perf lock: Introduce new tool "perf lock", for analyzing lock statistics
Hitoshi Mitake [Sat, 30 Jan 2010 11:43:33 +0000 (20:43 +0900)]
Adding new subcommand "perf lock" to perf.

I have a lot of remaining ToDos, but for now perf lock can
already provide minimal functionality for analyzing lock
statistics.

Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1264851813-8413-12-git-send-email-mitake@dcl.info.waseda.ac.jp>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tools/perf/Makefile
tools/perf/builtin-lock.c [new file with mode: 0644]
tools/perf/builtin.h
tools/perf/perf.c

index b2bce1f..4296930 100644 (file)
@@ -456,6 +456,7 @@ BUILTIN_OBJS += builtin-top.o
 BUILTIN_OBJS += builtin-trace.o
 BUILTIN_OBJS += builtin-probe.o
 BUILTIN_OBJS += builtin-kmem.o
+BUILTIN_OBJS += builtin-lock.o
 
 PERFLIBS = $(LIB_FILE)
 
diff --git a/tools/perf/builtin-lock.c b/tools/perf/builtin-lock.c
new file mode 100644 (file)
index 0000000..2b5f887
--- /dev/null
@@ -0,0 +1,724 @@
+#include "builtin.h"
+#include "perf.h"
+
+#include "util/util.h"
+#include "util/cache.h"
+#include "util/symbol.h"
+#include "util/thread.h"
+#include "util/header.h"
+
+#include "util/parse-options.h"
+#include "util/trace-event.h"
+
+#include "util/debug.h"
+#include "util/session.h"
+
+#include <sys/types.h>
+#include <sys/prctl.h>
+#include <semaphore.h>
+#include <pthread.h>
+#include <math.h>
+#include <limits.h>
+
+#include <linux/list.h>
+#include <linux/hash.h>
+
+/* based on kernel/lockdep.c */
+#define LOCKHASH_BITS          12
+#define LOCKHASH_SIZE          (1UL << LOCKHASH_BITS)
+
+static struct list_head lockhash_table[LOCKHASH_SIZE];
+
+#define __lockhashfn(key)      hash_long((unsigned long)key, LOCKHASH_BITS)
+#define lockhashentry(key)     (lockhash_table + __lockhashfn((key)))
+
+#define LOCK_STATE_UNLOCKED 0         /* initial state */
+#define LOCK_STATE_LOCKED 1
+
+struct lock_stat {
+       struct list_head hash_entry;
+       struct rb_node rb;      /* used for sorting */
+
+       /* FIXME: raw_field_value() returns unsigned long long,
+        * so address of lockdep_map should be dealed as 64bit.
+        * Is there more better solution? */
+       void *addr;            /* address of lockdep_map, used as ID */
+       char *name;            /* for strcpy(), we cannot use const */
+       char *file;
+       unsigned int line;
+
+       int state;
+       u64 prev_event_time;    /* timestamp of previous event */
+
+       unsigned int nr_acquired;
+       unsigned int nr_acquire;
+       unsigned int nr_contended;
+       unsigned int nr_release;
+
+       /* these times are in nano sec. */
+       u64 wait_time_total;
+       u64 wait_time_min;
+       u64 wait_time_max;
+};
+
+/* build simple key function one is bigger than two */
+#define SINGLE_KEY(member)                                     \
+       static int lock_stat_key_ ## member(struct lock_stat *one,      \
+                                        struct lock_stat *two)         \
+       {                                                               \
+               return one->member > two->member;                       \
+       }
+
+SINGLE_KEY(nr_acquired)
+SINGLE_KEY(nr_contended)
+SINGLE_KEY(wait_time_total)
+SINGLE_KEY(wait_time_min)
+SINGLE_KEY(wait_time_max)
+
+struct lock_key {
+       /*
+        * name: the value for specify by user
+        * this should be simpler than raw name of member
+        * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
+        */
+       const char *name;
+       int (*key)(struct lock_stat*, struct lock_stat*);
+};
+
+static const char *sort_key = "acquired";
+static int (*compare)(struct lock_stat *, struct lock_stat *);
+
+#define DEF_KEY_LOCK(name, fn_suffix)  \
+       { #name, lock_stat_key_ ## fn_suffix }
+struct lock_key keys[] = {
+       DEF_KEY_LOCK(acquired, nr_acquired),
+       DEF_KEY_LOCK(contended, nr_contended),
+       DEF_KEY_LOCK(wait_total, wait_time_total),
+       DEF_KEY_LOCK(wait_min, wait_time_min),
+       DEF_KEY_LOCK(wait_max, wait_time_max),
+
+       /* extra comparisons much complicated should be here */
+
+       { NULL, NULL }
+};
+
+static void select_key(void)
+{
+       int i;
+
+       for (i = 0; keys[i].name; i++) {
+               if (!strcmp(keys[i].name, sort_key)) {
+                       compare = keys[i].key;
+                       return;
+               }
+       }
+
+       die("Unknown compare key:%s\n", sort_key);
+}
+
+static struct rb_root result;  /* place to store sorted data */
+
+static void insert_to_result(struct lock_stat *st,
+                            int (*bigger)(struct lock_stat *,
+                                          struct lock_stat *))
+{
+       struct rb_node **rb = &result.rb_node;
+       struct rb_node *parent = NULL;
+       struct lock_stat *p;
+
+       while (*rb) {
+               p = container_of(*rb, struct lock_stat, rb);
+               parent = *rb;
+
+               if (bigger(st, p))
+                       rb = &(*rb)->rb_left;
+               else
+                       rb = &(*rb)->rb_right;
+       }
+
+       rb_link_node(&st->rb, parent, rb);
+       rb_insert_color(&st->rb, &result);
+}
+
+/* returns left most element of result, and erase it */
+static struct lock_stat *pop_from_result(void)
+{
+       struct rb_node *node = result.rb_node;
+
+       if (!node)
+               return NULL;
+
+       while (node->rb_left)
+               node = node->rb_left;
+
+       rb_erase(node, &result);
+       return container_of(node, struct lock_stat, rb);
+}
+
+static struct lock_stat *lock_stat_findnew(void *addr, const char *name,
+                                          const char *file, unsigned int line)
+{
+       struct list_head *entry = lockhashentry(addr);
+       struct lock_stat *ret, *new;
+
+       list_for_each_entry(ret, entry, hash_entry) {
+               if (ret->addr == addr)
+                       return ret;
+       }
+
+       new = zalloc(sizeof(struct lock_stat));
+       if (!new)
+               goto alloc_failed;
+
+       new->addr = addr;
+       new->name = zalloc(sizeof(char) * strlen(name) + 1);
+       if (!new->name)
+               goto alloc_failed;
+       strcpy(new->name, name);
+       new->file = zalloc(sizeof(char) * strlen(file) + 1);
+       if (!new->file)
+               goto alloc_failed;
+       strcpy(new->file, file);
+       new->line = line;
+
+       /* LOCK_STATE_UNLOCKED == 0 isn't guaranteed forever */
+       new->state = LOCK_STATE_UNLOCKED;
+       new->wait_time_min = ULLONG_MAX;
+
+       list_add(&new->hash_entry, entry);
+       return new;
+
+alloc_failed:
+       die("memory allocation failed\n");
+}
+
+static char                    const *input_name = "perf.data";
+
+static int                     profile_cpu = -1;
+
+struct raw_event_sample {
+       u32 size;
+       char data[0];
+};
+
+struct trace_acquire_event {
+       void *addr;
+       const char *name;
+       const char *file;
+       unsigned int line;
+};
+
+struct trace_acquired_event {
+       void *addr;
+       const char *name;
+       const char *file;
+       unsigned int line;
+};
+
+struct trace_contended_event {
+       void *addr;
+       const char *name;
+       const char *file;
+       unsigned int line;
+};
+
+struct trace_release_event {
+       void *addr;
+       const char *name;
+       const char *file;
+       unsigned int line;
+};
+
+struct trace_lock_handler {
+       void (*acquire_event)(struct trace_acquire_event *,
+                             struct event *,
+                             int cpu,
+                             u64 timestamp,
+                             struct thread *thread);
+
+       void (*acquired_event)(struct trace_acquired_event *,
+                              struct event *,
+                              int cpu,
+                              u64 timestamp,
+                              struct thread *thread);
+
+       void (*contended_event)(struct trace_contended_event *,
+                               struct event *,
+                               int cpu,
+                               u64 timestamp,
+                               struct thread *thread);
+
+       void (*release_event)(struct trace_release_event *,
+                             struct event *,
+                             int cpu,
+                             u64 timestamp,
+                             struct thread *thread);
+};
+
+static void prof_lock_acquire_event(struct trace_acquire_event *acquire_event,
+                       struct event *__event __used,
+                       int cpu __used,
+                       u64 timestamp,
+                       struct thread *thread __used)
+{
+       struct lock_stat *st;
+
+       st = lock_stat_findnew(acquire_event->addr, acquire_event->name,
+                              acquire_event->file, acquire_event->line);
+
+       switch (st->state) {
+       case LOCK_STATE_UNLOCKED:
+               break;
+       case LOCK_STATE_LOCKED:
+               break;
+       default:
+               BUG_ON(1);
+               break;
+       }
+
+       st->prev_event_time = timestamp;
+}
+
+static void prof_lock_acquired_event(struct trace_acquired_event *acquired_event,
+                        struct event *__event __used,
+                        int cpu __used,
+                        u64 timestamp,
+                        struct thread *thread __used)
+{
+       struct lock_stat *st;
+
+       st = lock_stat_findnew(acquired_event->addr, acquired_event->name,
+                              acquired_event->file, acquired_event->line);
+
+       switch (st->state) {
+       case LOCK_STATE_UNLOCKED:
+               st->state = LOCK_STATE_LOCKED;
+               st->nr_acquired++;
+               break;
+       case LOCK_STATE_LOCKED:
+               break;
+       default:
+               BUG_ON(1);
+               break;
+       }
+
+       st->prev_event_time = timestamp;
+}
+
+static void prof_lock_contended_event(struct trace_contended_event *contended_event,
+                         struct event *__event __used,
+                         int cpu __used,
+                         u64 timestamp,
+                         struct thread *thread __used)
+{
+       struct lock_stat *st;
+
+       st = lock_stat_findnew(contended_event->addr, contended_event->name,
+                              contended_event->file, contended_event->line);
+
+       switch (st->state) {
+       case LOCK_STATE_UNLOCKED:
+               break;
+       case LOCK_STATE_LOCKED:
+               st->nr_contended++;
+               break;
+       default:
+               BUG_ON(1);
+               break;
+       }
+
+       st->prev_event_time = timestamp;
+}
+
+static void prof_lock_release_event(struct trace_release_event *release_event,
+                       struct event *__event __used,
+                       int cpu __used,
+                       u64 timestamp,
+                       struct thread *thread __used)
+{
+       struct lock_stat *st;
+       u64 hold_time;
+
+       st = lock_stat_findnew(release_event->addr, release_event->name,
+                              release_event->file, release_event->line);
+
+       switch (st->state) {
+       case LOCK_STATE_UNLOCKED:
+               break;
+       case LOCK_STATE_LOCKED:
+               st->state = LOCK_STATE_UNLOCKED;
+               hold_time = timestamp - st->prev_event_time;
+
+               if (timestamp < st->prev_event_time) {
+                       /* terribly, this can happen... */
+                       goto end;
+               }
+
+               if (st->wait_time_min > hold_time)
+                       st->wait_time_min = hold_time;
+               if (st->wait_time_max < hold_time)
+                       st->wait_time_max = hold_time;
+               st->wait_time_total += hold_time;
+
+               st->nr_release++;
+               break;
+       default:
+               BUG_ON(1);
+               break;
+       }
+
+end:
+       st->prev_event_time = timestamp;
+}
+
+/* lock oriented handlers */
+/* TODO: handlers for CPU oriented, thread oriented */
+static struct trace_lock_handler prof_lock_ops  = {
+       .acquire_event          = prof_lock_acquire_event,
+       .acquired_event         = prof_lock_acquired_event,
+       .contended_event        = prof_lock_contended_event,
+       .release_event          = prof_lock_release_event,
+};
+
+static struct trace_lock_handler *trace_handler;
+
+static void
+process_lock_acquire_event(void *data,
+                          struct event *event __used,
+                          int cpu __used,
+                          u64 timestamp __used,
+                          struct thread *thread __used)
+{
+       struct trace_acquire_event acquire_event;
+       u64 tmp;                /* this is required for casting... */
+
+       tmp = raw_field_value(event, "lockdep_addr", data);
+       memcpy(&acquire_event.addr, &tmp, sizeof(void *));
+       acquire_event.name = (char *)raw_field_ptr(event, "name", data);
+       acquire_event.file = (char *)raw_field_ptr(event, "file", data);
+       acquire_event.line =
+               (unsigned int)raw_field_value(event, "line", data);
+
+       if (trace_handler->acquire_event) {
+               trace_handler->acquire_event(&acquire_event,
+                                            event, cpu, timestamp, thread);
+       }
+}
+
+static void
+process_lock_acquired_event(void *data,
+                           struct event *event __used,
+                           int cpu __used,
+                           u64 timestamp __used,
+                           struct thread *thread __used)
+{
+       struct trace_acquired_event acquired_event;
+       u64 tmp;                /* this is required for casting... */
+
+       tmp = raw_field_value(event, "lockdep_addr", data);
+       memcpy(&acquired_event.addr, &tmp, sizeof(void *));
+       acquired_event.name = (char *)raw_field_ptr(event, "name", data);
+       acquired_event.file = (char *)raw_field_ptr(event, "file", data);
+       acquired_event.line =
+               (unsigned int)raw_field_value(event, "line", data);
+
+       if (trace_handler->acquire_event) {
+               trace_handler->acquired_event(&acquired_event,
+                                            event, cpu, timestamp, thread);
+       }
+}
+
+static void
+process_lock_contended_event(void *data,
+                            struct event *event __used,
+                            int cpu __used,
+                            u64 timestamp __used,
+                            struct thread *thread __used)
+{
+       struct trace_contended_event contended_event;
+       u64 tmp;                /* this is required for casting... */
+
+       tmp = raw_field_value(event, "lockdep_addr", data);
+       memcpy(&contended_event.addr, &tmp, sizeof(void *));
+       contended_event.name = (char *)raw_field_ptr(event, "name", data);
+       contended_event.file = (char *)raw_field_ptr(event, "file", data);
+       contended_event.line =
+               (unsigned int)raw_field_value(event, "line", data);
+
+       if (trace_handler->acquire_event) {
+               trace_handler->contended_event(&contended_event,
+                                            event, cpu, timestamp, thread);
+       }
+}
+
+static void
+process_lock_release_event(void *data,
+                          struct event *event __used,
+                          int cpu __used,
+                          u64 timestamp __used,
+                          struct thread *thread __used)
+{
+       struct trace_release_event release_event;
+       u64 tmp;                /* this is required for casting... */
+
+       tmp = raw_field_value(event, "lockdep_addr", data);
+       memcpy(&release_event.addr, &tmp, sizeof(void *));
+       release_event.name = (char *)raw_field_ptr(event, "name", data);
+       release_event.file = (char *)raw_field_ptr(event, "file", data);
+       release_event.line =
+               (unsigned int)raw_field_value(event, "line", data);
+
+       if (trace_handler->acquire_event) {
+               trace_handler->release_event(&release_event,
+                                            event, cpu, timestamp, thread);
+       }
+}
+
+static void
+process_raw_event(void *data, int cpu,
+                 u64 timestamp, struct thread *thread)
+{
+       struct event *event;
+       int type;
+
+       type = trace_parse_common_type(data);
+       event = trace_find_event(type);
+
+       if (!strcmp(event->name, "lock_acquire"))
+               process_lock_acquire_event(data, event, cpu, timestamp, thread);
+       if (!strcmp(event->name, "lock_acquired"))
+               process_lock_acquired_event(data, event, cpu, timestamp, thread);
+       if (!strcmp(event->name, "lock_contended"))
+               process_lock_contended_event(data, event, cpu, timestamp, thread);
+       if (!strcmp(event->name, "lock_release"))
+               process_lock_release_event(data, event, cpu, timestamp, thread);
+}
+
+static int process_sample_event(event_t *event, struct perf_session *session)
+{
+       struct thread *thread;
+       struct sample_data data;
+
+       bzero(&data, sizeof(struct sample_data));
+       event__parse_sample(event, session->sample_type, &data);
+       thread = perf_session__findnew(session, data.pid);
+
+       /*
+        * FIXME: this causes warn on 32bit environment
+        * because of (void *)data.ip (type of data.ip is u64)
+        */
+/*     dump_printf("(IP, %d): %d/%d: %p period: %llu\n", */
+/*                 event->header.misc, */
+/*                 data.pid, data.tid, (void *)data.ip, data.period); */
+
+       if (thread == NULL) {
+               pr_debug("problem processing %d event, skipping it.\n",
+                        event->header.type);
+               return -1;
+       }
+
+       dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
+
+       if (profile_cpu != -1 && profile_cpu != (int) data.cpu)
+               return 0;
+
+       process_raw_event(data.raw_data, data.cpu, data.time, thread);
+
+       return 0;
+}
+
+/* TODO: various way to print, coloring, nano or milli sec */
+static void print_result(void)
+{
+       struct lock_stat *st;
+       char cut_name[20];
+
+       printf("%18s ", "ID");
+       printf("%20s ", "Name");
+       printf("%10s ", "acquired");
+       printf("%10s ", "contended");
+
+       printf("%15s ", "total wait (ns)");
+       printf("%15s ", "max wait (ns)");
+       printf("%15s ", "min wait (ns)");
+
+       printf("\n\n");
+
+       while ((st = pop_from_result())) {
+               bzero(cut_name, 20);
+
+               printf("%p ", st->addr);
+
+               if (strlen(st->name) < 16) {
+                       /* output raw name */
+                       printf("%20s ", st->name);
+               } else {
+                       strncpy(cut_name, st->name, 16);
+                       cut_name[16] = '.';
+                       cut_name[17] = '.';
+                       cut_name[18] = '.';
+                       cut_name[19] = '\0';
+                       /* cut off name for saving output style */
+                       printf("%20s ", cut_name);
+               }
+
+               printf("%10u ", st->nr_acquired);
+               printf("%10u ", st->nr_contended);
+
+               printf("%15llu ", st->wait_time_total);
+               printf("%15llu ", st->wait_time_max);
+               printf("%15llu ", st->wait_time_min == ULLONG_MAX ?
+                      0 : st->wait_time_min);
+               printf("\n");
+       }
+}
+
+static void dump_map(void)
+{
+       unsigned int i;
+       struct lock_stat *st;
+
+       for (i = 0; i < LOCKHASH_SIZE; i++) {
+               list_for_each_entry(st, &lockhash_table[i], hash_entry) {
+                       printf("%p: %s (src: %s, line: %u)\n",
+                              st->addr, st->name, st->file, st->line);
+               }
+       }
+}
+
+static struct perf_event_ops eops = {
+       .sample = process_sample_event,
+       .comm   = event__process_comm,
+};
+
+static struct perf_session *session;
+
+static int read_events(void)
+{
+       session = perf_session__new(input_name, O_RDONLY, 0);
+       if (!session)
+               die("Initializing perf session failed\n");
+
+       return perf_session__process_events(session, &eops);
+}
+
+static void sort_result(void)
+{
+       unsigned int i;
+       struct lock_stat *st;
+
+       for (i = 0; i < LOCKHASH_SIZE; i++) {
+               list_for_each_entry(st, &lockhash_table[i], hash_entry) {
+                       insert_to_result(st, compare);
+               }
+       }
+}
+
+static void __cmd_prof(void)
+{
+       setup_pager();
+       select_key();
+       read_events();
+       sort_result();
+       print_result();
+}
+
+static const char * const prof_usage[] = {
+       "perf sched prof [<options>]",
+       NULL
+};
+
+static const struct option prof_options[] = {
+       OPT_STRING('k', "key", &sort_key, "acquired",
+                   "key for sorting"),
+       /* TODO: type */
+       OPT_END()
+};
+
+static const char * const lock_usage[] = {
+       "perf lock [<options>] {record|trace|prof}",
+       NULL
+};
+
+static const struct option lock_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_END()
+};
+
+static const char *record_args[] = {
+       "record",
+       "-a",
+       "-R",
+       "-M",
+       "-f",
+       "-m", "1024",
+       "-c", "1",
+       "-e", "lock:lock_acquire:r",
+       "-e", "lock:lock_acquired:r",
+       "-e", "lock:lock_contended:r",
+       "-e", "lock:lock_release:r",
+};
+
+static int __cmd_record(int argc, const char **argv)
+{
+       unsigned int rec_argc, i, j;
+       const char **rec_argv;
+
+       rec_argc = ARRAY_SIZE(record_args) + argc - 1;
+       rec_argv = calloc(rec_argc + 1, sizeof(char *));
+
+       for (i = 0; i < ARRAY_SIZE(record_args); i++)
+               rec_argv[i] = strdup(record_args[i]);
+
+       for (j = 1; j < (unsigned int)argc; j++, i++)
+               rec_argv[i] = argv[j];
+
+       BUG_ON(i != rec_argc);
+
+       return cmd_record(i, rec_argv, NULL);
+}
+
+int cmd_lock(int argc, const char **argv, const char *prefix __used)
+{
+       unsigned int i;
+
+       symbol__init();
+       for (i = 0; i < LOCKHASH_SIZE; i++)
+               INIT_LIST_HEAD(lockhash_table + i);
+
+       argc = parse_options(argc, argv, lock_options, lock_usage,
+                            PARSE_OPT_STOP_AT_NON_OPTION);
+       if (!argc)
+               usage_with_options(lock_usage, lock_options);
+
+       if (!strncmp(argv[0], "rec", 3)) {
+               return __cmd_record(argc, argv);
+       } else if (!strncmp(argv[0], "prof", 4)) {
+               trace_handler = &prof_lock_ops;
+               if (argc) {
+                       argc = parse_options(argc, argv,
+                                            prof_options, prof_usage, 0);
+                       if (argc)
+                               usage_with_options(prof_usage, prof_options);
+               }
+               __cmd_prof();
+       } else if (!strcmp(argv[0], "trace")) {
+               /* Aliased to 'perf trace' */
+               return cmd_trace(argc, argv, prefix);
+       } else if (!strcmp(argv[0], "map")) {
+               /* recycling prof_lock_ops */
+               trace_handler = &prof_lock_ops;
+               setup_pager();
+               read_events();
+               dump_map();
+       } else {
+               usage_with_options(lock_usage, lock_options);
+       }
+
+       return 0;
+}
index dee97cf..10fe49e 100644 (file)
@@ -31,5 +31,6 @@ extern int cmd_trace(int argc, const char **argv, const char *prefix);
 extern int cmd_version(int argc, const char **argv, const char *prefix);
 extern int cmd_probe(int argc, const char **argv, const char *prefix);
 extern int cmd_kmem(int argc, const char **argv, const char *prefix);
+extern int cmd_lock(int argc, const char **argv, const char *prefix);
 
 #endif
index 109b89b..57cb107 100644 (file)
@@ -303,6 +303,7 @@ static void handle_internal_command(int argc, const char **argv)
                { "sched",      cmd_sched,      0 },
                { "probe",      cmd_probe,      0 },
                { "kmem",       cmd_kmem,       0 },
+               { "lock",       cmd_lock,       0 },
        };
        unsigned int i;
        static const char ext[] = STRIP_EXTENSION;