perf_counter tools: Prepare for 'perf annotate'
authorIngo Molnar <mingo@elte.hu>
Sat, 6 Jun 2009 13:19:13 +0000 (15:19 +0200)
committerIngo Molnar <mingo@elte.hu>
Sat, 6 Jun 2009 16:58:30 +0000 (18:58 +0200)
Prepare for the 'perf annotate' implementation by splitting off
builtin-annotate.c from builtin-report.c.

( We keep this commit separate to ease the later librarization
  of the facilities that perf-report and perf-annotate shares. )

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Documentation/perf_counter/Documentation/perf-annotate.txt [new file with mode: 0644]
Documentation/perf_counter/Makefile
Documentation/perf_counter/builtin-annotate.c [new file with mode: 0644]
Documentation/perf_counter/builtin.h
Documentation/perf_counter/command-list.txt
Documentation/perf_counter/perf.c

diff --git a/Documentation/perf_counter/Documentation/perf-annotate.txt b/Documentation/perf_counter/Documentation/perf-annotate.txt
new file mode 100644 (file)
index 0000000..a9d6d5e
--- /dev/null
@@ -0,0 +1,26 @@
+perf-annotate(1)
+==============
+
+NAME
+----
+perf-annotate - Read perf.data (created by perf record) and annotate functions
+
+SYNOPSIS
+--------
+[verse]
+'perf annotate' [-i <file> | --input=file] symbol_name
+
+DESCRIPTION
+-----------
+This command displays the performance counter profile information recorded
+via perf record.
+
+OPTIONS
+-------
+-i::
+--input=::
+        Input file name. (default: perf.data)
+
+SEE ALSO
+--------
+linkperf:perf-record[1]
index 32c0bb2..0cbd5d6 100644 (file)
@@ -323,12 +323,13 @@ LIB_OBJS += util/symbol.o
 LIB_OBJS += util/color.o
 LIB_OBJS += util/pager.o
 
+BUILTIN_OBJS += builtin-annotate.o
 BUILTIN_OBJS += builtin-help.o
+BUILTIN_OBJS += builtin-list.o
 BUILTIN_OBJS += builtin-record.o
 BUILTIN_OBJS += builtin-report.o
 BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
-BUILTIN_OBJS += builtin-list.o
 
 PERFLIBS = $(LIB_FILE)
 EXTLIBS =
diff --git a/Documentation/perf_counter/builtin-annotate.c b/Documentation/perf_counter/builtin-annotate.c
new file mode 100644 (file)
index 0000000..d656484
--- /dev/null
@@ -0,0 +1,1291 @@
+/*
+ * builtin-annotate.c
+ *
+ * Builtin annotate command: Analyze the perf.data input file,
+ * look up and read DSOs and symbol information and display
+ * a histogram of results, along various sorting keys.
+ */
+#include "builtin.h"
+
+#include "util/util.h"
+
+#include "util/color.h"
+#include "util/list.h"
+#include "util/cache.h"
+#include "util/rbtree.h"
+#include "util/symbol.h"
+#include "util/string.h"
+
+#include "perf.h"
+
+#include "util/parse-options.h"
+#include "util/parse-events.h"
+
+#define SHOW_KERNEL    1
+#define SHOW_USER      2
+#define SHOW_HV                4
+
+static char            const *input_name = "perf.data";
+static char            *vmlinux = NULL;
+
+static char            default_sort_order[] = "comm,dso";
+static char            *sort_order = default_sort_order;
+
+static int             input;
+static int             show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
+
+static int             dump_trace = 0;
+#define dprintf(x...)  do { if (dump_trace) printf(x); } while (0)
+
+static int             verbose;
+static int             full_paths;
+
+static unsigned long   page_size;
+static unsigned long   mmap_window = 32;
+
+struct ip_event {
+       struct perf_event_header header;
+       __u64 ip;
+       __u32 pid, tid;
+};
+
+struct mmap_event {
+       struct perf_event_header header;
+       __u32 pid, tid;
+       __u64 start;
+       __u64 len;
+       __u64 pgoff;
+       char filename[PATH_MAX];
+};
+
+struct comm_event {
+       struct perf_event_header header;
+       __u32 pid, tid;
+       char comm[16];
+};
+
+struct fork_event {
+       struct perf_event_header header;
+       __u32 pid, ppid;
+};
+
+struct period_event {
+       struct perf_event_header header;
+       __u64 time;
+       __u64 id;
+       __u64 sample_period;
+};
+
+typedef union event_union {
+       struct perf_event_header        header;
+       struct ip_event                 ip;
+       struct mmap_event               mmap;
+       struct comm_event               comm;
+       struct fork_event               fork;
+       struct period_event             period;
+} event_t;
+
+static LIST_HEAD(dsos);
+static struct dso *kernel_dso;
+static struct dso *vdso;
+
+static void dsos__add(struct dso *dso)
+{
+       list_add_tail(&dso->node, &dsos);
+}
+
+static struct dso *dsos__find(const char *name)
+{
+       struct dso *pos;
+
+       list_for_each_entry(pos, &dsos, node)
+               if (strcmp(pos->name, name) == 0)
+                       return pos;
+       return NULL;
+}
+
+static struct dso *dsos__findnew(const char *name)
+{
+       struct dso *dso = dsos__find(name);
+       int nr;
+
+       if (dso)
+               return dso;
+
+       dso = dso__new(name, 0);
+       if (!dso)
+               goto out_delete_dso;
+
+       nr = dso__load(dso, NULL, verbose);
+       if (nr < 0) {
+               if (verbose)
+                       fprintf(stderr, "Failed to open: %s\n", name);
+               goto out_delete_dso;
+       }
+       if (!nr && verbose) {
+               fprintf(stderr,
+               "No symbols found in: %s, maybe install a debug package?\n",
+                               name);
+       }
+
+       dsos__add(dso);
+
+       return dso;
+
+out_delete_dso:
+       dso__delete(dso);
+       return NULL;
+}
+
+static void dsos__fprintf(FILE *fp)
+{
+       struct dso *pos;
+
+       list_for_each_entry(pos, &dsos, node)
+               dso__fprintf(pos, fp);
+}
+
+static struct symbol *vdso__find_symbol(struct dso *dso, uint64_t ip)
+{
+       return dso__find_symbol(kernel_dso, ip);
+}
+
+static int load_kernel(void)
+{
+       int err;
+
+       kernel_dso = dso__new("[kernel]", 0);
+       if (!kernel_dso)
+               return -1;
+
+       err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
+       if (err) {
+               dso__delete(kernel_dso);
+               kernel_dso = NULL;
+       } else
+               dsos__add(kernel_dso);
+
+       vdso = dso__new("[vdso]", 0);
+       if (!vdso)
+               return -1;
+
+       vdso->find_symbol = vdso__find_symbol;
+
+       dsos__add(vdso);
+
+       return err;
+}
+
+static char __cwd[PATH_MAX];
+static char *cwd = __cwd;
+static int cwdlen;
+
+static int strcommon(const char *pathname)
+{
+       int n = 0;
+
+       while (pathname[n] == cwd[n] && n < cwdlen)
+               ++n;
+
+       return n;
+}
+
+struct map {
+       struct list_head node;
+       uint64_t         start;
+       uint64_t         end;
+       uint64_t         pgoff;
+       uint64_t         (*map_ip)(struct map *, uint64_t);
+       struct dso       *dso;
+};
+
+static uint64_t map__map_ip(struct map *map, uint64_t ip)
+{
+       return ip - map->start + map->pgoff;
+}
+
+static uint64_t vdso__map_ip(struct map *map, uint64_t ip)
+{
+       return ip;
+}
+
+static struct map *map__new(struct mmap_event *event)
+{
+       struct map *self = malloc(sizeof(*self));
+
+       if (self != NULL) {
+               const char *filename = event->filename;
+               char newfilename[PATH_MAX];
+
+               if (cwd) {
+                       int n = strcommon(filename);
+
+                       if (n == cwdlen) {
+                               snprintf(newfilename, sizeof(newfilename),
+                                        ".%s", filename + n);
+                               filename = newfilename;
+                       }
+               }
+
+               self->start = event->start;
+               self->end   = event->start + event->len;
+               self->pgoff = event->pgoff;
+
+               self->dso = dsos__findnew(filename);
+               if (self->dso == NULL)
+                       goto out_delete;
+
+               if (self->dso == vdso)
+                       self->map_ip = vdso__map_ip;
+               else
+                       self->map_ip = map__map_ip;
+       }
+       return self;
+out_delete:
+       free(self);
+       return NULL;
+}
+
+static struct map *map__clone(struct map *self)
+{
+       struct map *map = malloc(sizeof(*self));
+
+       if (!map)
+               return NULL;
+
+       memcpy(map, self, sizeof(*self));
+
+       return map;
+}
+
+static int map__overlap(struct map *l, struct map *r)
+{
+       if (l->start > r->start) {
+               struct map *t = l;
+               l = r;
+               r = t;
+       }
+
+       if (l->end > r->start)
+               return 1;
+
+       return 0;
+}
+
+static size_t map__fprintf(struct map *self, FILE *fp)
+{
+       return fprintf(fp, " %"PRIx64"-%"PRIx64" %"PRIx64" %s\n",
+                      self->start, self->end, self->pgoff, self->dso->name);
+}
+
+
+struct thread {
+       struct rb_node   rb_node;
+       struct list_head maps;
+       pid_t            pid;
+       char             *comm;
+};
+
+static struct thread *thread__new(pid_t pid)
+{
+       struct thread *self = malloc(sizeof(*self));
+
+       if (self != NULL) {
+               self->pid = pid;
+               self->comm = malloc(32);
+               if (self->comm)
+                       snprintf(self->comm, 32, ":%d", self->pid);
+               INIT_LIST_HEAD(&self->maps);
+       }
+
+       return self;
+}
+
+static int thread__set_comm(struct thread *self, const char *comm)
+{
+       if (self->comm)
+               free(self->comm);
+       self->comm = strdup(comm);
+       return self->comm ? 0 : -ENOMEM;
+}
+
+static size_t thread__fprintf(struct thread *self, FILE *fp)
+{
+       struct map *pos;
+       size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
+
+       list_for_each_entry(pos, &self->maps, node)
+               ret += map__fprintf(pos, fp);
+
+       return ret;
+}
+
+
+static struct rb_root threads;
+static struct thread *last_match;
+
+static struct thread *threads__findnew(pid_t pid)
+{
+       struct rb_node **p = &threads.rb_node;
+       struct rb_node *parent = NULL;
+       struct thread *th;
+
+       /*
+        * Font-end cache - PID lookups come in blocks,
+        * so most of the time we dont have to look up
+        * the full rbtree:
+        */
+       if (last_match && last_match->pid == pid)
+               return last_match;
+
+       while (*p != NULL) {
+               parent = *p;
+               th = rb_entry(parent, struct thread, rb_node);
+
+               if (th->pid == pid) {
+                       last_match = th;
+                       return th;
+               }
+
+               if (pid < th->pid)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+
+       th = thread__new(pid);
+       if (th != NULL) {
+               rb_link_node(&th->rb_node, parent, p);
+               rb_insert_color(&th->rb_node, &threads);
+               last_match = th;
+       }
+
+       return th;
+}
+
+static void thread__insert_map(struct thread *self, struct map *map)
+{
+       struct map *pos, *tmp;
+
+       list_for_each_entry_safe(pos, tmp, &self->maps, node) {
+               if (map__overlap(pos, map)) {
+                       list_del_init(&pos->node);
+                       /* XXX leaks dsos */
+                       free(pos);
+               }
+       }
+
+       list_add_tail(&map->node, &self->maps);
+}
+
+static int thread__fork(struct thread *self, struct thread *parent)
+{
+       struct map *map;
+
+       if (self->comm)
+               free(self->comm);
+       self->comm = strdup(parent->comm);
+       if (!self->comm)
+               return -ENOMEM;
+
+       list_for_each_entry(map, &parent->maps, node) {
+               struct map *new = map__clone(map);
+               if (!new)
+                       return -ENOMEM;
+               thread__insert_map(self, new);
+       }
+
+       return 0;
+}
+
+static struct map *thread__find_map(struct thread *self, uint64_t ip)
+{
+       struct map *pos;
+
+       if (self == NULL)
+               return NULL;
+
+       list_for_each_entry(pos, &self->maps, node)
+               if (ip >= pos->start && ip <= pos->end)
+                       return pos;
+
+       return NULL;
+}
+
+static size_t threads__fprintf(FILE *fp)
+{
+       size_t ret = 0;
+       struct rb_node *nd;
+
+       for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
+               struct thread *pos = rb_entry(nd, struct thread, rb_node);
+
+               ret += thread__fprintf(pos, fp);
+       }
+
+       return ret;
+}
+
+/*
+ * histogram, sorted on item, collects counts
+ */
+
+static struct rb_root hist;
+
+struct hist_entry {
+       struct rb_node   rb_node;
+
+       struct thread    *thread;
+       struct map       *map;
+       struct dso       *dso;
+       struct symbol    *sym;
+       uint64_t         ip;
+       char             level;
+
+       uint32_t         count;
+};
+
+/*
+ * configurable sorting bits
+ */
+
+struct sort_entry {
+       struct list_head list;
+
+       char *header;
+
+       int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
+       int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
+       size_t  (*print)(FILE *fp, struct hist_entry *);
+};
+
+/* --sort pid */
+
+static int64_t
+sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
+{
+       return right->thread->pid - left->thread->pid;
+}
+
+static size_t
+sort__thread_print(FILE *fp, struct hist_entry *self)
+{
+       return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
+}
+
+static struct sort_entry sort_thread = {
+       .header = "         Command:  Pid",
+       .cmp    = sort__thread_cmp,
+       .print  = sort__thread_print,
+};
+
+/* --sort comm */
+
+static int64_t
+sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
+{
+       return right->thread->pid - left->thread->pid;
+}
+
+static int64_t
+sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
+{
+       char *comm_l = left->thread->comm;
+       char *comm_r = right->thread->comm;
+
+       if (!comm_l || !comm_r) {
+               if (!comm_l && !comm_r)
+                       return 0;
+               else if (!comm_l)
+                       return -1;
+               else
+                       return 1;
+       }
+
+       return strcmp(comm_l, comm_r);
+}
+
+static size_t
+sort__comm_print(FILE *fp, struct hist_entry *self)
+{
+       return fprintf(fp, "%16s", self->thread->comm);
+}
+
+static struct sort_entry sort_comm = {
+       .header         = "         Command",
+       .cmp            = sort__comm_cmp,
+       .collapse       = sort__comm_collapse,
+       .print          = sort__comm_print,
+};
+
+/* --sort dso */
+
+static int64_t
+sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
+{
+       struct dso *dso_l = left->dso;
+       struct dso *dso_r = right->dso;
+
+       if (!dso_l || !dso_r) {
+               if (!dso_l && !dso_r)
+                       return 0;
+               else if (!dso_l)
+                       return -1;
+               else
+                       return 1;
+       }
+
+       return strcmp(dso_l->name, dso_r->name);
+}
+
+static size_t
+sort__dso_print(FILE *fp, struct hist_entry *self)
+{
+       if (self->dso)
+               return fprintf(fp, "%-25s", self->dso->name);
+
+       return fprintf(fp, "%016llx         ", (__u64)self->ip);
+}
+
+static struct sort_entry sort_dso = {
+       .header = "Shared Object            ",
+       .cmp    = sort__dso_cmp,
+       .print  = sort__dso_print,
+};
+
+/* --sort symbol */
+
+static int64_t
+sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
+{
+       uint64_t ip_l, ip_r;
+
+       if (left->sym == right->sym)
+               return 0;
+
+       ip_l = left->sym ? left->sym->start : left->ip;
+       ip_r = right->sym ? right->sym->start : right->ip;
+
+       return (int64_t)(ip_r - ip_l);
+}
+
+static size_t
+sort__sym_print(FILE *fp, struct hist_entry *self)
+{
+       size_t ret = 0;
+
+       if (verbose)
+               ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
+
+       if (self->sym) {
+               ret += fprintf(fp, "[%c] %s",
+                       self->dso == kernel_dso ? 'k' : '.', self->sym->name);
+       } else {
+               ret += fprintf(fp, "%#016llx", (__u64)self->ip);
+       }
+
+       return ret;
+}
+
+static struct sort_entry sort_sym = {
+       .header = "Symbol",
+       .cmp    = sort__sym_cmp,
+       .print  = sort__sym_print,
+};
+
+static int sort__need_collapse = 0;
+
+struct sort_dimension {
+       char                    *name;
+       struct sort_entry       *entry;
+       int                     taken;
+};
+
+static struct sort_dimension sort_dimensions[] = {
+       { .name = "pid",        .entry = &sort_thread,  },
+       { .name = "comm",       .entry = &sort_comm,    },
+       { .name = "dso",        .entry = &sort_dso,     },
+       { .name = "symbol",     .entry = &sort_sym,     },
+};
+
+static LIST_HEAD(hist_entry__sort_list);
+
+static int sort_dimension__add(char *tok)
+{
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
+               struct sort_dimension *sd = &sort_dimensions[i];
+
+               if (sd->taken)
+                       continue;
+
+               if (strncasecmp(tok, sd->name, strlen(tok)))
+                       continue;
+
+               if (sd->entry->collapse)
+                       sort__need_collapse = 1;
+
+               list_add_tail(&sd->entry->list, &hist_entry__sort_list);
+               sd->taken = 1;
+
+               return 0;
+       }
+
+       return -ESRCH;
+}
+
+static int64_t
+hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
+{
+       struct sort_entry *se;
+       int64_t cmp = 0;
+
+       list_for_each_entry(se, &hist_entry__sort_list, list) {
+               cmp = se->cmp(left, right);
+               if (cmp)
+                       break;
+       }
+
+       return cmp;
+}
+
+static int64_t
+hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
+{
+       struct sort_entry *se;
+       int64_t cmp = 0;
+
+       list_for_each_entry(se, &hist_entry__sort_list, list) {
+               int64_t (*f)(struct hist_entry *, struct hist_entry *);
+
+               f = se->collapse ?: se->cmp;
+
+               cmp = f(left, right);
+               if (cmp)
+                       break;
+       }
+
+       return cmp;
+}
+
+static size_t
+hist_entry__fprintf(FILE *fp, struct hist_entry *self, uint64_t total_samples)
+{
+       struct sort_entry *se;
+       size_t ret;
+
+       if (total_samples) {
+               double percent = self->count * 100.0 / total_samples;
+               char *color = PERF_COLOR_NORMAL;
+
+               /*
+                * We color high-overhead entries in red, low-overhead
+                * entries in green - and keep the middle ground normal:
+                */
+               if (percent >= 5.0)
+                       color = PERF_COLOR_RED;
+               if (percent < 0.5)
+                       color = PERF_COLOR_GREEN;
+
+               ret = color_fprintf(fp, color, "   %6.2f%%",
+                               (self->count * 100.0) / total_samples);
+       } else
+               ret = fprintf(fp, "%12d ", self->count);
+
+       list_for_each_entry(se, &hist_entry__sort_list, list) {
+               fprintf(fp, "  ");
+               ret += se->print(fp, self);
+       }
+
+       ret += fprintf(fp, "\n");
+
+       return ret;
+}
+
+/*
+ * collect histogram counts
+ */
+
+static int
+hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
+               struct symbol *sym, uint64_t ip, char level)
+{
+       struct rb_node **p = &hist.rb_node;
+       struct rb_node *parent = NULL;
+       struct hist_entry *he;
+       struct hist_entry entry = {
+               .thread = thread,
+               .map    = map,
+               .dso    = dso,
+               .sym    = sym,
+               .ip     = ip,
+               .level  = level,
+               .count  = 1,
+       };
+       int cmp;
+
+       while (*p != NULL) {
+               parent = *p;
+               he = rb_entry(parent, struct hist_entry, rb_node);
+
+               cmp = hist_entry__cmp(&entry, he);
+
+               if (!cmp) {
+                       he->count++;
+                       return 0;
+               }
+
+               if (cmp < 0)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+
+       he = malloc(sizeof(*he));
+       if (!he)
+               return -ENOMEM;
+       *he = entry;
+       rb_link_node(&he->rb_node, parent, p);
+       rb_insert_color(&he->rb_node, &hist);
+
+       return 0;
+}
+
+static void hist_entry__free(struct hist_entry *he)
+{
+       free(he);
+}
+
+/*
+ * collapse the histogram
+ */
+
+static struct rb_root collapse_hists;
+
+static void collapse__insert_entry(struct hist_entry *he)
+{
+       struct rb_node **p = &collapse_hists.rb_node;
+       struct rb_node *parent = NULL;
+       struct hist_entry *iter;
+       int64_t cmp;
+
+       while (*p != NULL) {
+               parent = *p;
+               iter = rb_entry(parent, struct hist_entry, rb_node);
+
+               cmp = hist_entry__collapse(iter, he);
+
+               if (!cmp) {
+                       iter->count += he->count;
+                       hist_entry__free(he);
+                       return;
+               }
+
+               if (cmp < 0)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+
+       rb_link_node(&he->rb_node, parent, p);
+       rb_insert_color(&he->rb_node, &collapse_hists);
+}
+
+static void collapse__resort(void)
+{
+       struct rb_node *next;
+       struct hist_entry *n;
+
+       if (!sort__need_collapse)
+               return;
+
+       next = rb_first(&hist);
+       while (next) {
+               n = rb_entry(next, struct hist_entry, rb_node);
+               next = rb_next(&n->rb_node);
+
+               rb_erase(&n->rb_node, &hist);
+               collapse__insert_entry(n);
+       }
+}
+
+/*
+ * reverse the map, sort on count.
+ */
+
+static struct rb_root output_hists;
+
+static void output__insert_entry(struct hist_entry *he)
+{
+       struct rb_node **p = &output_hists.rb_node;
+       struct rb_node *parent = NULL;
+       struct hist_entry *iter;
+
+       while (*p != NULL) {
+               parent = *p;
+               iter = rb_entry(parent, struct hist_entry, rb_node);
+
+               if (he->count > iter->count)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+
+       rb_link_node(&he->rb_node, parent, p);
+       rb_insert_color(&he->rb_node, &output_hists);
+}
+
+static void output__resort(void)
+{
+       struct rb_node *next;
+       struct hist_entry *n;
+       struct rb_root *tree = &hist;
+
+       if (sort__need_collapse)
+               tree = &collapse_hists;
+
+       next = rb_first(tree);
+
+       while (next) {
+               n = rb_entry(next, struct hist_entry, rb_node);
+               next = rb_next(&n->rb_node);
+
+               rb_erase(&n->rb_node, tree);
+               output__insert_entry(n);
+       }
+}
+
+static size_t output__fprintf(FILE *fp, uint64_t total_samples)
+{
+       struct hist_entry *pos;
+       struct sort_entry *se;
+       struct rb_node *nd;
+       size_t ret = 0;
+
+       fprintf(fp, "\n");
+       fprintf(fp, "#\n");
+       fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
+       fprintf(fp, "#\n");
+
+       fprintf(fp, "# Overhead");
+       list_for_each_entry(se, &hist_entry__sort_list, list)
+               fprintf(fp, "  %s", se->header);
+       fprintf(fp, "\n");
+
+       fprintf(fp, "# ........");
+       list_for_each_entry(se, &hist_entry__sort_list, list) {
+               int i;
+
+               fprintf(fp, "  ");
+               for (i = 0; i < strlen(se->header); i++)
+                       fprintf(fp, ".");
+       }
+       fprintf(fp, "\n");
+
+       fprintf(fp, "#\n");
+
+       for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
+               pos = rb_entry(nd, struct hist_entry, rb_node);
+               ret += hist_entry__fprintf(fp, pos, total_samples);
+       }
+
+       if (!strcmp(sort_order, default_sort_order)) {
+               fprintf(fp, "#\n");
+               fprintf(fp, "# (For more details, try: perf annotate --sort comm,dso,symbol)\n");
+               fprintf(fp, "#\n");
+       }
+       fprintf(fp, "\n");
+
+       return ret;
+}
+
+static void register_idle_thread(void)
+{
+       struct thread *thread = threads__findnew(0);
+
+       if (thread == NULL ||
+                       thread__set_comm(thread, "[idle]")) {
+               fprintf(stderr, "problem inserting idle task.\n");
+               exit(-1);
+       }
+}
+
+static unsigned long total = 0,
+                    total_mmap = 0,
+                    total_comm = 0,
+                    total_fork = 0,
+                    total_unknown = 0;
+
+static int
+process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       char level;
+       int show = 0;
+       struct dso *dso = NULL;
+       struct thread *thread = threads__findnew(event->ip.pid);
+       uint64_t ip = event->ip.ip;
+       struct map *map = NULL;
+
+       dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n",
+               (void *)(offset + head),
+               (void *)(long)(event->header.size),
+               event->header.misc,
+               event->ip.pid,
+               (void *)(long)ip);
+
+       dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
+
+       if (thread == NULL) {
+               fprintf(stderr, "problem processing %d event, skipping it.\n",
+                       event->header.type);
+               return -1;
+       }
+
+       if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
+               show = SHOW_KERNEL;
+               level = 'k';
+
+               dso = kernel_dso;
+
+               dprintf(" ...... dso: %s\n", dso->name);
+
+       } else if (event->header.misc & PERF_EVENT_MISC_USER) {
+
+               show = SHOW_USER;
+               level = '.';
+
+               map = thread__find_map(thread, ip);
+               if (map != NULL) {
+                       ip = map->map_ip(map, ip);
+                       dso = map->dso;
+               } else {
+                       /*
+                        * If this is outside of all known maps,
+                        * and is a negative address, try to look it
+                        * up in the kernel dso, as it might be a
+                        * vsyscall (which executes in user-mode):
+                        */
+                       if ((long long)ip < 0)
+                               dso = kernel_dso;
+               }
+               dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
+
+       } else {
+               show = SHOW_HV;
+               level = 'H';
+               dprintf(" ...... dso: [hypervisor]\n");
+       }
+
+       if (show & show_mask) {
+               struct symbol *sym = NULL;
+
+               if (dso)
+                       sym = dso->find_symbol(dso, ip);
+
+               if (hist_entry__add(thread, map, dso, sym, ip, level)) {
+                       fprintf(stderr,
+               "problem incrementing symbol count, skipping event\n");
+                       return -1;
+               }
+       }
+       total++;
+
+       return 0;
+}
+
+static int
+process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       struct thread *thread = threads__findnew(event->mmap.pid);
+       struct map *map = map__new(&event->mmap);
+
+       dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
+               (void *)(offset + head),
+               (void *)(long)(event->header.size),
+               event->mmap.pid,
+               (void *)(long)event->mmap.start,
+               (void *)(long)event->mmap.len,
+               (void *)(long)event->mmap.pgoff,
+               event->mmap.filename);
+
+       if (thread == NULL || map == NULL) {
+               dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
+               return 0;
+       }
+
+       thread__insert_map(thread, map);
+       total_mmap++;
+
+       return 0;
+}
+
+static int
+process_comm_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       struct thread *thread = threads__findnew(event->comm.pid);
+
+       dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
+               (void *)(offset + head),
+               (void *)(long)(event->header.size),
+               event->comm.comm, event->comm.pid);
+
+       if (thread == NULL ||
+           thread__set_comm(thread, event->comm.comm)) {
+               dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
+               return -1;
+       }
+       total_comm++;
+
+       return 0;
+}
+
+static int
+process_fork_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       struct thread *thread = threads__findnew(event->fork.pid);
+       struct thread *parent = threads__findnew(event->fork.ppid);
+
+       dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
+               (void *)(offset + head),
+               (void *)(long)(event->header.size),
+               event->fork.pid, event->fork.ppid);
+
+       if (!thread || !parent || thread__fork(thread, parent)) {
+               dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
+               return -1;
+       }
+       total_fork++;
+
+       return 0;
+}
+
+static int
+process_period_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
+               (void *)(offset + head),
+               (void *)(long)(event->header.size),
+               event->period.time,
+               event->period.id,
+               event->period.sample_period);
+
+       return 0;
+}
+
+static int
+process_event(event_t *event, unsigned long offset, unsigned long head)
+{
+       if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
+               return process_overflow_event(event, offset, head);
+
+       switch (event->header.type) {
+       case PERF_EVENT_MMAP:
+               return process_mmap_event(event, offset, head);
+
+       case PERF_EVENT_COMM:
+               return process_comm_event(event, offset, head);
+
+       case PERF_EVENT_FORK:
+               return process_fork_event(event, offset, head);
+
+       case PERF_EVENT_PERIOD:
+               return process_period_event(event, offset, head);
+       /*
+        * We dont process them right now but they are fine:
+        */
+
+       case PERF_EVENT_THROTTLE:
+       case PERF_EVENT_UNTHROTTLE:
+               return 0;
+
+       default:
+               return -1;
+       }
+
+       return 0;
+}
+
+static int __cmd_annotate(void)
+{
+       int ret, rc = EXIT_FAILURE;
+       unsigned long offset = 0;
+       unsigned long head = 0;
+       struct stat stat;
+       event_t *event;
+       uint32_t size;
+       char *buf;
+
+       register_idle_thread();
+
+       input = open(input_name, O_RDONLY);
+       if (input < 0) {
+               perror("failed to open file");
+               exit(-1);
+       }
+
+       ret = fstat(input, &stat);
+       if (ret < 0) {
+               perror("failed to stat file");
+               exit(-1);
+       }
+
+       if (!stat.st_size) {
+               fprintf(stderr, "zero-sized file, nothing to do!\n");
+               exit(0);
+       }
+
+       if (load_kernel() < 0) {
+               perror("failed to load kernel symbols");
+               return EXIT_FAILURE;
+       }
+
+       if (!full_paths) {
+               if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
+                       perror("failed to get the current directory");
+                       return EXIT_FAILURE;
+               }
+               cwdlen = strlen(cwd);
+       } else {
+               cwd = NULL;
+               cwdlen = 0;
+       }
+remap:
+       buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
+                          MAP_SHARED, input, offset);
+       if (buf == MAP_FAILED) {
+               perror("failed to mmap file");
+               exit(-1);
+       }
+
+more:
+       event = (event_t *)(buf + head);
+
+       size = event->header.size;
+       if (!size)
+               size = 8;
+
+       if (head + event->header.size >= page_size * mmap_window) {
+               unsigned long shift = page_size * (head / page_size);
+               int ret;
+
+               ret = munmap(buf, page_size * mmap_window);
+               assert(ret == 0);
+
+               offset += shift;
+               head -= shift;
+               goto remap;
+       }
+
+       size = event->header.size;
+
+       dprintf("%p [%p]: event: %d\n",
+                       (void *)(offset + head),
+                       (void *)(long)event->header.size,
+                       event->header.type);
+
+       if (!size || process_event(event, offset, head) < 0) {
+
+               dprintf("%p [%p]: skipping unknown header type: %d\n",
+                       (void *)(offset + head),
+                       (void *)(long)(event->header.size),
+                       event->header.type);
+
+               total_unknown++;
+
+               /*
+                * assume we lost track of the stream, check alignment, and
+                * increment a single u64 in the hope to catch on again 'soon'.
+                */
+
+               if (unlikely(head & 7))
+                       head &= ~7ULL;
+
+               size = 8;
+       }
+
+       head += size;
+
+       if (offset + head < stat.st_size)
+               goto more;
+
+       rc = EXIT_SUCCESS;
+       close(input);
+
+       dprintf("      IP events: %10ld\n", total);
+       dprintf("    mmap events: %10ld\n", total_mmap);
+       dprintf("    comm events: %10ld\n", total_comm);
+       dprintf("    fork events: %10ld\n", total_fork);
+       dprintf(" unknown events: %10ld\n", total_unknown);
+
+       if (dump_trace)
+               return 0;
+
+       if (verbose >= 3)
+               threads__fprintf(stdout);
+
+       if (verbose >= 2)
+               dsos__fprintf(stdout);
+
+       collapse__resort();
+       output__resort();
+       output__fprintf(stdout, total);
+
+       return rc;
+}
+
+static const char * const annotate_usage[] = {
+       "perf annotate [<options>] <command>",
+       NULL
+};
+
+static const struct option options[] = {
+       OPT_STRING('i', "input", &input_name, "file",
+                   "input file name"),
+       OPT_BOOLEAN('v', "verbose", &verbose,
+                   "be more verbose (show symbol address, etc)"),
+       OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
+                   "dump raw trace in ASCII"),
+       OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
+       OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
+                  "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
+       OPT_BOOLEAN('P', "full-paths", &full_paths,
+                   "Don't shorten the pathnames taking into account the cwd"),
+       OPT_END()
+};
+
+static void setup_sorting(void)
+{
+       char *tmp, *tok, *str = strdup(sort_order);
+
+       for (tok = strtok_r(str, ", ", &tmp);
+                       tok; tok = strtok_r(NULL, ", ", &tmp)) {
+               if (sort_dimension__add(tok) < 0) {
+                       error("Unknown --sort key: `%s'", tok);
+                       usage_with_options(annotate_usage, options);
+               }
+       }
+
+       free(str);
+}
+
+int cmd_annotate(int argc, const char **argv, const char *prefix)
+{
+       symbol__init();
+
+       page_size = getpagesize();
+
+       argc = parse_options(argc, argv, options, annotate_usage, 0);
+
+       setup_sorting();
+
+       /*
+        * Any (unrecognized) arguments left?
+        */
+       if (argc)
+               usage_with_options(annotate_usage, options);
+
+       setup_pager();
+
+       return __cmd_annotate();
+}
index e7de47d..51d1682 100644 (file)
@@ -14,6 +14,7 @@ extern void prune_packed_objects(int);
 extern int read_line_with_nul(char *buf, int size, FILE *file);
 extern int check_pager_config(const char *cmd);
 
+extern int cmd_annotate(int argc, const char **argv, const char *prefix);
 extern int cmd_help(int argc, const char **argv, const char *prefix);
 extern int cmd_record(int argc, const char **argv, const char *prefix);
 extern int cmd_report(int argc, const char **argv, const char *prefix);
index f0b922c..eebce30 100644 (file)
@@ -2,8 +2,9 @@
 # List of known perf commands.
 # command name                 category [deprecated] [common]
 #
+perf-annotate                  mainporcelain common
+perf-list                      mainporcelain common
 perf-record                    mainporcelain common
 perf-report                    mainporcelain common
 perf-stat                      mainporcelain common
 perf-top                       mainporcelain common
-perf-list                      mainporcelain common
index 161824f..4eb7259 100644 (file)
@@ -263,6 +263,7 @@ static void handle_internal_command(int argc, const char **argv)
                { "report", cmd_report, 0 },
                { "stat", cmd_stat, 0 },
                { "top", cmd_top, 0 },
+               { "annotate", cmd_annotate, 0 },
                { "version", cmd_version, 0 },
        };
        int i;
@@ -402,9 +403,11 @@ int main(int argc, const char **argv)
        while (1) {
                static int done_help = 0;
                static int was_alias = 0;
+
                was_alias = run_argv(&argc, &argv);
                if (errno != ENOENT)
                        break;
+
                if (was_alias) {
                        fprintf(stderr, "Expansion of alias '%s' failed; "
                                "'%s' is not a perf-command\n",