perf kmem: Analyze page allocator events also
authorNamhyung Kim <namhyung@kernel.org>
Mon, 6 Apr 2015 05:36:10 +0000 (14:36 +0900)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 13 Apr 2015 14:44:52 +0000 (11:44 -0300)
The perf kmem command records and analyze kernel memory allocation only
for SLAB objects.  This patch implement a simple page allocator analyzer
using kmem:mm_page_alloc and kmem:mm_page_free events.

It adds two new options of --slab and --page.  The --slab option is for
analyzing SLAB allocator and that's what perf kmem currently does.

The new --page option enables page allocator events and analyze kernel
memory usage in page unit.  Currently, 'stat --alloc' subcommand is
implemented only.

If none of these --slab nor --page is specified, --slab is implied.

First run 'perf kmem record' to generate a suitable perf.data file:

  # perf kmem record --page sleep 5

Then run 'perf kmem stat' to postprocess the perf.data file:

  # perf kmem stat --page --alloc --line 10

  -------------------------------------------------------------------------------
   PFN              | Total alloc (KB) | Hits     | Order | Mig.type | GFP flags
  -------------------------------------------------------------------------------
            4045014 |               16 |        1 |     2 |  RECLAIM |  00285250
            4143980 |               16 |        1 |     2 |  RECLAIM |  00285250
            3938658 |               16 |        1 |     2 |  RECLAIM |  00285250
            4045400 |               16 |        1 |     2 |  RECLAIM |  00285250
            3568708 |               16 |        1 |     2 |  RECLAIM |  00285250
            3729824 |               16 |        1 |     2 |  RECLAIM |  00285250
            3657210 |               16 |        1 |     2 |  RECLAIM |  00285250
            4120750 |               16 |        1 |     2 |  RECLAIM |  00285250
            3678850 |               16 |        1 |     2 |  RECLAIM |  00285250
            3693874 |               16 |        1 |     2 |  RECLAIM |  00285250
   ...              | ...              | ...      | ...   | ...      | ...
  -------------------------------------------------------------------------------

  SUMMARY (page allocator)
  ========================
  Total allocation requests     :           44,260   [          177,256 KB ]
  Total free requests           :              117   [              468 KB ]

  Total alloc+freed requests    :               49   [              196 KB ]
  Total alloc-only requests     :           44,211   [          177,060 KB ]
  Total free-only requests      :               68   [              272 KB ]

  Total allocation failures     :                0   [                0 KB ]

  Order     Unmovable   Reclaimable       Movable      Reserved  CMA/Isolated
  -----  ------------  ------------  ------------  ------------  ------------
      0            32             .        44,210             .             .
      1             .             .             .             .             .
      2             .            18             .             .             .
      3             .             .             .             .             .
      4             .             .             .             .             .
      5             .             .             .             .             .
      6             .             .             .             .             .
      7             .             .             .             .             .
      8             .             .             .             .             .
      9             .             .             .             .             .
     10             .             .             .             .             .

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/1428298576-9785-4-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/Documentation/perf-kmem.txt
tools/perf/builtin-kmem.c

index 150253cc3c97dd9561610ea780b1b83b871346eb..23219c65c16f77892a30b9b85f38609f1b1dc1f5 100644 (file)
@@ -3,7 +3,7 @@ perf-kmem(1)
 
 NAME
 ----
-perf-kmem - Tool to trace/measure kernel memory(slab) properties
+perf-kmem - Tool to trace/measure kernel memory properties
 
 SYNOPSIS
 --------
@@ -46,6 +46,12 @@ OPTIONS
 --raw-ip::
        Print raw ip instead of symbol
 
+--slab::
+       Analyze SLAB allocator events.
+
+--page::
+       Analyze page allocator events
+
 SEE ALSO
 --------
 linkperf:perf-record[1]
index 4ebf65c7943443291db80f3f3cd36aa9f17e85e6..63ea01349b6e2b6c6bf4fb16ce3722c93d625695 100644 (file)
 #include <linux/string.h>
 #include <locale.h>
 
+static int     kmem_slab;
+static int     kmem_page;
+
+static long    kmem_page_size;
+
 struct alloc_stat;
 typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *);
 
@@ -226,6 +231,244 @@ static int perf_evsel__process_free_event(struct perf_evsel *evsel,
        return 0;
 }
 
+static u64 total_page_alloc_bytes;
+static u64 total_page_free_bytes;
+static u64 total_page_nomatch_bytes;
+static u64 total_page_fail_bytes;
+static unsigned long nr_page_allocs;
+static unsigned long nr_page_frees;
+static unsigned long nr_page_fails;
+static unsigned long nr_page_nomatch;
+
+static bool use_pfn;
+
+#define MAX_MIGRATE_TYPES  6
+#define MAX_PAGE_ORDER     11
+
+static int order_stats[MAX_PAGE_ORDER][MAX_MIGRATE_TYPES];
+
+struct page_stat {
+       struct rb_node  node;
+       u64             page;
+       int             order;
+       unsigned        gfp_flags;
+       unsigned        migrate_type;
+       u64             alloc_bytes;
+       u64             free_bytes;
+       int             nr_alloc;
+       int             nr_free;
+};
+
+static struct rb_root page_tree;
+static struct rb_root page_alloc_tree;
+static struct rb_root page_alloc_sorted;
+
+static struct page_stat *search_page(unsigned long page, bool create)
+{
+       struct rb_node **node = &page_tree.rb_node;
+       struct rb_node *parent = NULL;
+       struct page_stat *data;
+
+       while (*node) {
+               s64 cmp;
+
+               parent = *node;
+               data = rb_entry(*node, struct page_stat, node);
+
+               cmp = data->page - page;
+               if (cmp < 0)
+                       node = &parent->rb_left;
+               else if (cmp > 0)
+                       node = &parent->rb_right;
+               else
+                       return data;
+       }
+
+       if (!create)
+               return NULL;
+
+       data = zalloc(sizeof(*data));
+       if (data != NULL) {
+               data->page = page;
+
+               rb_link_node(&data->node, parent, node);
+               rb_insert_color(&data->node, &page_tree);
+       }
+
+       return data;
+}
+
+static int page_stat_cmp(struct page_stat *a, struct page_stat *b)
+{
+       if (a->page > b->page)
+               return -1;
+       if (a->page < b->page)
+               return 1;
+       if (a->order > b->order)
+               return -1;
+       if (a->order < b->order)
+               return 1;
+       if (a->migrate_type > b->migrate_type)
+               return -1;
+       if (a->migrate_type < b->migrate_type)
+               return 1;
+       if (a->gfp_flags > b->gfp_flags)
+               return -1;
+       if (a->gfp_flags < b->gfp_flags)
+               return 1;
+       return 0;
+}
+
+static struct page_stat *search_page_alloc_stat(struct page_stat *stat, bool create)
+{
+       struct rb_node **node = &page_alloc_tree.rb_node;
+       struct rb_node *parent = NULL;
+       struct page_stat *data;
+
+       while (*node) {
+               s64 cmp;
+
+               parent = *node;
+               data = rb_entry(*node, struct page_stat, node);
+
+               cmp = page_stat_cmp(data, stat);
+               if (cmp < 0)
+                       node = &parent->rb_left;
+               else if (cmp > 0)
+                       node = &parent->rb_right;
+               else
+                       return data;
+       }
+
+       if (!create)
+               return NULL;
+
+       data = zalloc(sizeof(*data));
+       if (data != NULL) {
+               data->page = stat->page;
+               data->order = stat->order;
+               data->gfp_flags = stat->gfp_flags;
+               data->migrate_type = stat->migrate_type;
+
+               rb_link_node(&data->node, parent, node);
+               rb_insert_color(&data->node, &page_alloc_tree);
+       }
+
+       return data;
+}
+
+static bool valid_page(u64 pfn_or_page)
+{
+       if (use_pfn && pfn_or_page == -1UL)
+               return false;
+       if (!use_pfn && pfn_or_page == 0)
+               return false;
+       return true;
+}
+
+static int perf_evsel__process_page_alloc_event(struct perf_evsel *evsel,
+                                               struct perf_sample *sample)
+{
+       u64 page;
+       unsigned int order = perf_evsel__intval(evsel, sample, "order");
+       unsigned int gfp_flags = perf_evsel__intval(evsel, sample, "gfp_flags");
+       unsigned int migrate_type = perf_evsel__intval(evsel, sample,
+                                                      "migratetype");
+       u64 bytes = kmem_page_size << order;
+       struct page_stat *stat;
+       struct page_stat this = {
+               .order = order,
+               .gfp_flags = gfp_flags,
+               .migrate_type = migrate_type,
+       };
+
+       if (use_pfn)
+               page = perf_evsel__intval(evsel, sample, "pfn");
+       else
+               page = perf_evsel__intval(evsel, sample, "page");
+
+       nr_page_allocs++;
+       total_page_alloc_bytes += bytes;
+
+       if (!valid_page(page)) {
+               nr_page_fails++;
+               total_page_fail_bytes += bytes;
+
+               return 0;
+       }
+
+       /*
+        * This is to find the current page (with correct gfp flags and
+        * migrate type) at free event.
+        */
+       stat = search_page(page, true);
+       if (stat == NULL)
+               return -ENOMEM;
+
+       stat->order = order;
+       stat->gfp_flags = gfp_flags;
+       stat->migrate_type = migrate_type;
+
+       this.page = page;
+       stat = search_page_alloc_stat(&this, true);
+       if (stat == NULL)
+               return -ENOMEM;
+
+       stat->nr_alloc++;
+       stat->alloc_bytes += bytes;
+
+       order_stats[order][migrate_type]++;
+
+       return 0;
+}
+
+static int perf_evsel__process_page_free_event(struct perf_evsel *evsel,
+                                               struct perf_sample *sample)
+{
+       u64 page;
+       unsigned int order = perf_evsel__intval(evsel, sample, "order");
+       u64 bytes = kmem_page_size << order;
+       struct page_stat *stat;
+       struct page_stat this = {
+               .order = order,
+       };
+
+       if (use_pfn)
+               page = perf_evsel__intval(evsel, sample, "pfn");
+       else
+               page = perf_evsel__intval(evsel, sample, "page");
+
+       nr_page_frees++;
+       total_page_free_bytes += bytes;
+
+       stat = search_page(page, false);
+       if (stat == NULL) {
+               pr_debug2("missing free at page %"PRIx64" (order: %d)\n",
+                         page, order);
+
+               nr_page_nomatch++;
+               total_page_nomatch_bytes += bytes;
+
+               return 0;
+       }
+
+       this.page = page;
+       this.gfp_flags = stat->gfp_flags;
+       this.migrate_type = stat->migrate_type;
+
+       rb_erase(&stat->node, &page_tree);
+       free(stat);
+
+       stat = search_page_alloc_stat(&this, false);
+       if (stat == NULL)
+               return -ENOENT;
+
+       stat->nr_free++;
+       stat->free_bytes += bytes;
+
+       return 0;
+}
+
 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
                                  struct perf_sample *sample);
 
@@ -270,8 +513,9 @@ static double fragmentation(unsigned long n_req, unsigned long n_alloc)
                return 100.0 - (100.0 * n_req / n_alloc);
 }
 
-static void __print_result(struct rb_root *root, struct perf_session *session,
-                          int n_lines, int is_caller)
+static void __print_slab_result(struct rb_root *root,
+                               struct perf_session *session,
+                               int n_lines, int is_caller)
 {
        struct rb_node *next;
        struct machine *machine = &session->machines.host;
@@ -323,9 +567,56 @@ static void __print_result(struct rb_root *root, struct perf_session *session,
        printf("%.105s\n", graph_dotted_line);
 }
 
-static void print_summary(void)
+static const char * const migrate_type_str[] = {
+       "UNMOVABL",
+       "RECLAIM",
+       "MOVABLE",
+       "RESERVED",
+       "CMA/ISLT",
+       "UNKNOWN",
+};
+
+static void __print_page_result(struct rb_root *root,
+                               struct perf_session *session __maybe_unused,
+                               int n_lines)
+{
+       struct rb_node *next = rb_first(root);
+       const char *format;
+
+       printf("\n%.80s\n", graph_dotted_line);
+       printf(" %-16s | Total alloc (KB) | Hits      | Order | Mig.type | GFP flags\n",
+              use_pfn ? "PFN" : "Page");
+       printf("%.80s\n", graph_dotted_line);
+
+       if (use_pfn)
+               format = " %16llu | %'16llu | %'9d | %5d | %8s |  %08lx\n";
+       else
+               format = " %016llx | %'16llu | %'9d | %5d | %8s |  %08lx\n";
+
+       while (next && n_lines--) {
+               struct page_stat *data;
+
+               data = rb_entry(next, struct page_stat, node);
+
+               printf(format, (unsigned long long)data->page,
+                      (unsigned long long)data->alloc_bytes / 1024,
+                      data->nr_alloc, data->order,
+                      migrate_type_str[data->migrate_type],
+                      (unsigned long)data->gfp_flags);
+
+               next = rb_next(next);
+       }
+
+       if (n_lines == -1)
+               printf(" ...              | ...              | ...       | ...   | ...      | ...     \n");
+
+       printf("%.80s\n", graph_dotted_line);
+}
+
+static void print_slab_summary(void)
 {
-       printf("\nSUMMARY\n=======\n");
+       printf("\nSUMMARY (SLAB allocator)");
+       printf("\n========================\n");
        printf("Total bytes requested: %'lu\n", total_requested);
        printf("Total bytes allocated: %'lu\n", total_allocated);
        printf("Total bytes wasted on internal fragmentation: %'lu\n",
@@ -335,13 +626,73 @@ static void print_summary(void)
        printf("Cross CPU allocations: %'lu/%'lu\n", nr_cross_allocs, nr_allocs);
 }
 
-static void print_result(struct perf_session *session)
+static void print_page_summary(void)
+{
+       int o, m;
+       u64 nr_alloc_freed = nr_page_frees - nr_page_nomatch;
+       u64 total_alloc_freed_bytes = total_page_free_bytes - total_page_nomatch_bytes;
+
+       printf("\nSUMMARY (page allocator)");
+       printf("\n========================\n");
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total allocation requests",
+              nr_page_allocs, total_page_alloc_bytes / 1024);
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total free requests",
+              nr_page_frees, total_page_free_bytes / 1024);
+       printf("\n");
+
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total alloc+freed requests",
+              nr_alloc_freed, (total_alloc_freed_bytes) / 1024);
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total alloc-only requests",
+              nr_page_allocs - nr_alloc_freed,
+              (total_page_alloc_bytes - total_alloc_freed_bytes) / 1024);
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total free-only requests",
+              nr_page_nomatch, total_page_nomatch_bytes / 1024);
+       printf("\n");
+
+       printf("%-30s: %'16lu   [ %'16"PRIu64" KB ]\n", "Total allocation failures",
+              nr_page_fails, total_page_fail_bytes / 1024);
+       printf("\n");
+
+       printf("%5s  %12s  %12s  %12s  %12s  %12s\n", "Order",  "Unmovable",
+              "Reclaimable", "Movable", "Reserved", "CMA/Isolated");
+       printf("%.5s  %.12s  %.12s  %.12s  %.12s  %.12s\n", graph_dotted_line,
+              graph_dotted_line, graph_dotted_line, graph_dotted_line,
+              graph_dotted_line, graph_dotted_line);
+
+       for (o = 0; o < MAX_PAGE_ORDER; o++) {
+               printf("%5d", o);
+               for (m = 0; m < MAX_MIGRATE_TYPES - 1; m++) {
+                       if (order_stats[o][m])
+                               printf("  %'12d", order_stats[o][m]);
+                       else
+                               printf("  %12c", '.');
+               }
+               printf("\n");
+       }
+}
+
+static void print_slab_result(struct perf_session *session)
 {
        if (caller_flag)
-               __print_result(&root_caller_sorted, session, caller_lines, 1);
+               __print_slab_result(&root_caller_sorted, session, caller_lines, 1);
+       if (alloc_flag)
+               __print_slab_result(&root_alloc_sorted, session, alloc_lines, 0);
+       print_slab_summary();
+}
+
+static void print_page_result(struct perf_session *session)
+{
        if (alloc_flag)
-               __print_result(&root_alloc_sorted, session, alloc_lines, 0);
-       print_summary();
+               __print_page_result(&page_alloc_sorted, session, alloc_lines);
+       print_page_summary();
+}
+
+static void print_result(struct perf_session *session)
+{
+       if (kmem_slab)
+               print_slab_result(session);
+       if (kmem_page)
+               print_page_result(session);
 }
 
 struct sort_dimension {
@@ -353,8 +704,8 @@ struct sort_dimension {
 static LIST_HEAD(caller_sort);
 static LIST_HEAD(alloc_sort);
 
-static void sort_insert(struct rb_root *root, struct alloc_stat *data,
-                       struct list_head *sort_list)
+static void sort_slab_insert(struct rb_root *root, struct alloc_stat *data,
+                            struct list_head *sort_list)
 {
        struct rb_node **new = &(root->rb_node);
        struct rb_node *parent = NULL;
@@ -383,8 +734,8 @@ static void sort_insert(struct rb_root *root, struct alloc_stat *data,
        rb_insert_color(&data->node, root);
 }
 
-static void __sort_result(struct rb_root *root, struct rb_root *root_sorted,
-                         struct list_head *sort_list)
+static void __sort_slab_result(struct rb_root *root, struct rb_root *root_sorted,
+                              struct list_head *sort_list)
 {
        struct rb_node *node;
        struct alloc_stat *data;
@@ -396,26 +747,79 @@ static void __sort_result(struct rb_root *root, struct rb_root *root_sorted,
 
                rb_erase(node, root);
                data = rb_entry(node, struct alloc_stat, node);
-               sort_insert(root_sorted, data, sort_list);
+               sort_slab_insert(root_sorted, data, sort_list);
+       }
+}
+
+static void sort_page_insert(struct rb_root *root, struct page_stat *data)
+{
+       struct rb_node **new = &root->rb_node;
+       struct rb_node *parent = NULL;
+
+       while (*new) {
+               struct page_stat *this;
+               int cmp = 0;
+
+               this = rb_entry(*new, struct page_stat, node);
+               parent = *new;
+
+               /* TODO: support more sort key */
+               cmp = data->alloc_bytes - this->alloc_bytes;
+
+               if (cmp > 0)
+                       new = &parent->rb_left;
+               else
+                       new = &parent->rb_right;
+       }
+
+       rb_link_node(&data->node, parent, new);
+       rb_insert_color(&data->node, root);
+}
+
+static void __sort_page_result(struct rb_root *root, struct rb_root *root_sorted)
+{
+       struct rb_node *node;
+       struct page_stat *data;
+
+       for (;;) {
+               node = rb_first(root);
+               if (!node)
+                       break;
+
+               rb_erase(node, root);
+               data = rb_entry(node, struct page_stat, node);
+               sort_page_insert(root_sorted, data);
        }
 }
 
 static void sort_result(void)
 {
-       __sort_result(&root_alloc_stat, &root_alloc_sorted, &alloc_sort);
-       __sort_result(&root_caller_stat, &root_caller_sorted, &caller_sort);
+       if (kmem_slab) {
+               __sort_slab_result(&root_alloc_stat, &root_alloc_sorted,
+                                  &alloc_sort);
+               __sort_slab_result(&root_caller_stat, &root_caller_sorted,
+                                  &caller_sort);
+       }
+       if (kmem_page) {
+               __sort_page_result(&page_alloc_tree, &page_alloc_sorted);
+       }
 }
 
 static int __cmd_kmem(struct perf_session *session)
 {
        int err = -EINVAL;
+       struct perf_evsel *evsel;
        const struct perf_evsel_str_handler kmem_tracepoints[] = {
+               /* slab allocator */
                { "kmem:kmalloc",               perf_evsel__process_alloc_event, },
                { "kmem:kmem_cache_alloc",      perf_evsel__process_alloc_event, },
                { "kmem:kmalloc_node",          perf_evsel__process_alloc_node_event, },
                { "kmem:kmem_cache_alloc_node", perf_evsel__process_alloc_node_event, },
                { "kmem:kfree",                 perf_evsel__process_free_event, },
                { "kmem:kmem_cache_free",       perf_evsel__process_free_event, },
+               /* page allocator */
+               { "kmem:mm_page_alloc",         perf_evsel__process_page_alloc_event, },
+               { "kmem:mm_page_free",          perf_evsel__process_page_free_event, },
        };
 
        if (!perf_session__has_traces(session, "kmem record"))
@@ -426,10 +830,20 @@ static int __cmd_kmem(struct perf_session *session)
                goto out;
        }
 
+       evlist__for_each(session->evlist, evsel) {
+               if (!strcmp(perf_evsel__name(evsel), "kmem:mm_page_alloc") &&
+                   perf_evsel__field(evsel, "pfn")) {
+                       use_pfn = true;
+                       break;
+               }
+       }
+
        setup_pager();
        err = perf_session__process_events(session);
-       if (err != 0)
+       if (err != 0) {
+               pr_err("error during process events: %d\n", err);
                goto out;
+       }
        sort_result();
        print_result(session);
 out:
@@ -612,6 +1026,22 @@ static int parse_alloc_opt(const struct option *opt __maybe_unused,
        return 0;
 }
 
+static int parse_slab_opt(const struct option *opt __maybe_unused,
+                         const char *arg __maybe_unused,
+                         int unset __maybe_unused)
+{
+       kmem_slab = (kmem_page + 1);
+       return 0;
+}
+
+static int parse_page_opt(const struct option *opt __maybe_unused,
+                         const char *arg __maybe_unused,
+                         int unset __maybe_unused)
+{
+       kmem_page = (kmem_slab + 1);
+       return 0;
+}
+
 static int parse_line_opt(const struct option *opt __maybe_unused,
                          const char *arg, int unset __maybe_unused)
 {
@@ -634,6 +1064,8 @@ static int __cmd_record(int argc, const char **argv)
 {
        const char * const record_args[] = {
        "record", "-a", "-R", "-c", "1",
+       };
+       const char * const slab_events[] = {
        "-e", "kmem:kmalloc",
        "-e", "kmem:kmalloc_node",
        "-e", "kmem:kfree",
@@ -641,10 +1073,19 @@ static int __cmd_record(int argc, const char **argv)
        "-e", "kmem:kmem_cache_alloc_node",
        "-e", "kmem:kmem_cache_free",
        };
+       const char * const page_events[] = {
+       "-e", "kmem:mm_page_alloc",
+       "-e", "kmem:mm_page_free",
+       };
        unsigned int rec_argc, i, j;
        const char **rec_argv;
 
        rec_argc = ARRAY_SIZE(record_args) + argc - 1;
+       if (kmem_slab)
+               rec_argc += ARRAY_SIZE(slab_events);
+       if (kmem_page)
+               rec_argc += ARRAY_SIZE(page_events);
+
        rec_argv = calloc(rec_argc + 1, sizeof(char *));
 
        if (rec_argv == NULL)
@@ -653,6 +1094,15 @@ static int __cmd_record(int argc, const char **argv)
        for (i = 0; i < ARRAY_SIZE(record_args); i++)
                rec_argv[i] = strdup(record_args[i]);
 
+       if (kmem_slab) {
+               for (j = 0; j < ARRAY_SIZE(slab_events); j++, i++)
+                       rec_argv[i] = strdup(slab_events[j]);
+       }
+       if (kmem_page) {
+               for (j = 0; j < ARRAY_SIZE(page_events); j++, i++)
+                       rec_argv[i] = strdup(page_events[j]);
+       }
+
        for (j = 1; j < (unsigned int)argc; j++, i++)
                rec_argv[i] = argv[j];
 
@@ -679,6 +1129,10 @@ int cmd_kmem(int argc, const char **argv, const char *prefix __maybe_unused)
        OPT_CALLBACK('l', "line", NULL, "num", "show n lines", parse_line_opt),
        OPT_BOOLEAN(0, "raw-ip", &raw_ip, "show raw ip instead of symbol"),
        OPT_BOOLEAN('f', "force", &file.force, "don't complain, do it"),
+       OPT_CALLBACK_NOOPT(0, "slab", NULL, NULL, "Analyze slab allocator",
+                          parse_slab_opt),
+       OPT_CALLBACK_NOOPT(0, "page", NULL, NULL, "Analyze page allocator",
+                          parse_page_opt),
        OPT_END()
        };
        const char *const kmem_subcommands[] = { "record", "stat", NULL };
@@ -695,6 +1149,9 @@ int cmd_kmem(int argc, const char **argv, const char *prefix __maybe_unused)
        if (!argc)
                usage_with_options(kmem_usage, kmem_options);
 
+       if (kmem_slab == 0 && kmem_page == 0)
+               kmem_slab = 1;  /* for backward compatibility */
+
        if (!strncmp(argv[0], "rec", 3)) {
                symbol__init(NULL);
                return __cmd_record(argc, argv);
@@ -706,6 +1163,17 @@ int cmd_kmem(int argc, const char **argv, const char *prefix __maybe_unused)
        if (session == NULL)
                return -1;
 
+       if (kmem_page) {
+               struct perf_evsel *evsel = perf_evlist__first(session->evlist);
+
+               if (evsel == NULL || evsel->tp_format == NULL) {
+                       pr_err("invalid event found.. aborting\n");
+                       return -1;
+               }
+
+               kmem_page_size = pevent_get_page_size(evsel->tp_format->pevent);
+       }
+
        symbol__init(&session->header.env);
 
        if (!strcmp(argv[0], "stat")) {