static char const *input_name = "perf.data";
-static unsigned long total_comm = 0;
-
-static struct rb_root threads;
-static struct thread *last_match;
-
static struct perf_header *header;
static u64 sample_type;
static char default_sort_order[] = "avg, max, switch, runtime";
static char *sort_order = default_sort_order;
-static char *cwd;
-static int cwdlen;
+static int profile_cpu = -1;
#define PR_SET_NAME 15 /* Set process name */
#define MAX_CPUS 4096
-#define BUG_ON(x) assert(!(x))
-
static u64 run_measurement_overhead;
static u64 sleep_measurement_overhead;
SCHED_EVENT_RUN,
SCHED_EVENT_SLEEP,
SCHED_EVENT_WAKEUP,
+ SCHED_EVENT_MIGRATION,
};
struct sched_atom {
static struct sched_atom *
get_new_event(struct task_desc *task, u64 timestamp)
{
- struct sched_atom *event = calloc(1, sizeof(*event));
+ struct sched_atom *event = zalloc(sizeof(*event));
unsigned long idx = task->nr_events;
size_t size;
return;
}
- wakee_event->wait_sem = calloc(1, sizeof(*wakee_event->wait_sem));
+ wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
sem_init(wakee_event->wait_sem, 0, 0);
wakee_event->specific_wait = 1;
event->wait_sem = wakee_event->wait_sem;
if (task)
return task;
- task = calloc(1, sizeof(*task));
+ task = zalloc(sizeof(*task));
task->pid = pid;
task->nr = nr_tasks;
strcpy(task->comm, comm);
ret = sem_post(atom->wait_sem);
BUG_ON(ret);
break;
+ case SCHED_EVENT_MIGRATION:
+ break;
default:
BUG_ON(1);
}
printf("the sleep test took %Ld nsecs\n", T1-T0);
}
-static int
-process_comm_event(event_t *event, unsigned long offset, unsigned long head)
-{
- struct thread *thread;
-
- thread = threads__findnew(event->comm.pid, &threads, &last_match);
-
- dump_printf("%p [%p]: perf_event_comm: %s:%d\n",
- (void *)(offset + head),
- (void *)(long)(event->header.size),
- event->comm.comm, event->comm.pid);
-
- if (thread == NULL ||
- thread__set_comm(thread, event->comm.comm)) {
- dump_printf("problem processing perf_event_comm, skipping event.\n");
- return -1;
- }
- total_comm++;
-
- return 0;
-}
-
-
struct raw_event_sample {
u32 size;
char data[0];
u32 child_pid;
};
+struct trace_migrate_task_event {
+ u32 size;
+
+ u16 common_type;
+ u8 common_flags;
+ u8 common_preempt_count;
+ u32 common_pid;
+ u32 common_tgid;
+
+ char comm[16];
+ u32 pid;
+
+ u32 prio;
+ u32 cpu;
+};
+
struct trace_sched_handler {
void (*switch_event)(struct trace_switch_event *,
struct event *,
int cpu,
u64 timestamp,
struct thread *thread);
+
+ void (*migrate_task_event)(struct trace_migrate_task_event *,
+ struct event *,
+ int cpu,
+ u64 timestamp,
+ struct thread *thread);
};
static void thread_atoms_insert(struct thread *thread)
{
- struct work_atoms *atoms;
-
- atoms = calloc(sizeof(*atoms), 1);
+ struct work_atoms *atoms = zalloc(sizeof(*atoms));
if (!atoms)
die("No memory");
char run_state,
u64 timestamp)
{
- struct work_atom *atom;
-
- atom = calloc(sizeof(*atom), 1);
+ struct work_atom *atom = zalloc(sizeof(*atom));
if (!atom)
die("Non memory");
die("hm, delta: %Ld < 0 ?\n", delta);
- sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
- sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
+ sched_out = threads__findnew(switch_event->prev_pid);
+ sched_in = threads__findnew(switch_event->next_pid);
out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
if (!out_events) {
u64 timestamp,
struct thread *this_thread __used)
{
- struct work_atoms *atoms;
- struct thread *thread;
+ struct thread *thread = threads__findnew(runtime_event->pid);
+ struct work_atoms *atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
BUG_ON(cpu >= MAX_CPUS || cpu < 0);
-
- thread = threads__findnew(runtime_event->pid, &threads, &last_match);
- atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
if (!atoms) {
thread_atoms_insert(thread);
atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
if (!wakeup_event->success)
return;
- wakee = threads__findnew(wakeup_event->pid, &threads, &last_match);
+ wakee = threads__findnew(wakeup_event->pid);
atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
if (!atoms) {
thread_atoms_insert(wakee);
atom = list_entry(atoms->work_list.prev, struct work_atom, list);
- if (atom->state != THREAD_SLEEPING)
+ /*
+ * You WILL be missing events if you've recorded only
+ * one CPU, or are only looking at only one, so don't
+ * make useless noise.
+ */
+ if (profile_cpu == -1 && atom->state != THREAD_SLEEPING)
nr_state_machine_bugs++;
nr_timestamps++;
atom->wake_up_time = timestamp;
}
+static void
+latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event,
+ struct event *__event __used,
+ int cpu __used,
+ u64 timestamp,
+ struct thread *thread __used)
+{
+ struct work_atoms *atoms;
+ struct work_atom *atom;
+ struct thread *migrant;
+
+ /*
+ * Only need to worry about migration when profiling one CPU.
+ */
+ if (profile_cpu == -1)
+ return;
+
+ migrant = threads__findnew(migrate_task_event->pid);
+ atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
+ if (!atoms) {
+ thread_atoms_insert(migrant);
+ register_pid(migrant->pid, migrant->comm);
+ atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
+ if (!atoms)
+ die("migration-event: Internal tree error");
+ add_sched_out_event(atoms, 'R', timestamp);
+ }
+
+ BUG_ON(list_empty(&atoms->work_list));
+
+ atom = list_entry(atoms->work_list.prev, struct work_atom, list);
+ atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
+
+ nr_timestamps++;
+
+ if (atom->sched_out_time > timestamp)
+ nr_unordered_timestamps++;
+}
+
static struct trace_sched_handler lat_ops = {
.wakeup_event = latency_wakeup_event,
.switch_event = latency_switch_event,
.runtime_event = latency_runtime_event,
.fork_event = latency_fork_event,
+ .migrate_task_event = latency_migrate_task_event,
};
static void output_lat_thread(struct work_atoms *work_list)
static LIST_HEAD(sort_list);
-static int sort_dimension__add(char *tok, struct list_head *list)
+static int sort_dimension__add(const char *tok, struct list_head *list)
{
int i;
die("hm, delta: %Ld < 0 ?\n", delta);
- sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
- sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
+ sched_out = threads__findnew(switch_event->prev_pid);
+ sched_in = threads__findnew(switch_event->next_pid);
curr_thread[this_cpu] = sched_in;
}
static void
+process_sched_migrate_task_event(struct raw_event_sample *raw,
+ struct event *event,
+ int cpu __used,
+ u64 timestamp __used,
+ struct thread *thread __used)
+{
+ struct trace_migrate_task_event migrate_task_event;
+
+ FILL_COMMON_FIELDS(migrate_task_event, event, raw->data);
+
+ FILL_ARRAY(migrate_task_event, comm, event, raw->data);
+ FILL_FIELD(migrate_task_event, pid, event, raw->data);
+ FILL_FIELD(migrate_task_event, prio, event, raw->data);
+ FILL_FIELD(migrate_task_event, cpu, event, raw->data);
+
+ if (trace_handler->migrate_task_event)
+ trace_handler->migrate_task_event(&migrate_task_event, event, cpu, timestamp, thread);
+}
+
+static void
process_raw_event(event_t *raw_event __used, void *more_data,
int cpu, u64 timestamp, struct thread *thread)
{
process_sched_fork_event(raw, event, cpu, timestamp, thread);
if (!strcmp(event->name, "sched_process_exit"))
process_sched_exit_event(event, cpu, timestamp, thread);
+ if (!strcmp(event->name, "sched_migrate_task"))
+ process_sched_migrate_task_event(raw, event, cpu, timestamp, thread);
}
-static int
-process_sample_event(event_t *event, unsigned long offset, unsigned long head)
+static int process_sample_event(event_t *event)
{
struct thread *thread;
u64 ip = event->ip.ip;
if (!(sample_type & PERF_SAMPLE_RAW))
return 0;
- thread = threads__findnew(event->ip.pid, &threads, &last_match);
+ thread = threads__findnew(event->ip.pid);
if (sample_type & PERF_SAMPLE_TIME) {
timestamp = *(u64 *)more_data;
more_data += sizeof(u64);
}
- dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
- (void *)(offset + head),
- (void *)(long)(event->header.size),
+ dump_printf("(IP, %d): %d/%d: %p period: %Ld\n",
event->header.misc,
event->ip.pid, event->ip.tid,
(void *)(long)ip,
(long long)period);
- dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
-
if (thread == NULL) {
- eprintf("problem processing %d event, skipping it.\n",
- event->header.type);
+ pr_debug("problem processing %d event, skipping it.\n",
+ event->header.type);
return -1;
}
+ dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
+
+ if (profile_cpu != -1 && profile_cpu != (int) cpu)
+ return 0;
+
process_raw_event(event, more_data, cpu, timestamp, thread);
return 0;
}
-static int
-process_lost_event(event_t *event __used,
- unsigned long offset __used,
- unsigned long head __used)
+static int process_lost_event(event_t *event __used)
{
nr_lost_chunks++;
nr_lost_events += event->lost.lost;
static struct perf_file_handler file_handler = {
.process_sample_event = process_sample_event,
- .process_comm_event = process_comm_event,
+ .process_comm_event = event__process_comm,
.process_lost_event = process_lost_event,
.sample_type_check = sample_type_check,
};
static int read_events(void)
{
- register_idle_thread(&threads, &last_match);
+ register_idle_thread();
register_perf_file_handler(&file_handler);
- return mmap_dispatch_perf_file(&header, input_name, 0, 0, &cwdlen, &cwd);
+ return mmap_dispatch_perf_file(&header, input_name, 0, 0,
+ &event__cwdlen, &event__cwd);
}
static void print_bad_events(void)
"sort by key(s): runtime, switch, avg, max"),
OPT_BOOLEAN('v', "verbose", &verbose,
"be more verbose (show symbol address, etc)"),
+ OPT_INTEGER('C', "CPU", &profile_cpu,
+ "CPU to profile on"),
OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
"dump raw trace in ASCII"),
OPT_END()
free(str);
- sort_dimension__add((char *)"pid", &cmp_pid);
+ sort_dimension__add("pid", &cmp_pid);
}
static const char *record_args[] = {
int cmd_sched(int argc, const char **argv, const char *prefix __used)
{
- symbol__init();
+ symbol__init(0);
argc = parse_options(argc, argv, sched_options, sched_usage,
PARSE_OPT_STOP_AT_NON_OPTION);