Commit a1225dec authored by Thomas Gleixner's avatar Thomas Gleixner Committed by Arnaldo Carvalho de Melo

perf session: Fix list sort algorithm

The homebrewn sort algorithm fails to sort in time order. One of the problem
spots is that it fails to deal with equal timestamps correctly.

My first gut reaction was to replace the fancy list with an rbtree, but the
performance is 3 times worse.

Rewrite it so it works.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20101130163819.908482530@linutronix.de>
Signed-off-by: default avatarThomas Gleixner <tglx@linutronix.de>
Signed-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
parent c320c7b7
...@@ -104,7 +104,7 @@ struct perf_session *perf_session__new(const char *filename, int mode, bool forc ...@@ -104,7 +104,7 @@ struct perf_session *perf_session__new(const char *filename, int mode, bool forc
self->mmap_window = 32; self->mmap_window = 32;
self->machines = RB_ROOT; self->machines = RB_ROOT;
self->repipe = repipe; self->repipe = repipe;
INIT_LIST_HEAD(&self->ordered_samples.samples_head); INIT_LIST_HEAD(&self->ordered_samples.samples);
machine__init(&self->host_machine, "", HOST_KERNEL_ID); machine__init(&self->host_machine, "", HOST_KERNEL_ID);
if (mode == O_RDONLY) { if (mode == O_RDONLY) {
...@@ -393,27 +393,33 @@ struct sample_queue { ...@@ -393,27 +393,33 @@ struct sample_queue {
static void flush_sample_queue(struct perf_session *s, static void flush_sample_queue(struct perf_session *s,
struct perf_event_ops *ops) struct perf_event_ops *ops)
{ {
struct list_head *head = &s->ordered_samples.samples_head; struct ordered_samples *os = &s->ordered_samples;
u64 limit = s->ordered_samples.next_flush; struct list_head *head = &os->samples;
struct sample_queue *tmp, *iter; struct sample_queue *tmp, *iter;
u64 limit = os->next_flush;
u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
if (!ops->ordered_samples || !limit) if (!ops->ordered_samples || !limit)
return; return;
list_for_each_entry_safe(iter, tmp, head, list) { list_for_each_entry_safe(iter, tmp, head, list) {
if (iter->timestamp > limit) if (iter->timestamp > limit)
return; break;
if (iter == s->ordered_samples.last_inserted)
s->ordered_samples.last_inserted = NULL;
ops->sample((event_t *)iter->event, s); ops->sample((event_t *)iter->event, s);
s->ordered_samples.last_flush = iter->timestamp; os->last_flush = iter->timestamp;
list_del(&iter->list); list_del(&iter->list);
free(iter->event); free(iter->event);
free(iter); free(iter);
} }
if (list_empty(head)) {
os->last_sample = NULL;
} else if (last_ts <= limit) {
os->last_sample =
list_entry(head->prev, struct sample_queue, list);
}
} }
/* /*
...@@ -465,71 +471,50 @@ static int process_finished_round(event_t *event __used, ...@@ -465,71 +471,50 @@ static int process_finished_round(event_t *event __used,
return 0; return 0;
} }
static void __queue_sample_end(struct sample_queue *new, struct list_head *head)
{
struct sample_queue *iter;
list_for_each_entry_reverse(iter, head, list) {
if (iter->timestamp < new->timestamp) {
list_add(&new->list, &iter->list);
return;
}
}
list_add(&new->list, head);
}
static void __queue_sample_before(struct sample_queue *new,
struct sample_queue *iter,
struct list_head *head)
{
list_for_each_entry_continue_reverse(iter, head, list) {
if (iter->timestamp < new->timestamp) {
list_add(&new->list, &iter->list);
return;
}
}
list_add(&new->list, head);
}
static void __queue_sample_after(struct sample_queue *new,
struct sample_queue *iter,
struct list_head *head)
{
list_for_each_entry_continue(iter, head, list) {
if (iter->timestamp > new->timestamp) {
list_add_tail(&new->list, &iter->list);
return;
}
}
list_add_tail(&new->list, head);
}
/* The queue is ordered by time */ /* The queue is ordered by time */
static void __queue_sample_event(struct sample_queue *new, static void __queue_sample_event(struct sample_queue *new,
struct perf_session *s) struct perf_session *s)
{ {
struct sample_queue *last_inserted = s->ordered_samples.last_inserted; struct ordered_samples *os = &s->ordered_samples;
struct list_head *head = &s->ordered_samples.samples_head; struct sample_queue *sample = os->last_sample;
u64 timestamp = new->timestamp;
struct list_head *p;
os->last_sample = new;
if (!last_inserted) { if (!sample) {
__queue_sample_end(new, head); list_add(&new->list, &os->samples);
os->max_timestamp = timestamp;
return; return;
} }
/* /*
* Most of the time the current event has a timestamp * last_sample might point to some random place in the list as it's
* very close to the last event inserted, unless we just switched * the last queued event. We expect that the new event is close to
* to another event buffer. Having a sorting based on a list and * this.
* 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) if (sample->timestamp <= timestamp) {
__queue_sample_before(new, last_inserted, head); while (sample->timestamp <= timestamp) {
else p = sample->list.next;
__queue_sample_after(new, last_inserted, head); if (p == &os->samples) {
list_add_tail(&new->list, &os->samples);
os->max_timestamp = timestamp;
return;
}
sample = list_entry(p, struct sample_queue, list);
}
list_add_tail(&new->list, &sample->list);
} else {
while (sample->timestamp > timestamp) {
p = sample->list.prev;
if (p == &os->samples) {
list_add(&new->list, &os->samples);
return;
}
sample = list_entry(p, struct sample_queue, list);
}
list_add(&new->list, &sample->list);
}
} }
static int queue_sample_event(event_t *event, struct sample_data *data, static int queue_sample_event(event_t *event, struct sample_data *data,
...@@ -559,10 +544,6 @@ static int queue_sample_event(event_t *event, struct sample_data *data, ...@@ -559,10 +544,6 @@ static int queue_sample_event(event_t *event, struct sample_data *data,
memcpy(new->event, event, event->header.size); memcpy(new->event, event, event->header.size);
__queue_sample_event(new, s); __queue_sample_event(new, s);
s->ordered_samples.last_inserted = new;
if (new->timestamp > s->ordered_samples.max_timestamp)
s->ordered_samples.max_timestamp = new->timestamp;
return 0; return 0;
} }
......
...@@ -17,8 +17,8 @@ struct ordered_samples { ...@@ -17,8 +17,8 @@ struct ordered_samples {
u64 last_flush; u64 last_flush;
u64 next_flush; u64 next_flush;
u64 max_timestamp; u64 max_timestamp;
struct list_head samples_head; struct list_head samples;
struct sample_queue *last_inserted; struct sample_queue *last_sample;
}; };
struct perf_session { struct perf_session {
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment