depending on the kernel version and command > line options, the kernel would relinquish a different number of pages > (that contain struct pages) when a hugetlb page is reserved (e.g., 0, 6 > or 7 for a 2MB hugepage). The userspace application would want to know > the exact savings achieved through page metadata deallocation without > dealing with the intricacies of the kernel. > > Observability: Struct page overhead can only be calculated on-paper at > boot time (e.g., 1.5% machine capacity). Beyond boot once hugepages are > reserved or memory is hotplugged, the computation becomes complex. > Per-page metrics will help explain part of the system memory overhead, > which shall help guide memory optimizations and memory cgroup sizing. > > Debugging: Tracking the changes or absolute value in struct pages can > help detect anomalies as they can be correlated with other metrics in > the machine (e.g., memtotal, number of huge pages, etc). > > page_ext overheads: Some kernel features such as page_owner > page_table_check that use page_ext can be optionally enabled via kernel > parameters. Having the total per-page metadata information helps users > precisely measure impact. > > Suggested-by: Pasha Tatashin <pasha.tatashin@soleen.com> > Signed-off-by: Sourav Panda <souravpanda@google.com> Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
On Wed, Mar 13, 2024 at 6:40 PM Pasha Tatashin <pasha.tatashin@soleen.com> wrote: > > On Tue, Feb 20, 2024 at 4:46 PM Sourav Panda <souravpanda@google.com> wrote: > > > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > > to /proc/meminfo. This information can be used by users to see how > > much memory is being used by per-page metadata, which can vary > > depending on build configuration, machine architecture, and system > > use. > > > > Per-page metadata is the amount of memory that Linux needs in order to > > manage memory at the page granularity. The majority of such memory is > > used by "struct page" and "page_ext" data structures. In contrast to > > most other memory consumption statistics, per-page metadata might not > > be included in MemTotal. For example, MemTotal does not include memblock > > allocations but includes buddy allocations. In this patch, exported > > field nr_memmap in /sys/devices/system/node/nodeN/vmstat would > > exclusively track buddy allocations while nr_memmap_boot would > > exclusively track memblock allocations. Furthermore, Memmap in > > /proc/meminfo would exclusively track buddy allocations allowing it to > > be compared against MemTotal. > > > > This memory depends on build configurations, machine architectures, and > > the way system is used: > > > > Build configuration may include extra fields into "struct page", > > and enable / disable "page_ext" > > Machine architecture defines base page sizes. For example 4K x86, > > 8K SPARC, 64K ARM64 (optionally), etc. The per-page metadata > > overhead is smaller on machines with larger page sizes. > > System use can change per-page overhead by using vmemmap > > optimizations with hugetlb pages, and emulated pmem devdax pages. > > Also, boot parameters can determine whether page_ext is needed > > to be allocated. This memory can be part of MemTotal or be outside > > MemTotal depending on whether the memory was hot-plugged, booted with, > > or hugetlb memory was returned back to the system. > > > > Utility for userspace: > > > > Application Optimization: Depending on the kernel version and command > > line options, the kernel would relinquish a different number of pages > > (that contain struct pages) when a hugetlb page is reserved (e.g., 0, 6 > > or 7 for a 2MB hugepage). The userspace application would want to know > > the exact savings achieved through page metadata deallocation without > > dealing with the intricacies of the kernel. > > > > Observability: Struct page overhead can only be calculated on-paper at > > boot time (e.g., 1.5% machine capacity). Beyond boot once hugepages are > > reserved or memory is hotplugged, the computation becomes complex. > > Per-page metrics will help explain part of the system memory overhead, > > which shall help guide memory optimizations and memory cgroup sizing. > > > > Debugging: Tracking the changes or absolute value in struct pages can > > help detect anomalies as they can be correlated with other metrics in > > the machine (e.g., memtotal, number of huge pages, etc). > > > > page_ext overheads: Some kernel features such as page_owner > > page_table_check that use page_ext can be optionally enabled via kernel > > parameters. Having the total per-page metadata information helps users > > precisely measure impact. Hi Andrew, Can you please give this patch another look, does it require more reviews before you can take it in? Thank you, Pasha
On Tue, 20 Feb 2024 13:45:58 -0800 Sourav Panda <souravpanda@google.com> wrote: > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > to /proc/meminfo. This information can be used by users to see how > much memory is being used by per-page metadata, which can vary > depending on build configuration, machine architecture, and system > use. I yield to no man in my admiration of changelogging but boy, that's a lot of changelogging. Would it be possible to consolidate the [0/N] coverletter and the [1/N] changelog into a single thing please? > Documentation/filesystems/proc.rst | 3 +++ > fs/proc/meminfo.c | 4 ++++ > include/linux/mmzone.h | 4 ++++ > include/linux/vmstat.h | 4 ++++ > mm/hugetlb_vmemmap.c | 17 ++++++++++++---- > mm/mm_init.c | 3 +++ > mm/page_alloc.c | 1 + > mm/page_ext.c | 32 +++++++++++++++++++++--------- > mm/sparse-vmemmap.c | 8 ++++++++ > mm/sparse.c | 7 ++++++- > mm/vmstat.c | 26 +++++++++++++++++++++++- > 11 files changed, 94 insertions(+), 15 deletions(-) And yet we offer the users basically no documentation. The new sysfs file should be documented under Documentation/ABI somewhere and perhaps we could prepare some more expansive user-facing documentation elsewhere? I'd like to hear others' views on the overall usefulness/utility of this change, please?
On Tue, 19 Mar 2024, Andrew Morton wrote: > On Tue, 20 Feb 2024 13:45:58 -0800 Sourav Panda <souravpanda@google.com> wrote: > > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > > to /proc/meminfo. This information can be used by users to see how > > much memory is being used by per-page metadata, which can vary > > depending on build configuration, machine architecture, and system > > use. > > I yield to no man in my admiration of changelogging but boy, that's a > lot of changelogging. Would it be possible to consolidate the [0/N] > coverletter and the [1/N] changelog into a single thing please? > > > Documentation/filesystems/proc.rst | 3 +++ > > fs/proc/meminfo.c | 4 ++++ > > include/linux/mmzone.h | 4 ++++ > > include/linux/vmstat.h | 4 ++++ > > mm/hugetlb_vmemmap.c | 17 ++++++++++++---- > > mm/mm_init.c | 3 +++ > > mm/page_alloc.c | 1 + > > mm/page_ext.c | 32 +++++++++++++++++++++--------- > > mm/sparse-vmemmap.c | 8 ++++++++ > > mm/sparse.c | 7 ++++++- > > mm/vmstat.c | 26 +++++++++++++++++++++++- > > 11 files changed, 94 insertions(+), 15 deletions(-) > > And yet we offer the users basically no documentation. The new sysfs > file should be documented under Documentation/ABI somewhere and > perhaps we could prepare some more expansive user-facing documentation > elsewhere? > Sourav, is it possible to refresh this series into a v10 on top of the latest upstream kernel with a single condensed changelog that details the current behavior, what extension this is adding, and how it is generally useful? As noted here, the cover letter has great material that discusses the rationale for this change but would be lost if only this patch is merged. So typically the cover letter material gets concatenated into the changelog, but in this case there's a lot of overlap. A single patch that includes a succinct changelog would be awesome. And then the requested documentation in Documentation/ABI either included in the same patch or as a second patch in the series? I don't think the resulting patch series will actually need a cover letter after that, it will be able to stand on its own. > I'd like to hear others' views on the overall usefulness/utility of this > change, please? > Likely true for all hyperscalers, the immediate use case that this could be applied to is to track boot memory overhead and any regression over time (across kernel upgrades, firmware upgrades, etc) that may change the amount of total memory available. We'd want to subtract out the boot overhead that we know about (like struct page here) and then alert on any regression where we're losing memory from reboot to reboot for any reason. This increased visibility into boot memory overhead allows us to create a mechanism to track changes over time when otherwise that attribution of that memory is not available.
On Wed, Apr 10, 2024 at 4:58 PM David Rientjes <rientjes@google.com> wrote: > On Tue, 19 Mar 2024, Andrew Morton wrote: > > > On Tue, 20 Feb 2024 13:45:58 -0800 Sourav Panda <souravpanda@google.com> > wrote: > > > > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, > > > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field > > > to /proc/meminfo. This information can be used by users to see how > > > much memory is being used by per-page metadata, which can vary > > > depending on build configuration, machine architecture, and system > > > use. > > > > I yield to no man in my admiration of changelogging but boy, that's a > > lot of changelogging. Would it be possible to consolidate the [0/N] > > coverletter and the [1/N] changelog into a single thing please? > > > > > Documentation/filesystems/proc.rst | 3 +++ > > > fs/proc/meminfo.c | 4 ++++ > > > include/linux/mmzone.h | 4 ++++ > > > include/linux/vmstat.h | 4 ++++ > > > mm/hugetlb_vmemmap.c | 17 ++++++++++++---- > > > mm/mm_init.c | 3 +++ > > > mm/page_alloc.c | 1 + > > > mm/page_ext.c | 32 +++++++++++++++++++++--------- > > > mm/sparse-vmemmap.c | 8 ++++++++ > > > mm/sparse.c | 7 ++++++- > > > mm/vmstat.c | 26 +++++++++++++++++++++++- > > > 11 files changed, 94 insertions(+), 15 deletions(-) > > > > And yet we offer the users basically no documentation. The new sysfs > > file should be documented under Documentation/ABI somewhere and > > perhaps we could prepare some more expansive user-facing documentation > > elsewhere? > > > > Sourav, is it possible to refresh this series into a v10 on top of the > latest upstream kernel with a single condensed changelog that details the > current behavior, what extension this is adding, and how it is generally > useful? > > As noted here, the cover letter has great material that discusses the > rationale for this change but would be lost if only this patch is merged. > So typically the cover letter material gets concatenated into the > changelog, but in this case there's a lot of overlap. > > A single patch that includes a succinct changelog would be awesome. > > And then the requested documentation in Documentation/ABI either included > in the same patch or as a second patch in the series? > > I don't think the resulting patch series will actually need a cover letter > after that, it will be able to stand on its own. > Thanks David, I will send v10 soon. > > > I'd like to hear others' views on the overall usefulness/utility of this > > change, please? > > > > Likely true for all hyperscalers, the immediate use case that this could > be applied to is to track boot memory overhead and any regression over > time (across kernel upgrades, firmware upgrades, etc) that may change the > amount of total memory available. We'd want to subtract out the boot > overhead that we know about (like struct page here) and then alert on any > regression where we're losing memory from reboot to reboot for any reason. > > This increased visibility into boot memory overhead allows us to create a > mechanism to track changes over time when otherwise that attribution of > that memory is not available. >
> > Documentation/filesystems/proc.rst | 3 +++ > > fs/proc/meminfo.c | 4 ++++ > > include/linux/mmzone.h | 4 ++++ > > include/linux/vmstat.h | 4 ++++ > > mm/hugetlb_vmemmap.c | 17 ++++++++++++---- > > mm/mm_init.c | 3 +++ > > mm/page_alloc.c | 1 + > > mm/page_ext.c | 32 +++++++++++++++++++++--------- > > mm/sparse-vmemmap.c | 8 ++++++++ > > mm/sparse.c | 7 ++++++- > > mm/vmstat.c | 26 +++++++++++++++++++++++- > > 11 files changed, 94 insertions(+), 15 deletions(-) > > And yet we offer the users basically no documentation. The new sysfs > file should be documented under Documentation/ABI somewhere and There are no new sysfs files in this change. The new Memmap field in /proc/meminfo is documented. > perhaps we could prepare some more expansive user-facing documentation > elsewhere? > > I'd like to hear others' views on the overall usefulness/utility of this > change, please? Sourav, could you please consolidate the cover letter and the patch into one email, sync it with the upstream kernel, and send the new version putting the necessary background information into the stat area in the patch. Thanks, Pasha
diff --git a/Documentation/filesystems/proc.rst b/Documentation/filesystems/proc.rst index 104c6d047d9b..c9b4de65f162 100644 --- a/Documentation/filesystems/proc.rst +++ b/Documentation/filesystems/proc.rst @@ -993,6 +993,7 @@ Example output. You may not have all of these fields. AnonPages: 4654780 kB Mapped: 266244 kB Shmem: 9976 kB + Memmap: 513419 kB KReclaimable: 517708 kB Slab: 660044 kB SReclaimable: 517708 kB @@ -1095,6 +1096,8 @@ Mapped files which have been mmapped, such as libraries Shmem Total memory used by shared memory (shmem) and tmpfs +Memmap + Memory used for per-page metadata KReclaimable Kernel allocations that the kernel will attempt to reclaim under memory pressure. Includes SReclaimable (below), and other diff --git a/fs/proc/meminfo.c b/fs/proc/meminfo.c index 45af9a989d40..3d3db55cfeab 100644 --- a/fs/proc/meminfo.c +++ b/fs/proc/meminfo.c @@ -39,6 +39,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v) long available; unsigned long pages[NR_LRU_LISTS]; unsigned long sreclaimable, sunreclaim; + unsigned long nr_memmap; int lru; si_meminfo(&i); @@ -57,6 +58,8 @@ static int meminfo_proc_show(struct seq_file *m, void *v) sreclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B); sunreclaim = global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B); + nr_memmap = global_node_page_state_pages(NR_MEMMAP); + show_val_kb(m, "MemTotal: ", i.totalram); show_val_kb(m, "MemFree: ", i.freeram); show_val_kb(m, "MemAvailable: ", available); @@ -104,6 +107,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v) show_val_kb(m, "Mapped: ", global_node_page_state(NR_FILE_MAPPED)); show_val_kb(m, "Shmem: ", i.sharedram); + show_val_kb(m, "Memmap: ", nr_memmap); show_val_kb(m, "KReclaimable: ", sreclaimable + global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE)); show_val_kb(m, "Slab: ", sreclaimable + sunreclaim); diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index a497f189d988..59b244092325 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -214,6 +214,10 @@ enum node_stat_item { PGDEMOTE_KSWAPD, PGDEMOTE_DIRECT, PGDEMOTE_KHUGEPAGED, + NR_MEMMAP, /* Page metadata size (struct page and page_ext) + * in pages + */ + NR_MEMMAP_BOOT, /* NR_MEMMAP for bootmem */ NR_VM_NODE_STAT_ITEMS }; diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h index 343906a98d6e..c3785fdd3668 100644 --- a/include/linux/vmstat.h +++ b/include/linux/vmstat.h @@ -632,4 +632,8 @@ static inline void lruvec_stat_sub_folio(struct folio *folio, { lruvec_stat_mod_folio(folio, idx, -folio_nr_pages(folio)); } + +void __meminit mod_node_early_perpage_metadata(int nid, long delta); +void __meminit store_early_perpage_metadata(void); + #endif /* _LINUX_VMSTAT_H */ diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c index da177e49d956..2da8689aeb93 100644 --- a/mm/hugetlb_vmemmap.c +++ b/mm/hugetlb_vmemmap.c @@ -184,10 +184,13 @@ static int vmemmap_remap_range(unsigned long start, unsigned long end, */ static inline void free_vmemmap_page(struct page *page) { - if (PageReserved(page)) + if (PageReserved(page)) { free_bootmem_page(page); - else + mod_node_page_state(page_pgdat(page), NR_MEMMAP_BOOT, -1); + } else { __free_page(page); + mod_node_page_state(page_pgdat(page), NR_MEMMAP, -1); + } } /* Free a list of the vmemmap pages */ @@ -338,6 +341,7 @@ static int vmemmap_remap_free(unsigned long start, unsigned long end, copy_page(page_to_virt(walk.reuse_page), (void *)walk.reuse_addr); list_add(&walk.reuse_page->lru, vmemmap_pages); + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, 1); } /* @@ -384,14 +388,19 @@ static int alloc_vmemmap_page_list(unsigned long start, unsigned long end, unsigned long nr_pages = (end - start) >> PAGE_SHIFT; int nid = page_to_nid((struct page *)start); struct page *page, *next; + int i; - while (nr_pages--) { + for (i = 0; i < nr_pages; i++) { page = alloc_pages_node(nid, gfp_mask, 0); - if (!page) + if (!page) { + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, i); goto out; + } list_add(&page->lru, list); } + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, nr_pages); + return 0; out: list_for_each_entry_safe(page, next, list, lru) diff --git a/mm/mm_init.c b/mm/mm_init.c index 2c19f5515e36..b61372431b7d 100644 --- a/mm/mm_init.c +++ b/mm/mm_init.c @@ -27,6 +27,7 @@ #include <linux/swap.h> #include <linux/cma.h> #include <linux/crash_dump.h> +#include <linux/vmstat.h> #include "internal.h" #include "slab.h" #include "shuffle.h" @@ -1656,6 +1657,8 @@ static void __init alloc_node_mem_map(struct pglist_data *pgdat) panic("Failed to allocate %ld bytes for node %d memory map\n", size, pgdat->node_id); pgdat->node_mem_map = map + offset; + mod_node_early_perpage_metadata(pgdat->node_id, + DIV_ROUND_UP(size, PAGE_SIZE)); pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", __func__, pgdat->node_id, (unsigned long)pgdat, (unsigned long)pgdat->node_mem_map); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 150d4f23b010..236cfdf5a8fa 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -5635,6 +5635,7 @@ void __init setup_per_cpu_pageset(void) for_each_online_pgdat(pgdat) pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + store_early_perpage_metadata(); } __meminit void zone_pcp_init(struct zone *zone) diff --git a/mm/page_ext.c b/mm/page_ext.c index 4548fcc66d74..c1e324a1427e 100644 --- a/mm/page_ext.c +++ b/mm/page_ext.c @@ -201,6 +201,8 @@ static int __init alloc_node_page_ext(int nid) return -ENOMEM; NODE_DATA(nid)->node_page_ext = base; total_usage += table_size; + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP_BOOT, + DIV_ROUND_UP(table_size, PAGE_SIZE)); return 0; } @@ -255,12 +257,15 @@ static void *__meminit alloc_page_ext(size_t size, int nid) void *addr = NULL; addr = alloc_pages_exact_nid(nid, size, flags); - if (addr) { + if (addr) kmemleak_alloc(addr, size, 1, flags); - return addr; - } + else + addr = vzalloc_node(size, nid); - addr = vzalloc_node(size, nid); + if (addr) { + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, + DIV_ROUND_UP(size, PAGE_SIZE)); + } return addr; } @@ -303,18 +308,27 @@ static int __meminit init_section_page_ext(unsigned long pfn, int nid) static void free_page_ext(void *addr) { + size_t table_size; + struct page *page; + struct pglist_data *pgdat; + + table_size = page_ext_size * PAGES_PER_SECTION; + if (is_vmalloc_addr(addr)) { + page = vmalloc_to_page(addr); + pgdat = page_pgdat(page); vfree(addr); } else { - struct page *page = virt_to_page(addr); - size_t table_size; - - table_size = page_ext_size * PAGES_PER_SECTION; - + page = virt_to_page(addr); + pgdat = page_pgdat(page); BUG_ON(PageReserved(page)); kmemleak_free(addr); free_pages_exact(addr, table_size); } + + mod_node_page_state(pgdat, NR_MEMMAP, + -1L * (DIV_ROUND_UP(table_size, PAGE_SIZE))); + } static void __free_page_ext(unsigned long pfn) diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c index a2cbe44c48e1..1dda6c53370b 100644 --- a/mm/sparse-vmemmap.c +++ b/mm/sparse-vmemmap.c @@ -469,5 +469,13 @@ struct page * __meminit __populate_section_memmap(unsigned long pfn, if (r < 0) return NULL; + if (system_state == SYSTEM_BOOTING) { + mod_node_early_perpage_metadata(nid, DIV_ROUND_UP(end - start, + PAGE_SIZE)); + } else { + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP, + DIV_ROUND_UP(end - start, PAGE_SIZE)); + } + return pfn_to_page(pfn); } diff --git a/mm/sparse.c b/mm/sparse.c index 338cf946dee8..eb2aeb4e226b 100644 --- a/mm/sparse.c +++ b/mm/sparse.c @@ -14,7 +14,7 @@ #include <linux/swap.h> #include <linux/swapops.h> #include <linux/bootmem_info.h> - +#include <linux/vmstat.h> #include "internal.h" #include <asm/dma.h> @@ -465,6 +465,9 @@ static void __init sparse_buffer_init(unsigned long size, int nid) */ sparsemap_buf = memmap_alloc(size, section_map_size(), addr, nid, true); sparsemap_buf_end = sparsemap_buf + size; +#ifndef CONFIG_SPARSEMEM_VMEMMAP + mod_node_early_perpage_metadata(nid, DIV_ROUND_UP(size, PAGE_SIZE)); +#endif } static void __init sparse_buffer_fini(void) @@ -641,6 +644,8 @@ static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages, unsigned long start = (unsigned long) pfn_to_page(pfn); unsigned long end = start + nr_pages * sizeof(struct page); + mod_node_page_state(page_pgdat(pfn_to_page(pfn)), NR_MEMMAP, + -1L * (DIV_ROUND_UP(end - start, PAGE_SIZE))); vmemmap_free(start, end, altmap); } static void free_map_bootmem(struct page *memmap) diff --git a/mm/vmstat.c b/mm/vmstat.c index db79935e4a54..79466450040e 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -1252,7 +1252,8 @@ const char * const vmstat_text[] = { "pgdemote_kswapd", "pgdemote_direct", "pgdemote_khugepaged", - + "nr_memmap", + "nr_memmap_boot", /* enum writeback_stat_item counters */ "nr_dirty_threshold", "nr_dirty_background_threshold", @@ -2279,4 +2280,27 @@ static int __init extfrag_debug_init(void) } module_init(extfrag_debug_init); + #endif + +/* + * Page metadata size (struct page and page_ext) in pages + */ +static unsigned long early_perpage_metadata[MAX_NUMNODES] __meminitdata; + +void __meminit mod_node_early_perpage_metadata(int nid, long delta) +{ + early_perpage_metadata[nid] += delta; +} + +void __meminit store_early_perpage_metadata(void) +{ + int nid; + struct pglist_data *pgdat; + + for_each_online_pgdat(pgdat) { + nid = pgdat->node_id; + mod_node_page_state(NODE_DATA(nid), NR_MEMMAP_BOOT, + early_perpage_metadata[nid]); + } +}
Adds two new per-node fields, namely nr_memmap and nr_memmap_boot, to /sys/devices/system/node/nodeN/vmstat and a global Memmap field to /proc/meminfo. This information can be used by users to see how much memory is being used by per-page metadata, which can vary depending on build configuration, machine architecture, and system use. Per-page metadata is the amount of memory that Linux needs in order to manage memory at the page granularity. The majority of such memory is used by "struct page" and "page_ext" data structures. In contrast to most other memory consumption statistics, per-page metadata might not be included in MemTotal. For example, MemTotal does not include memblock allocations but includes buddy allocations. In this patch, exported field nr_memmap in /sys/devices/system/node/nodeN/vmstat would exclusively track buddy allocations while nr_memmap_boot would exclusively track memblock allocations. Furthermore, Memmap in /proc/meminfo would exclusively track buddy allocations allowing it to be compared against MemTotal. This memory depends on build configurations, machine architectures, and the way system is used: Build configuration may include extra fields into "struct page", and enable / disable "page_ext" Machine architecture defines base page sizes. For example 4K x86, 8K SPARC, 64K ARM64 (optionally), etc. The per-page metadata overhead is smaller on machines with larger page sizes. System use can change per-page overhead by using vmemmap optimizations with hugetlb pages, and emulated pmem devdax pages. Also, boot parameters can determine whether page_ext is needed to be allocated. This memory can be part of MemTotal or be outside MemTotal depending on whether the memory was hot-plugged, booted with, or hugetlb memory was returned back to the system. Utility for userspace: Application Optimization: Depending on the kernel version and command line options, the kernel would relinquish a different number of pages (that contain struct pages) when a hugetlb page is reserved (e.g., 0, 6 or 7 for a 2MB hugepage). The userspace application would want to know the exact savings achieved through page metadata deallocation without dealing with the intricacies of the kernel. Observability: Struct page overhead can only be calculated on-paper at boot time (e.g., 1.5% machine capacity). Beyond boot once hugepages are reserved or memory is hotplugged, the computation becomes complex. Per-page metrics will help explain part of the system memory overhead, which shall help guide memory optimizations and memory cgroup sizing. Debugging: Tracking the changes or absolute value in struct pages can help detect anomalies as they can be correlated with other metrics in the machine (e.g., memtotal, number of huge pages, etc). page_ext overheads: Some kernel features such as page_owner page_table_check that use page_ext can be optionally enabled via kernel parameters. Having the total per-page metadata information helps users precisely measure impact. Suggested-by: Pasha Tatashin <pasha.tatashin@soleen.com> Signed-off-by: Sourav Panda <souravpanda@google.com> --- Documentation/filesystems/proc.rst | 3 +++ fs/proc/meminfo.c | 4 ++++ include/linux/mmzone.h | 4 ++++ include/linux/vmstat.h | 4 ++++ mm/hugetlb_vmemmap.c | 17 ++++++++++++---- mm/mm_init.c | 3 +++ mm/page_alloc.c | 1 + mm/page_ext.c | 32 +++++++++++++++++++++--------- mm/sparse-vmemmap.c | 8 ++++++++ mm/sparse.c | 7 ++++++- mm/vmstat.c | 26 +++++++++++++++++++++++- 11 files changed, 94 insertions(+), 15 deletions(-)