perf: Generalize perf lock's sample event reordering to the session layer

The sample events recorded by perf record are not time ordered
because we have one buffer per cpu for each event (even demultiplexed
per task/per cpu for task bound events). But when we read trace events
we want them to be ordered by time because many state machines are
involved.

There are currently two ways perf tools deal with that:

- use -M to multiplex every buffers (perf sched, perf kmem)
  But this creates a lot of contention in SMP machines on
  record time.

- use a post-processing time reordering (perf timechart, perf lock)
  The reordering used by timechart is simple but doesn't scale well
  with huge flow of events, in terms of performance and memory use
  (unusable with perf lock for example).
  Perf lock has its own samples reordering that flushes its memory
  use in a regular basis and that uses a sorting based on the
  previous event queued (a new event to be queued is close to the
  previous one most of the time).

This patch proposes to export perf lock's samples reordering facility
to the session layer that reads the events. So if a tool wants to
get ordered sample events, it needs to set its
struct perf_event_ops::ordered_samples to true and that's it.

This prepares tracing based perf tools to get rid of the need to
use buffers multiplexing (-M) or to implement their own
reordering.

Also lower the flush period to 2 as it's sufficient already.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Tom Zanussi <tzanussi@gmail.com>

1 files changed, 22 insertions, 175 deletions

diff --git a/tools/perf/builtin-lock.c b/tools/perf/builtin-lock.c
index 716d8c544a56..ce276750b140 100644
--- a/tools/perf/builtin-lock.c
+++ b/tools/perf/builtin-lock.c
@@ -316,8 +316,6 @@ alloc_failed:
 
 static char                     const *input_name = "perf.data";
 
-static int                      profile_cpu = -1;
-
 struct raw_event_sample {
         u32                     size;
         char                    data[0];
@@ -697,8 +695,7 @@ process_lock_release_event(void *data,
 }
 
 static void
-process_raw_event(void *data, int cpu __used,
-                  u64 timestamp __used, struct thread *thread __used)
+process_raw_event(void *data, int cpu, u64 timestamp, struct thread *thread)
 {
         struct event *event;
         int type;
@@ -716,176 +713,6 @@ process_raw_event(void *data, int cpu __used,
                 process_lock_release_event(data, event, cpu, timestamp, thread);
 }
 
-struct raw_event_queue {
-        u64                     timestamp;
-        int                     cpu;
-        void                    *data;
-        struct thread           *thread;
-        struct list_head        list;
-};
-
-static LIST_HEAD(raw_event_head);
-
-#define FLUSH_PERIOD    (5 * NSEC_PER_SEC)
-
-static u64 flush_limit = ULLONG_MAX;
-static u64 last_flush = 0;
-struct raw_event_queue *last_inserted;
-
-static void flush_raw_event_queue(u64 limit)
-{
-        struct raw_event_queue *tmp, *iter;
-
-        list_for_each_entry_safe(iter, tmp, &raw_event_head, list) {
-                if (iter->timestamp > limit)
-                        return;
-
-                if (iter == last_inserted)
-                        last_inserted = NULL;
-
-                process_raw_event(iter->data, iter->cpu, iter->timestamp,
-                                  iter->thread);
-
-                last_flush = iter->timestamp;
-                list_del(&iter->list);
-                free(iter->data);
-                free(iter);
-        }
-}
-
-static void __queue_raw_event_end(struct raw_event_queue *new)
-{
-        struct raw_event_queue *iter;
-
-        list_for_each_entry_reverse(iter, &raw_event_head, list) {
-                if (iter->timestamp < new->timestamp) {
-                        list_add(&new->list, &iter->list);
-                        return;
-                }
-        }
-
-        list_add(&new->list, &raw_event_head);
-}
-
-static void __queue_raw_event_before(struct raw_event_queue *new,
-                                     struct raw_event_queue *iter)
-{
-        list_for_each_entry_continue_reverse(iter, &raw_event_head, list) {
-                if (iter->timestamp < new->timestamp) {
-                        list_add(&new->list, &iter->list);
-                        return;
-                }
-        }
-
-        list_add(&new->list, &raw_event_head);
-}
-
-static void __queue_raw_event_after(struct raw_event_queue *new,
-                                     struct raw_event_queue *iter)
-{
-        list_for_each_entry_continue(iter, &raw_event_head, list) {
-                if (iter->timestamp > new->timestamp) {
-                        list_add_tail(&new->list, &iter->list);
-                        return;
-                }
-        }
-        list_add_tail(&new->list, &raw_event_head);
-}
-
-/* The queue is ordered by time */
-static void __queue_raw_event(struct raw_event_queue *new)
-{
-        if (!last_inserted) {
-                __queue_raw_event_end(new);
-                return;
-        }
-
-        /*
-         * Most of the time the current event has a timestamp
-         * very close to the last event inserted, unless we just switched
-         * to another event buffer. Having a sorting based on a list and
-         * on the last inserted event that is close to the current one is
-         * probably more efficient than an rbtree based sorting.
-         */
-        if (last_inserted->timestamp >= new->timestamp)
-                __queue_raw_event_before(new, last_inserted);
-        else
-                __queue_raw_event_after(new, last_inserted);
-}
-
-static void queue_raw_event(void *data, int raw_size, int cpu,
-                            u64 timestamp, struct thread *thread)
-{
-        struct raw_event_queue *new;
-
-        if (flush_limit == ULLONG_MAX)
-                flush_limit = timestamp + FLUSH_PERIOD;
-
-        if (timestamp < last_flush) {
-                printf("Warning: Timestamp below last timeslice flush\n");
-                return;
-        }
-
-        new = malloc(sizeof(*new));
-        if (!new)
-                die("Not enough memory\n");
-
-        new->timestamp = timestamp;
-        new->cpu = cpu;
-        new->thread = thread;
-
-        new->data = malloc(raw_size);
-        if (!new->data)
-                die("Not enough memory\n");
-
-        memcpy(new->data, data, raw_size);
-
-        __queue_raw_event(new);
-        last_inserted = new;
-
-        /*
-         * We want to have a slice of events covering 2 * FLUSH_PERIOD
-         * If FLUSH_PERIOD is big enough, it ensures every events that occured
-         * in the first half of the timeslice have all been buffered and there
-         * are none remaining (we need that because of the weakly ordered
-         * event recording we have). Then once we reach the 2 * FLUSH_PERIOD
-         * timeslice, we flush the first half to be gentle with the memory
-         * (the second half can still get new events in the middle, so wait
-         * another period to flush it)
-         */
-        if (new->timestamp > flush_limit &&
-                new->timestamp - flush_limit > FLUSH_PERIOD) {
-                flush_limit += FLUSH_PERIOD;
-                flush_raw_event_queue(flush_limit);
-        }
-}
-
-static int process_sample_event(event_t *event, struct perf_session *s)
-{
-        struct thread *thread;
-        struct sample_data data;
-
-        bzero(&data, sizeof(struct sample_data));
-        event__parse_sample(event, s->sample_type, &data);
-        /* CAUTION: using tid as thread.pid */
-        thread = perf_session__findnew(s, data.tid);
-
-        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;
-
-        queue_raw_event(data.raw_data, data.raw_size, data.cpu, data.time, thread);
-
-        return 0;
-}
-
 /* TODO: various way to print, coloring, nano or milli sec */
 static void print_result(void)
 {
@@ -963,9 +790,30 @@ static void dump_map(void)
         }
 }
 
+static int process_sample_event(event_t *self, struct perf_session *s)
+{
+        struct sample_data data;
+        struct thread *thread;
+
+        bzero(&data, sizeof(data));
+        event__parse_sample(self, s->sample_type, &data);
+
+        thread = perf_session__findnew(s, data.tid);
+        if (thread == NULL) {
+                pr_debug("problem processing %d event, skipping it.\n",
+                        self->header.type);
+                return -1;
+        }
+
+        process_raw_event(data.raw_data, data.cpu, data.time, thread);
+
+        return 0;
+}
+
 static struct perf_event_ops eops = {
         .sample                 = process_sample_event,
         .comm                   = event__process_comm,
+        .ordered_samples        = true,
 };
 
 static int read_events(void)
@@ -994,7 +842,6 @@ static void __cmd_report(void)
         setup_pager();
         select_key();
         read_events();
-        flush_raw_event_queue(ULLONG_MAX);
         sort_result();
         print_result();
 }