Print out statistics in relation to fragmentation avoidance to /proc/pagetypeinfo
authorMel Gorman <mel@csn.ul.ie>
Tue, 16 Oct 2007 08:26:02 +0000 (01:26 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Tue, 16 Oct 2007 16:43:00 +0000 (09:43 -0700)
This patch provides fragmentation avoidance statistics via /proc/pagetypeinfo.
 The information is collected only on request so there is no runtime overhead.
 The statistics are in three parts:

The first part prints information on the size of blocks that pages are
being grouped on and looks like

Page block order: 10
Pages per block:  1024

The second part is a more detailed version of /proc/buddyinfo and looks like

Free pages count per migrate type at order       0      1      2      3      4      5      6      7      8      9     10
Node    0, zone      DMA, type    Unmovable      0      0      0      0      0      0      0      0      0      0      0
Node    0, zone      DMA, type  Reclaimable      1      0      0      0      0      0      0      0      0      0      0
Node    0, zone      DMA, type      Movable      0      0      0      0      0      0      0      0      0      0      0
Node    0, zone      DMA, type      Reserve      0      4      4      0      0      0      0      1      0      1      0
Node    0, zone   Normal, type    Unmovable    111      8      4      4      2      3      1      0      0      0      0
Node    0, zone   Normal, type  Reclaimable    293     89      8      0      0      0      0      0      0      0      0
Node    0, zone   Normal, type      Movable      1      6     13      9      7      6      3      0      0      0      0
Node    0, zone   Normal, type      Reserve      0      0      0      0      0      0      0      0      0      0      4

The third part looks like

Number of blocks type     Unmovable  Reclaimable      Movable      Reserve
Node 0, zone      DMA            0            1            2            1
Node 0, zone   Normal            3           17           94            4

To walk the zones within a node with interrupts disabled, walk_zones_in_node()
is introduced and shared between /proc/buddyinfo, /proc/zoneinfo and
/proc/pagetypeinfo to reduce code duplication.  It seems specific to what
vmstat.c requires but could be broken out as a general utility function in
mmzone.c if there were other other potential users.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/proc_misc.c
include/linux/gfp.h
include/linux/mmzone.h
mm/page_alloc.c
mm/vmstat.c

index 0071939c00955aa538711eb5d287f7c2af75a151..5de7f874d95c0fce684261bb52d7e44d1d957271 100644 (file)
@@ -229,6 +229,19 @@ static const struct file_operations fragmentation_file_operations = {
        .release        = seq_release,
 };
 
+extern struct seq_operations pagetypeinfo_op;
+static int pagetypeinfo_open(struct inode *inode, struct file *file)
+{
+       return seq_open(file, &pagetypeinfo_op);
+}
+
+static const struct file_operations pagetypeinfo_file_ops = {
+       .open           = pagetypeinfo_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
 extern struct seq_operations zoneinfo_op;
 static int zoneinfo_open(struct inode *inode, struct file *file)
 {
@@ -724,6 +737,7 @@ void __init proc_misc_init(void)
 #endif
 #endif
        create_seq_entry("buddyinfo",S_IRUGO, &fragmentation_file_operations);
+       create_seq_entry("pagetypeinfo", S_IRUGO, &pagetypeinfo_file_ops);
        create_seq_entry("vmstat",S_IRUGO, &proc_vmstat_file_operations);
        create_seq_entry("zoneinfo",S_IRUGO, &proc_zoneinfo_file_operations);
 #ifdef CONFIG_BLOCK
index f8ffcd401c5f5aff0c44547584a4b5ad9bd3e300..7e93a9ae7064c3a59b0223763964181500f6d3cf 100644 (file)
@@ -104,6 +104,18 @@ struct vm_area_struct;
 /* 4GB DMA on some platforms */
 #define GFP_DMA32      __GFP_DMA32
 
+/* Convert GFP flags to their corresponding migrate type */
+static inline int allocflags_to_migratetype(gfp_t gfp_flags)
+{
+       WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
+
+       if (unlikely(page_group_by_mobility_disabled))
+               return MIGRATE_UNMOVABLE;
+
+       /* Group based on mobility */
+       return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
+               ((gfp_flags & __GFP_RECLAIMABLE) != 0);
+}
 
 static inline enum zone_type gfp_zone(gfp_t flags)
 {
index 87a4045580f2789a636ad7782befa14b3110cca8..20ea42c45e4adde226432a755c323a390bd88232 100644 (file)
        for (order = 0; order < MAX_ORDER; order++) \
                for (type = 0; type < MIGRATE_TYPES; type++)
 
+extern int page_group_by_mobility_disabled;
+
+static inline int get_pageblock_migratetype(struct page *page)
+{
+       if (unlikely(page_group_by_mobility_disabled))
+               return MIGRATE_UNMOVABLE;
+
+       return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
+}
+
 struct free_area {
        struct list_head        free_list[MIGRATE_TYPES];
        unsigned long           nr_free;
index b9bc7369c48dddb88475b1783db6a4a5a33f7458..17797d062c3b832aade2dcfa664230934479c47d 100644 (file)
@@ -164,32 +164,12 @@ EXPORT_SYMBOL(nr_node_ids);
 
 int page_group_by_mobility_disabled __read_mostly;
 
-static inline int get_pageblock_migratetype(struct page *page)
-{
-       if (unlikely(page_group_by_mobility_disabled))
-               return MIGRATE_UNMOVABLE;
-
-       return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
-}
-
 static void set_pageblock_migratetype(struct page *page, int migratetype)
 {
        set_pageblock_flags_group(page, (unsigned long)migratetype,
                                        PB_migrate, PB_migrate_end);
 }
 
-static inline int allocflags_to_migratetype(gfp_t gfp_flags)
-{
-       WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
-
-       if (unlikely(page_group_by_mobility_disabled))
-               return MIGRATE_UNMOVABLE;
-
-       /* Cluster based on mobility */
-       return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
-               ((gfp_flags & __GFP_RECLAIMABLE) != 0);
-}
-
 #ifdef CONFIG_DEBUG_VM
 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
 {
index c64d169537bfcfe65fc963b009779de85c9b051a..2f19e4f867c10b069b68f533f3038440b50939d8 100644 (file)
@@ -398,6 +398,13 @@ void zone_statistics(struct zonelist *zonelist, struct zone *z)
 
 #include <linux/seq_file.h>
 
+static char * const migratetype_names[MIGRATE_TYPES] = {
+       "Unmovable",
+       "Reclaimable",
+       "Movable",
+       "Reserve",
+};
+
 static void *frag_start(struct seq_file *m, loff_t *pos)
 {
        pg_data_t *pgdat;
@@ -422,28 +429,144 @@ static void frag_stop(struct seq_file *m, void *arg)
 {
 }
 
-/*
- * This walks the free areas for each zone.
- */
-static int frag_show(struct seq_file *m, void *arg)
+/* Walk all the zones in a node and print using a callback */
+static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
+               void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
 {
-       pg_data_t *pgdat = (pg_data_t *)arg;
        struct zone *zone;
        struct zone *node_zones = pgdat->node_zones;
        unsigned long flags;
-       int order;
 
        for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
                if (!populated_zone(zone))
                        continue;
 
                spin_lock_irqsave(&zone->lock, flags);
-               seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
-               for (order = 0; order < MAX_ORDER; ++order)
-                       seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
+               print(m, pgdat, zone);
                spin_unlock_irqrestore(&zone->lock, flags);
+       }
+}
+
+static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
+                                               struct zone *zone)
+{
+       int order;
+
+       seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
+       for (order = 0; order < MAX_ORDER; ++order)
+               seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
+       seq_putc(m, '\n');
+}
+
+/*
+ * This walks the free areas for each zone.
+ */
+static int frag_show(struct seq_file *m, void *arg)
+{
+       pg_data_t *pgdat = (pg_data_t *)arg;
+       walk_zones_in_node(m, pgdat, frag_show_print);
+       return 0;
+}
+
+static void pagetypeinfo_showfree_print(struct seq_file *m,
+                                       pg_data_t *pgdat, struct zone *zone)
+{
+       int order, mtype;
+
+       for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
+               seq_printf(m, "Node %4d, zone %8s, type %12s ",
+                                       pgdat->node_id,
+                                       zone->name,
+                                       migratetype_names[mtype]);
+               for (order = 0; order < MAX_ORDER; ++order) {
+                       unsigned long freecount = 0;
+                       struct free_area *area;
+                       struct list_head *curr;
+
+                       area = &(zone->free_area[order]);
+
+                       list_for_each(curr, &area->free_list[mtype])
+                               freecount++;
+                       seq_printf(m, "%6lu ", freecount);
+               }
                seq_putc(m, '\n');
        }
+}
+
+/* Print out the free pages at each order for each migatetype */
+static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
+{
+       int order;
+       pg_data_t *pgdat = (pg_data_t *)arg;
+
+       /* Print header */
+       seq_printf(m, "%-43s ", "Free pages count per migrate type at order");
+       for (order = 0; order < MAX_ORDER; ++order)
+               seq_printf(m, "%6d ", order);
+       seq_putc(m, '\n');
+
+       walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print);
+
+       return 0;
+}
+
+static void pagetypeinfo_showblockcount_print(struct seq_file *m,
+                                       pg_data_t *pgdat, struct zone *zone)
+{
+       int mtype;
+       unsigned long pfn;
+       unsigned long start_pfn = zone->zone_start_pfn;
+       unsigned long end_pfn = start_pfn + zone->spanned_pages;
+       unsigned long count[MIGRATE_TYPES] = { 0, };
+
+       for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
+               struct page *page;
+
+               if (!pfn_valid(pfn))
+                       continue;
+
+               page = pfn_to_page(pfn);
+               mtype = get_pageblock_migratetype(page);
+
+               count[mtype]++;
+       }
+
+       /* Print counts */
+       seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
+       for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
+               seq_printf(m, "%12lu ", count[mtype]);
+       seq_putc(m, '\n');
+}
+
+/* Print out the free pages at each order for each migratetype */
+static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
+{
+       int mtype;
+       pg_data_t *pgdat = (pg_data_t *)arg;
+
+       seq_printf(m, "\n%-23s", "Number of blocks type ");
+       for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
+               seq_printf(m, "%12s ", migratetype_names[mtype]);
+       seq_putc(m, '\n');
+       walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print);
+
+       return 0;
+}
+
+/*
+ * This prints out statistics in relation to grouping pages by mobility.
+ * It is expensive to collect so do not constantly read the file.
+ */
+static int pagetypeinfo_show(struct seq_file *m, void *arg)
+{
+       pg_data_t *pgdat = (pg_data_t *)arg;
+
+       seq_printf(m, "Page block order: %d\n", pageblock_order);
+       seq_printf(m, "Pages per block:  %lu\n", pageblock_nr_pages);
+       seq_putc(m, '\n');
+       pagetypeinfo_showfree(m, pgdat);
+       pagetypeinfo_showblockcount(m, pgdat);
+
        return 0;
 }
 
@@ -454,6 +577,13 @@ const struct seq_operations fragmentation_op = {
        .show   = frag_show,
 };
 
+const struct seq_operations pagetypeinfo_op = {
+       .start  = frag_start,
+       .next   = frag_next,
+       .stop   = frag_stop,
+       .show   = pagetypeinfo_show,
+};
+
 #ifdef CONFIG_ZONE_DMA
 #define TEXT_FOR_DMA(xx) xx "_dma",
 #else
@@ -532,84 +662,78 @@ static const char * const vmstat_text[] = {
 #endif
 };
 
-/*
- * Output information about zones in @pgdat.
- */
-static int zoneinfo_show(struct seq_file *m, void *arg)
+static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
+                                                       struct zone *zone)
 {
-       pg_data_t *pgdat = arg;
-       struct zone *zone;
-       struct zone *node_zones = pgdat->node_zones;
-       unsigned long flags;
-
-       for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
-               int i;
-
-               if (!populated_zone(zone))
-                       continue;
-
-               spin_lock_irqsave(&zone->lock, flags);
-               seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
-               seq_printf(m,
-                          "\n  pages free     %lu"
-                          "\n        min      %lu"
-                          "\n        low      %lu"
-                          "\n        high     %lu"
-                          "\n        scanned  %lu (a: %lu i: %lu)"
-                          "\n        spanned  %lu"
-                          "\n        present  %lu",
-                          zone_page_state(zone, NR_FREE_PAGES),
-                          zone->pages_min,
-                          zone->pages_low,
-                          zone->pages_high,
-                          zone->pages_scanned,
-                          zone->nr_scan_active, zone->nr_scan_inactive,
-                          zone->spanned_pages,
-                          zone->present_pages);
+       int i;
+       seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
+       seq_printf(m,
+                  "\n  pages free     %lu"
+                  "\n        min      %lu"
+                  "\n        low      %lu"
+                  "\n        high     %lu"
+                  "\n        scanned  %lu (a: %lu i: %lu)"
+                  "\n        spanned  %lu"
+                  "\n        present  %lu",
+                  zone_page_state(zone, NR_FREE_PAGES),
+                  zone->pages_min,
+                  zone->pages_low,
+                  zone->pages_high,
+                  zone->pages_scanned,
+                  zone->nr_scan_active, zone->nr_scan_inactive,
+                  zone->spanned_pages,
+                  zone->present_pages);
 
-               for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
-                       seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
-                                       zone_page_state(zone, i));
-
-               seq_printf(m,
-                          "\n        protection: (%lu",
-                          zone->lowmem_reserve[0]);
-               for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
-                       seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
-               seq_printf(m,
-                          ")"
-                          "\n  pagesets");
-               for_each_online_cpu(i) {
-                       struct per_cpu_pageset *pageset;
-                       int j;
-
-                       pageset = zone_pcp(zone, i);
-                       for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
-                               seq_printf(m,
-                                          "\n    cpu: %i pcp: %i"
-                                          "\n              count: %i"
-                                          "\n              high:  %i"
-                                          "\n              batch: %i",
-                                          i, j,
-                                          pageset->pcp[j].count,
-                                          pageset->pcp[j].high,
-                                          pageset->pcp[j].batch);
+       for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
+               seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
+                               zone_page_state(zone, i));
+
+       seq_printf(m,
+                  "\n        protection: (%lu",
+                  zone->lowmem_reserve[0]);
+       for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
+               seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
+       seq_printf(m,
+                  ")"
+                  "\n  pagesets");
+       for_each_online_cpu(i) {
+               struct per_cpu_pageset *pageset;
+               int j;
+
+               pageset = zone_pcp(zone, i);
+               for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
+                       seq_printf(m,
+                                  "\n    cpu: %i pcp: %i"
+                                  "\n              count: %i"
+                                  "\n              high:  %i"
+                                  "\n              batch: %i",
+                                  i, j,
+                                  pageset->pcp[j].count,
+                                  pageset->pcp[j].high,
+                                  pageset->pcp[j].batch);
                        }
 #ifdef CONFIG_SMP
-                       seq_printf(m, "\n  vm stats threshold: %d",
-                                       pageset->stat_threshold);
+               seq_printf(m, "\n  vm stats threshold: %d",
+                               pageset->stat_threshold);
 #endif
-               }
-               seq_printf(m,
-                          "\n  all_unreclaimable: %u"
-                          "\n  prev_priority:     %i"
-                          "\n  start_pfn:         %lu",
-                          zone->all_unreclaimable,
-                          zone->prev_priority,
-                          zone->zone_start_pfn);
-               spin_unlock_irqrestore(&zone->lock, flags);
-               seq_putc(m, '\n');
        }
+       seq_printf(m,
+                  "\n  all_unreclaimable: %u"
+                  "\n  prev_priority:     %i"
+                  "\n  start_pfn:         %lu",
+                  zone->all_unreclaimable,
+                  zone->prev_priority,
+                  zone->zone_start_pfn);
+       seq_putc(m, '\n');
+}
+
+/*
+ * Output information about zones in @pgdat.
+ */
+static int zoneinfo_show(struct seq_file *m, void *arg)
+{
+       pg_data_t *pgdat = (pg_data_t *)arg;
+       walk_zones_in_node(m, pgdat, zoneinfo_show_print);
        return 0;
 }