linux_dsm_epyc7002/tools/perf/builtin-lock.c

1018 lines
23 KiB
C
Raw Normal View History

#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>
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
static struct perf_session *session;
/* 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)))
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 */
unsigned int nr_acquire;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
unsigned int nr_acquired;
unsigned int nr_contended;
unsigned int nr_release;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
unsigned int nr_readlock;
unsigned int nr_trylock;
/* these times are in nano sec. */
u64 wait_time_total;
u64 wait_time_min;
u64 wait_time_max;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
int discard; /* flag of blacklist */
};
/*
* States of lock_seq_stat
*
* UNINITIALIZED is required for detecting first event of acquire.
* As the nature of lock events, there is no guarantee
* that the first event for the locks are acquire,
* it can be acquired, contended or release.
*/
#define SEQ_STATE_UNINITIALIZED 0 /* initial state */
#define SEQ_STATE_RELEASED 1
#define SEQ_STATE_ACQUIRING 2
#define SEQ_STATE_ACQUIRED 3
#define SEQ_STATE_READ_ACQUIRED 4
#define SEQ_STATE_CONTENDED 5
/*
* MAX_LOCK_DEPTH
* Imported from include/linux/sched.h.
* Should this be synchronized?
*/
#define MAX_LOCK_DEPTH 48
/*
* struct lock_seq_stat:
* Place to put on state of one lock sequence
* 1) acquire -> acquired -> release
* 2) acquire -> contended -> acquired -> release
* 3) acquire (with read or try) -> release
* 4) Are there other patterns?
*/
struct lock_seq_stat {
struct list_head list;
int state;
u64 prev_event_time;
void *addr;
int read_count;
};
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
struct thread_stat {
struct rb_node rb;
u32 tid;
struct list_head seq_list;
};
static struct rb_root thread_stats;
static struct thread_stat *thread_stat_find(u32 tid)
{
struct rb_node *node;
struct thread_stat *st;
node = thread_stats.rb_node;
while (node) {
st = container_of(node, struct thread_stat, rb);
if (st->tid == tid)
return st;
else if (tid < st->tid)
node = node->rb_left;
else
node = node->rb_right;
}
return NULL;
}
static void thread_stat_insert(struct thread_stat *new)
{
struct rb_node **rb = &thread_stats.rb_node;
struct rb_node *parent = NULL;
struct thread_stat *p;
while (*rb) {
p = container_of(*rb, struct thread_stat, rb);
parent = *rb;
if (new->tid < p->tid)
rb = &(*rb)->rb_left;
else if (new->tid > p->tid)
rb = &(*rb)->rb_right;
else
BUG_ON("inserting invalid thread_stat\n");
}
rb_link_node(&new->rb, parent, rb);
rb_insert_color(&new->rb, &thread_stats);
}
static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
{
struct thread_stat *st;
st = thread_stat_find(tid);
if (st)
return st;
st = zalloc(sizeof(struct thread_stat));
if (!st)
die("memory allocation failed\n");
st->tid = tid;
INIT_LIST_HEAD(&st->seq_list);
thread_stat_insert(st);
return st;
}
static struct thread_stat *thread_stat_findnew_first(u32 tid);
static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
thread_stat_findnew_first;
static struct thread_stat *thread_stat_findnew_first(u32 tid)
{
struct thread_stat *st;
st = zalloc(sizeof(struct thread_stat));
if (!st)
die("memory allocation failed\n");
st->tid = tid;
INIT_LIST_HEAD(&st->seq_list);
rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
rb_insert_color(&st->rb, &thread_stats);
thread_stat_findnew = thread_stat_findnew_after_first;
return st;
}
/* 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_max)
static int lock_stat_key_wait_time_min(struct lock_stat *one,
struct lock_stat *two)
{
u64 s1 = one->wait_time_min;
u64 s2 = two->wait_time_min;
if (s1 == ULLONG_MAX)
s1 = 0;
if (s2 == ULLONG_MAX)
s2 = 0;
return s1 > s2;
}
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 *);
static struct rb_root result; /* place to store sorted data */
#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 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)
{
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->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";
struct raw_event_sample {
u32 size;
char data[0];
};
struct trace_acquire_event {
void *addr;
const char *name;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
int flag;
};
struct trace_acquired_event {
void *addr;
const char *name;
};
struct trace_contended_event {
void *addr;
const char *name;
};
struct trace_release_event {
void *addr;
const char *name;
};
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);
};
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
{
struct lock_seq_stat *seq;
list_for_each_entry(seq, &ts->seq_list, list) {
if (seq->addr == addr)
return seq;
}
seq = zalloc(sizeof(struct lock_seq_stat));
if (!seq)
die("Not enough memory\n");
seq->state = SEQ_STATE_UNINITIALIZED;
seq->addr = addr;
list_add(&seq->list, &ts->seq_list);
return seq;
}
enum broken_state {
BROKEN_ACQUIRE,
BROKEN_ACQUIRED,
BROKEN_CONTENDED,
BROKEN_RELEASE,
BROKEN_MAX,
};
static int bad_hist[BROKEN_MAX];
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
enum acquire_flags {
TRY_LOCK = 1,
READ_LOCK = 2,
};
static void
report_lock_acquire_event(struct trace_acquire_event *acquire_event,
struct event *__event __used,
int cpu __used,
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
u64 timestamp __used,
struct thread *thread __used)
{
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
struct lock_stat *ls;
struct thread_stat *ts;
struct lock_seq_stat *seq;
ls = lock_stat_findnew(acquire_event->addr, acquire_event->name);
if (ls->discard)
return;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ts = thread_stat_findnew(thread->pid);
seq = get_seq(ts, acquire_event->addr);
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
switch (seq->state) {
case SEQ_STATE_UNINITIALIZED:
case SEQ_STATE_RELEASED:
if (!acquire_event->flag) {
seq->state = SEQ_STATE_ACQUIRING;
} else {
if (acquire_event->flag & TRY_LOCK)
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls->nr_trylock++;
if (acquire_event->flag & READ_LOCK)
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls->nr_readlock++;
seq->state = SEQ_STATE_READ_ACQUIRED;
seq->read_count = 1;
ls->nr_acquired++;
}
break;
case SEQ_STATE_READ_ACQUIRED:
if (acquire_event->flag & READ_LOCK) {
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
seq->read_count++;
ls->nr_acquired++;
goto end;
} else {
goto broken;
}
break;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
case SEQ_STATE_ACQUIRED:
case SEQ_STATE_ACQUIRING:
case SEQ_STATE_CONTENDED:
broken:
/* broken lock sequence, discard it */
ls->discard = 1;
bad_hist[BROKEN_ACQUIRE]++;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
list_del(&seq->list);
free(seq);
goto end;
break;
default:
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
BUG_ON("Unknown state of lock sequence found!\n");
break;
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls->nr_acquire++;
seq->prev_event_time = timestamp;
end:
return;
}
static void
report_lock_acquired_event(struct trace_acquired_event *acquired_event,
struct event *__event __used,
int cpu __used,
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
u64 timestamp __used,
struct thread *thread __used)
{
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
struct lock_stat *ls;
struct thread_stat *ts;
struct lock_seq_stat *seq;
u64 contended_term;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls = lock_stat_findnew(acquired_event->addr, acquired_event->name);
if (ls->discard)
return;
ts = thread_stat_findnew(thread->pid);
seq = get_seq(ts, acquired_event->addr);
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
switch (seq->state) {
case SEQ_STATE_UNINITIALIZED:
/* orphan event, do nothing */
return;
case SEQ_STATE_ACQUIRING:
break;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
case SEQ_STATE_CONTENDED:
contended_term = timestamp - seq->prev_event_time;
ls->wait_time_total += contended_term;
if (contended_term < ls->wait_time_min)
ls->wait_time_min = contended_term;
if (ls->wait_time_max < contended_term)
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls->wait_time_max = contended_term;
break;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
case SEQ_STATE_RELEASED:
case SEQ_STATE_ACQUIRED:
case SEQ_STATE_READ_ACQUIRED:
/* broken lock sequence, discard it */
ls->discard = 1;
bad_hist[BROKEN_ACQUIRED]++;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
list_del(&seq->list);
free(seq);
goto end;
break;
default:
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
BUG_ON("Unknown state of lock sequence found!\n");
break;
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
seq->state = SEQ_STATE_ACQUIRED;
ls->nr_acquired++;
seq->prev_event_time = timestamp;
end:
return;
}
static void
report_lock_contended_event(struct trace_contended_event *contended_event,
struct event *__event __used,
int cpu __used,
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
u64 timestamp __used,
struct thread *thread __used)
{
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
struct lock_stat *ls;
struct thread_stat *ts;
struct lock_seq_stat *seq;
ls = lock_stat_findnew(contended_event->addr, contended_event->name);
if (ls->discard)
return;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ts = thread_stat_findnew(thread->pid);
seq = get_seq(ts, contended_event->addr);
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
switch (seq->state) {
case SEQ_STATE_UNINITIALIZED:
/* orphan event, do nothing */
return;
case SEQ_STATE_ACQUIRING:
break;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
case SEQ_STATE_RELEASED:
case SEQ_STATE_ACQUIRED:
case SEQ_STATE_READ_ACQUIRED:
case SEQ_STATE_CONTENDED:
/* broken lock sequence, discard it */
ls->discard = 1;
bad_hist[BROKEN_CONTENDED]++;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
list_del(&seq->list);
free(seq);
goto end;
break;
default:
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
BUG_ON("Unknown state of lock sequence found!\n");
break;
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
seq->state = SEQ_STATE_CONTENDED;
ls->nr_contended++;
seq->prev_event_time = timestamp;
end:
return;
}
static void
report_lock_release_event(struct trace_release_event *release_event,
struct event *__event __used,
int cpu __used,
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
u64 timestamp __used,
struct thread *thread __used)
{
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
struct lock_stat *ls;
struct thread_stat *ts;
struct lock_seq_stat *seq;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls = lock_stat_findnew(release_event->addr, release_event->name);
if (ls->discard)
return;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ts = thread_stat_findnew(thread->pid);
seq = get_seq(ts, release_event->addr);
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
switch (seq->state) {
case SEQ_STATE_UNINITIALIZED:
goto end;
break;
case SEQ_STATE_ACQUIRED:
break;
case SEQ_STATE_READ_ACQUIRED:
seq->read_count--;
BUG_ON(seq->read_count < 0);
if (!seq->read_count) {
ls->nr_release++;
goto end;
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
break;
case SEQ_STATE_ACQUIRING:
case SEQ_STATE_CONTENDED:
case SEQ_STATE_RELEASED:
/* broken lock sequence, discard it */
ls->discard = 1;
bad_hist[BROKEN_RELEASE]++;
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
goto free_seq;
break;
default:
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
BUG_ON("Unknown state of lock sequence found!\n");
break;
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
ls->nr_release++;
free_seq:
list_del(&seq->list);
free(seq);
end:
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
return;
}
/* lock oriented handlers */
/* TODO: handlers for CPU oriented, thread oriented */
static struct trace_lock_handler report_lock_ops = {
.acquire_event = report_lock_acquire_event,
.acquired_event = report_lock_acquired_event,
.contended_event = report_lock_contended_event,
.release_event = report_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);
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
acquire_event.flag = (int)raw_field_value(event, "flag", 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);
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);
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);
if (trace_handler->acquire_event)
trace_handler->release_event(&release_event, event, cpu, timestamp, thread);
}
static void
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>
2010-04-24 05:04:12 +07:00
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 void print_bad_events(int bad, int total)
{
/* Output for debug, this have to be removed */
int i;
const char *name[4] =
{ "acquire", "acquired", "contended", "release" };
pr_info("\n=== output for debug===\n\n");
pr_info("bad: %d, total: %d\n", bad, total);
pr_info("bad rate: %f %%\n", (double)bad / (double)total * 100);
pr_info("histogram of events caused bad sequence\n");
for (i = 0; i < BROKEN_MAX; i++)
pr_info(" %10s: %d\n", name[i], bad_hist[i]);
}
/* TODO: various way to print, coloring, nano or milli sec */
static void print_result(void)
{
struct lock_stat *st;
char cut_name[20];
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
int bad, total;
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("%20s ", "Name");
pr_info("%10s ", "acquired");
pr_info("%10s ", "contended");
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("%15s ", "total wait (ns)");
pr_info("%15s ", "max wait (ns)");
pr_info("%15s ", "min wait (ns)");
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("\n\n");
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
bad = total = 0;
while ((st = pop_from_result())) {
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
total++;
if (st->discard) {
bad++;
continue;
}
bzero(cut_name, 20);
if (strlen(st->name) < 16) {
/* output raw name */
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("%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 */
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("%20s ", cut_name);
}
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("%10u ", st->nr_acquired);
pr_info("%10u ", st->nr_contended);
pr_info("%15" PRIu64 " ", st->wait_time_total);
pr_info("%15" PRIu64 " ", st->wait_time_max);
pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
0 : st->wait_time_min);
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("\n");
}
perf lock: Fix state machine to recognize lock sequence Previous state machine of perf lock was really broken. This patch improves it a little. This patch prepares the list of state machine that represents lock sequences for each threads. These state machines can be one of these sequences: 1) acquire -> acquired -> release 2) acquire -> contended -> acquired -> release 3) acquire (w/ try) -> release 4) acquire (w/ read) -> release The case of 4) is a little special. Double acquire of read lock is allowed, so the state machine counts read lock number, and permits double acquire and release. But, things are not so simple. Something in my model is still wrong. I counted the number of lock instances with bad sequence, and ratio is like this (case of tracing whoami): bad:233, total:2279 version 2: * threads are now identified with tid, not pid * prepared SEQ_STATE_READ_ACQUIRED for read lock. * bunch of struct lock_seq_stat is now linked list * debug information enhanced (this have to be removed someday) e.g. | === output for debug=== | | bad:233, total:2279 | bad rate:0.000000 | histogram of events caused bad sequence | acquire: 165 | acquired: 0 | contended: 0 | release: 68 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: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1271852634-9351-1-git-send-email-mitake@dcl.info.waseda.ac.jp> [rename SEQ_STATE_UNINITED to SEQ_STATE_UNINITIALIZED] Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-04-21 19:23:54 +07:00
print_bad_events(bad, total);
}
static bool info_threads, info_map;
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
static void dump_threads(void)
{
struct thread_stat *st;
struct rb_node *node;
struct thread *t;
pr_info("%10s: comm\n", "Thread ID");
node = rb_first(&thread_stats);
while (node) {
st = container_of(node, struct thread_stat, rb);
t = perf_session__findnew(session, st->tid);
pr_info("%10d: %s\n", st->tid, t->comm);
node = rb_next(node);
};
}
static void dump_map(void)
{
unsigned int i;
struct lock_stat *st;
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info("Address of instance: name of class\n");
for (i = 0; i < LOCKHASH_SIZE; i++) {
list_for_each_entry(st, &lockhash_table[i], hash_entry) {
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
pr_info(" %p: %s\n", st->addr, st->name);
}
}
}
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
static void dump_info(void)
{
if (info_threads)
dump_threads();
else if (info_map)
dump_map();
else
die("Unknown type of information\n");
}
static int process_sample_event(union perf_event *event,
struct perf_sample *sample,
struct perf_evsel *evsel __used,
struct perf_session *s)
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>
2010-04-24 05:04:12 +07:00
{
struct thread *thread = perf_session__findnew(s, sample->tid);
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>
2010-04-24 05:04:12 +07:00
if (thread == NULL) {
pr_debug("problem processing %d event, skipping it.\n",
event->header.type);
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>
2010-04-24 05:04:12 +07:00
return -1;
}
process_raw_event(sample->raw_data, sample->cpu, sample->time, thread);
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>
2010-04-24 05:04:12 +07:00
return 0;
}
static struct perf_event_ops eops = {
.sample = process_sample_event,
.comm = perf_event__process_comm,
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>
2010-04-24 05:04:12 +07:00
.ordered_samples = true,
};
static int read_events(void)
{
perf session: Fallback to unordered processing if no sample_id_all If we are running the new perf on an old kernel without support for sample_id_all, we should fall back to the old unordered processing of events. If we didn't than we would *always* process events without timestamps out of order, whether or not we hit a reordering race. In other words, instead of there being a chance of not attributing samples correctly, we would guarantee that samples would not be attributed. While processing all events without timestamps before events with timestamps may seem like an intuitive solution, it falls down as PERF_RECORD_EXIT events would also be processed before any samples. Even with a workaround for that case, samples before/after an exec would not be attributed correctly. This patch allows commands to indicate whether they need to fall back to unordered processing, so that commands that do not care about timestamps on every event will not be affected. If we do fallback, this will print out a warning if report -D was invoked. This patch adds the test in perf_session__new so that we only need to test once per session. Commands that do not use an event_ops (such as record and top) can simply pass NULL in it's place. Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> LKML-Reference: <1291951882-sup-6069@au1.ibm.com> Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-12-10 10:09:16 +07:00
session = perf_session__new(input_name, O_RDONLY, 0, false, &eops);
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_report(void)
{
setup_pager();
select_key();
read_events();
sort_result();
print_result();
}
static const char * const report_usage[] = {
"perf lock report [<options>]",
NULL
};
static const struct option report_options[] = {
OPT_STRING('k', "key", &sort_key, "acquired",
"key for sorting (acquired / contended / wait_total / wait_max / wait_min)"),
/* TODO: type */
OPT_END()
};
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
static const char * const info_usage[] = {
"perf lock info [<options>]",
NULL
};
static const struct option info_options[] = {
OPT_BOOLEAN('t', "threads", &info_threads,
"dump thread list in perf.data"),
OPT_BOOLEAN('m', "map", &info_map,
"map of lock instances (name:address table)"),
OPT_END()
};
static const char * const lock_usage[] = {
"perf lock [<options>] {record|trace|report}",
NULL
};
static const struct option lock_options[] = {
OPT_STRING('i', "input", &input_name, "file", "input file name"),
perf: Fix endianness argument compatibility with OPT_BOOLEAN() and introduce OPT_INCR() Parsing an option from the command line with OPT_BOOLEAN on a bool data type would not work on a big-endian machine due to the manner in which the boolean was being cast into an int and incremented. For example, running 'perf probe --list' on a PowerPC machine would fail to properly set the list_events bool and would therefore print out the usage information and terminate. This patch makes OPT_BOOLEAN work as expected with a bool datatype. For cases where the original OPT_BOOLEAN was intentionally being used to increment an int each time it was passed in on the command line, this patch introduces OPT_INCR with the old behaviour of OPT_BOOLEAN (the verbose variable is currently the only such example of this). I have reviewed every use of OPT_BOOLEAN to verify that a true C99 bool was passed. Where integers were used, I verified that they were only being used for boolean logic and changed them to bools to ensure that they would not be mistakenly used as ints. The major exception was the verbose variable which now uses OPT_INCR instead of OPT_BOOLEAN. Signed-off-by: Ian Munsie <imunsie@au.ibm.com> Acked-by: David S. Miller <davem@davemloft.net> Cc: <stable@kernel.org> # NOTE: wont apply to .3[34].x cleanly, please backport Cc: Git development list <git@vger.kernel.org> Cc: Ian Munsie <imunsie@au1.ibm.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Eric B Munson <ebmunson@us.ibm.com> Cc: Valdis.Kletnieks@vt.edu Cc: WANG Cong <amwang@redhat.com> Cc: Thiago Farina <tfransosi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Jaswinder Singh Rajput <jaswinderrajput@gmail.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> Cc: Mike Galbraith <efault@gmx.de> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Anton Blanchard <anton@samba.org> Cc: John Kacur <jkacur@redhat.com> Cc: Li Zefan <lizf@cn.fujitsu.com> Cc: Steven Rostedt <rostedt@goodmis.org> LKML-Reference: <1271147857-11604-1-git-send-email-imunsie@au.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-04-13 15:37:33 +07:00
OPT_INCR('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",
"-R",
"-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 *));
if (rec_argv == NULL)
return -ENOMEM;
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], "report", 6)) {
trace_handler = &report_lock_ops;
if (argc) {
argc = parse_options(argc, argv,
report_options, report_usage, 0);
if (argc)
usage_with_options(report_usage, report_options);
}
__cmd_report();
} else if (!strcmp(argv[0], "script")) {
/* Aliased to 'perf script' */
return cmd_script(argc, argv, prefix);
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
} else if (!strcmp(argv[0], "info")) {
if (argc) {
argc = parse_options(argc, argv,
info_options, info_usage, 0);
if (argc)
usage_with_options(info_usage, info_options);
}
/* recycling report_lock_ops */
trace_handler = &report_lock_ops;
setup_pager();
read_events();
perf lock: Add "info" subcommand for dumping misc information This adds the "info" subcommand to perf lock which can be used to dump metadata like threads or addresses of lock instances. "map" was removed because info should do the work for it. This will be useful not only for debugging but also for ordinary analyzing. v2: adding example of usage % sudo ./perf lock info -t | Thread ID: comm | 0: swapper | 1: init | 18: migration/5 | 29: events/2 | 32: events/5 | 33: events/6 ... % sudo ./perf lock info -m | Address of instance: name of class | 0xffff8800b95adae0: &(&sighand->siglock)->rlock | 0xffff8800bbb41ae0: &(&sighand->siglock)->rlock | 0xffff8800bf165ae0: &(&sighand->siglock)->rlock | 0xffff8800b9576a98: &p->cred_guard_mutex | 0xffff8800bb890a08: &(&p->alloc_lock)->rlock | 0xffff8800b9522a08: &(&p->alloc_lock)->rlock | 0xffff8800bb8aaa08: &(&p->alloc_lock)->rlock | 0xffff8800bba72a08: &(&p->alloc_lock)->rlock | 0xffff8800bf18ea08: &(&p->alloc_lock)->rlock | 0xffff8800b8a0d8a0: &(&ip->i_lock)->mr_lock | 0xffff88009bf818a0: &(&ip->i_lock)->mr_lock | 0xffff88004c66b8a0: &(&ip->i_lock)->mr_lock | 0xffff8800bb6478a0: &(shost->host_lock)->rlock v3: fixed some problems Frederic pointed out * better rbtree tracking in dump_threads() * removed printf() and used pr_info() and pr_debug() Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1272863520-16179-1-git-send-email-mitake@dcl.info.waseda.ac.jp> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
2010-05-03 12:12:00 +07:00
dump_info();
} else {
usage_with_options(lock_usage, lock_options);
}
return 0;
}