perf sched: Use perf_tool as ancestor
authorArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 11 Sep 2012 20:29:27 +0000 (17:29 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 11 Sep 2012 20:29:27 +0000 (17:29 -0300)
So that we can remove all the globals.

Before:

   text    data     bss     dec     hex filename
1586833  110368 1438600 3135801  2fd939 /tmp/oldperf

After:

   text    data     bss     dec     hex filename
1629329   93568  848328 2571225  273bd9 /root/bin/perf

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-oph40vikij0crjz4eyapneov@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-sched.c

index af11b1aa1bd73f71f880076a5d9b402485c853a5..79f88fa3f7a3f72e943382162a894fe979cfbb53 100644 (file)
 #include <pthread.h>
 #include <math.h>
 
-static const char              *input_name;
-
-static char                    default_sort_order[] = "avg, max, switch, runtime";
-static const char              *sort_order = default_sort_order;
-
-static int                     profile_cpu = -1;
-
 #define PR_SET_NAME            15               /* Set process name */
 #define MAX_CPUS               4096
-
-static u64                     run_measurement_overhead;
-static u64                     sleep_measurement_overhead;
-
 #define COMM_LEN               20
 #define SYM_LEN                        129
-
 #define MAX_PID                        65536
 
-static unsigned long           nr_tasks;
-
 struct sched_atom;
 
 struct task_desc {
@@ -80,44 +66,6 @@ struct sched_atom {
        struct task_desc        *wakee;
 };
 
-static struct task_desc                *pid_to_task[MAX_PID];
-
-static struct task_desc                **tasks;
-
-static pthread_mutex_t         start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
-static u64                     start_time;
-
-static pthread_mutex_t         work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
-
-static unsigned long           nr_run_events;
-static unsigned long           nr_sleep_events;
-static unsigned long           nr_wakeup_events;
-
-static unsigned long           nr_sleep_corrections;
-static unsigned long           nr_run_events_optimized;
-
-static unsigned long           targetless_wakeups;
-static unsigned long           multitarget_wakeups;
-
-static u64                     cpu_usage;
-static u64                     runavg_cpu_usage;
-static u64                     parent_cpu_usage;
-static u64                     runavg_parent_cpu_usage;
-
-static unsigned long           nr_runs;
-static u64                     sum_runtime;
-static u64                     sum_fluct;
-static u64                     run_avg;
-
-static unsigned int            replay_repeat = 10;
-static unsigned long           nr_timestamps;
-static unsigned long           nr_unordered_timestamps;
-static unsigned long           nr_state_machine_bugs;
-static unsigned long           nr_context_switch_bugs;
-static unsigned long           nr_events;
-static unsigned long           nr_lost_chunks;
-static unsigned long           nr_lost_events;
-
 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
 
 enum thread_state {
@@ -149,11 +97,169 @@ struct work_atoms {
 
 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
 
-static struct rb_root          atom_root, sorted_atom_root;
+struct trace_switch_event {
+       u32  size;
+
+       u16  common_type;
+       u8   common_flags;
+       u8   common_preempt_count;
+       u32  common_pid;
+       u32  common_tgid;
+
+       char prev_comm[16];
+       u32  prev_pid;
+       u32  prev_prio;
+       u64  prev_state;
+       char next_comm[16];
+       u32  next_pid;
+       u32  next_prio;
+};
+
+struct trace_runtime_event {
+       u32  size;
+
+       u16  common_type;
+       u8   common_flags;
+       u8   common_preempt_count;
+       u32  common_pid;
+       u32  common_tgid;
+
+       char comm[16];
+       u32  pid;
+       u64  runtime;
+       u64  vruntime;
+};
+
+struct trace_wakeup_event {
+       u32  size;
 
-static u64                     all_runtime;
-static u64                     all_count;
+       u16  common_type;
+       u8   common_flags;
+       u8   common_preempt_count;
+       u32  common_pid;
+       u32  common_tgid;
 
+       char comm[16];
+       u32  pid;
+
+       u32  prio;
+       u32  success;
+       u32  cpu;
+};
+
+struct trace_fork_event {
+       u32  size;
+
+       u16  common_type;
+       u8   common_flags;
+       u8   common_preempt_count;
+       u32  common_pid;
+       u32  common_tgid;
+
+       char parent_comm[16];
+       u32  parent_pid;
+       char child_comm[16];
+       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 perf_sched;
+
+struct trace_sched_handler {
+       int (*switch_event)(struct perf_sched *sched,
+                           struct trace_switch_event *event,
+                           struct machine *machine,
+                           struct event_format *tp_format,
+                           struct perf_sample *sample);
+
+       int (*runtime_event)(struct perf_sched *sched,
+                            struct trace_runtime_event *event,
+                            struct machine *machine,
+                            struct perf_sample *sample);
+
+       int (*wakeup_event)(struct perf_sched *sched,
+                           struct trace_wakeup_event *event,
+                           struct machine *machine,
+                           struct event_format *tp_format,
+                           struct perf_sample *sample);
+
+       int (*fork_event)(struct perf_sched *sched,
+                         struct trace_fork_event *event,
+                         struct event_format *tp_format);
+
+       int (*migrate_task_event)(struct perf_sched *sched,
+                                 struct trace_migrate_task_event *event,
+                                 struct machine *machine,
+                                 struct perf_sample *sample);
+};
+
+struct perf_sched {
+       struct perf_tool tool;
+       const char       *input_name;
+       const char       *sort_order;
+       unsigned long    nr_tasks;
+       struct task_desc *pid_to_task[MAX_PID];
+       struct task_desc **tasks;
+       const struct trace_sched_handler *tp_handler;
+       pthread_mutex_t  start_work_mutex;
+       pthread_mutex_t  work_done_wait_mutex;
+       int              profile_cpu;
+/*
+ * Track the current task - that way we can know whether there's any
+ * weird events, such as a task being switched away that is not current.
+ */
+       int              max_cpu;
+       u32              curr_pid[MAX_CPUS];
+       struct thread    *curr_thread[MAX_CPUS];
+       char             next_shortname1;
+       char             next_shortname2;
+       unsigned int     replay_repeat;
+       unsigned long    nr_run_events;
+       unsigned long    nr_sleep_events;
+       unsigned long    nr_wakeup_events;
+       unsigned long    nr_sleep_corrections;
+       unsigned long    nr_run_events_optimized;
+       unsigned long    targetless_wakeups;
+       unsigned long    multitarget_wakeups;
+       unsigned long    nr_runs;
+       unsigned long    nr_timestamps;
+       unsigned long    nr_unordered_timestamps;
+       unsigned long    nr_state_machine_bugs;
+       unsigned long    nr_context_switch_bugs;
+       unsigned long    nr_events;
+       unsigned long    nr_lost_chunks;
+       unsigned long    nr_lost_events;
+       u64              run_measurement_overhead;
+       u64              sleep_measurement_overhead;
+       u64              start_time;
+       u64              cpu_usage;
+       u64              runavg_cpu_usage;
+       u64              parent_cpu_usage;
+       u64              runavg_parent_cpu_usage;
+       u64              sum_runtime;
+       u64              sum_fluct;
+       u64              run_avg;
+       u64              all_runtime;
+       u64              all_count;
+       u64              cpu_last_switched[MAX_CPUS];
+       struct rb_root   atom_root, sorted_atom_root;
+       struct list_head sort_list, cmp_pid;
+};
 
 static u64 get_nsecs(void)
 {
@@ -164,13 +270,13 @@ static u64 get_nsecs(void)
        return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
 }
 
-static void burn_nsecs(u64 nsecs)
+static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
 {
        u64 T0 = get_nsecs(), T1;
 
        do {
                T1 = get_nsecs();
-       } while (T1 + run_measurement_overhead < T0 + nsecs);
+       } while (T1 + sched->run_measurement_overhead < T0 + nsecs);
 }
 
 static void sleep_nsecs(u64 nsecs)
@@ -183,24 +289,24 @@ static void sleep_nsecs(u64 nsecs)
        nanosleep(&ts, NULL);
 }
 
-static void calibrate_run_measurement_overhead(void)
+static void calibrate_run_measurement_overhead(struct perf_sched *sched)
 {
        u64 T0, T1, delta, min_delta = 1000000000ULL;
        int i;
 
        for (i = 0; i < 10; i++) {
                T0 = get_nsecs();
-               burn_nsecs(0);
+               burn_nsecs(sched, 0);
                T1 = get_nsecs();
                delta = T1-T0;
                min_delta = min(min_delta, delta);
        }
-       run_measurement_overhead = min_delta;
+       sched->run_measurement_overhead = min_delta;
 
        printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
 }
 
-static void calibrate_sleep_measurement_overhead(void)
+static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
 {
        u64 T0, T1, delta, min_delta = 1000000000ULL;
        int i;
@@ -213,7 +319,7 @@ static void calibrate_sleep_measurement_overhead(void)
                min_delta = min(min_delta, delta);
        }
        min_delta -= 10000;
-       sleep_measurement_overhead = min_delta;
+       sched->sleep_measurement_overhead = min_delta;
 
        printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
 }
@@ -246,8 +352,8 @@ static struct sched_atom *last_event(struct task_desc *task)
        return task->atoms[task->nr_events - 1];
 }
 
-static void
-add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
+static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
+                               u64 timestamp, u64 duration)
 {
        struct sched_atom *event, *curr_event = last_event(task);
 
@@ -256,7 +362,7 @@ add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
         * to it:
         */
        if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
-               nr_run_events_optimized++;
+               sched->nr_run_events_optimized++;
                curr_event->duration += duration;
                return;
        }
@@ -266,12 +372,11 @@ add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
        event->type = SCHED_EVENT_RUN;
        event->duration = duration;
 
-       nr_run_events++;
+       sched->nr_run_events++;
 }
 
-static void
-add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
-                      struct task_desc *wakee)
+static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
+                                  u64 timestamp, struct task_desc *wakee)
 {
        struct sched_atom *event, *wakee_event;
 
@@ -281,11 +386,11 @@ add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
 
        wakee_event = last_event(wakee);
        if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
-               targetless_wakeups++;
+               sched->targetless_wakeups++;
                return;
        }
        if (wakee_event->wait_sem) {
-               multitarget_wakeups++;
+               sched->multitarget_wakeups++;
                return;
        }
 
@@ -294,89 +399,89 @@ add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
        wakee_event->specific_wait = 1;
        event->wait_sem = wakee_event->wait_sem;
 
-       nr_wakeup_events++;
+       sched->nr_wakeup_events++;
 }
 
-static void
-add_sched_event_sleep(struct task_desc *task, u64 timestamp,
-                     u64 task_state __maybe_unused)
+static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
+                                 u64 timestamp, u64 task_state __maybe_unused)
 {
        struct sched_atom *event = get_new_event(task, timestamp);
 
        event->type = SCHED_EVENT_SLEEP;
 
-       nr_sleep_events++;
+       sched->nr_sleep_events++;
 }
 
-static struct task_desc *register_pid(unsigned long pid, const char *comm)
+static struct task_desc *register_pid(struct perf_sched *sched,
+                                     unsigned long pid, const char *comm)
 {
        struct task_desc *task;
 
        BUG_ON(pid >= MAX_PID);
 
-       task = pid_to_task[pid];
+       task = sched->pid_to_task[pid];
 
        if (task)
                return task;
 
        task = zalloc(sizeof(*task));
        task->pid = pid;
-       task->nr = nr_tasks;
+       task->nr = sched->nr_tasks;
        strcpy(task->comm, comm);
        /*
         * every task starts in sleeping state - this gets ignored
         * if there's no wakeup pointing to this sleep state:
         */
-       add_sched_event_sleep(task, 0, 0);
+       add_sched_event_sleep(sched, task, 0, 0);
 
-       pid_to_task[pid] = task;
-       nr_tasks++;
-       tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *));
-       BUG_ON(!tasks);
-       tasks[task->nr] = task;
+       sched->pid_to_task[pid] = task;
+       sched->nr_tasks++;
+       sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_task *));
+       BUG_ON(!sched->tasks);
+       sched->tasks[task->nr] = task;
 
        if (verbose)
-               printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm);
+               printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
 
        return task;
 }
 
 
-static void print_task_traces(void)
+static void print_task_traces(struct perf_sched *sched)
 {
        struct task_desc *task;
        unsigned long i;
 
-       for (i = 0; i < nr_tasks; i++) {
-               task = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               task = sched->tasks[i];
                printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
                        task->nr, task->comm, task->pid, task->nr_events);
        }
 }
 
-static void add_cross_task_wakeups(void)
+static void add_cross_task_wakeups(struct perf_sched *sched)
 {
        struct task_desc *task1, *task2;
        unsigned long i, j;
 
-       for (i = 0; i < nr_tasks; i++) {
-               task1 = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               task1 = sched->tasks[i];
                j = i + 1;
-               if (j == nr_tasks)
+               if (j == sched->nr_tasks)
                        j = 0;
-               task2 = tasks[j];
-               add_sched_event_wakeup(task1, 0, task2);
+               task2 = sched->tasks[j];
+               add_sched_event_wakeup(sched, task1, 0, task2);
        }
 }
 
-static void process_sched_event(struct task_desc *this_task __maybe_unused,
-                               struct sched_atom *atom)
+static void perf_sched__process_event(struct perf_sched *sched,
+                                     struct sched_atom *atom)
 {
        int ret = 0;
 
        switch (atom->type) {
                case SCHED_EVENT_RUN:
-                       burn_nsecs(atom->duration);
+                       burn_nsecs(sched, atom->duration);
                        break;
                case SCHED_EVENT_SLEEP:
                        if (atom->wait_sem)
@@ -439,14 +544,23 @@ static u64 get_cpu_usage_nsec_self(int fd)
        return runtime;
 }
 
+struct sched_thread_parms {
+       struct task_desc  *task;
+       struct perf_sched *sched;
+};
+
 static void *thread_func(void *ctx)
 {
-       struct task_desc *this_task = ctx;
+       struct sched_thread_parms *parms = ctx;
+       struct task_desc *this_task = parms->task;
+       struct perf_sched *sched = parms->sched;
        u64 cpu_usage_0, cpu_usage_1;
        unsigned long i, ret;
        char comm2[22];
        int fd;
 
+       free(parms);
+
        sprintf(comm2, ":%s", this_task->comm);
        prctl(PR_SET_NAME, comm2);
        fd = self_open_counters();
@@ -455,16 +569,16 @@ static void *thread_func(void *ctx)
 again:
        ret = sem_post(&this_task->ready_for_work);
        BUG_ON(ret);
-       ret = pthread_mutex_lock(&start_work_mutex);
+       ret = pthread_mutex_lock(&sched->start_work_mutex);
        BUG_ON(ret);
-       ret = pthread_mutex_unlock(&start_work_mutex);
+       ret = pthread_mutex_unlock(&sched->start_work_mutex);
        BUG_ON(ret);
 
        cpu_usage_0 = get_cpu_usage_nsec_self(fd);
 
        for (i = 0; i < this_task->nr_events; i++) {
                this_task->curr_event = i;
-               process_sched_event(this_task, this_task->atoms[i]);
+               perf_sched__process_event(sched, this_task->atoms[i]);
        }
 
        cpu_usage_1 = get_cpu_usage_nsec_self(fd);
@@ -472,15 +586,15 @@ again:
        ret = sem_post(&this_task->work_done_sem);
        BUG_ON(ret);
 
-       ret = pthread_mutex_lock(&work_done_wait_mutex);
+       ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
        BUG_ON(ret);
-       ret = pthread_mutex_unlock(&work_done_wait_mutex);
+       ret = pthread_mutex_unlock(&sched->work_done_wait_mutex);
        BUG_ON(ret);
 
        goto again;
 }
 
-static void create_tasks(void)
+static void create_tasks(struct perf_sched *sched)
 {
        struct task_desc *task;
        pthread_attr_t attr;
@@ -492,128 +606,129 @@ static void create_tasks(void)
        err = pthread_attr_setstacksize(&attr,
                        (size_t) max(16 * 1024, PTHREAD_STACK_MIN));
        BUG_ON(err);
-       err = pthread_mutex_lock(&start_work_mutex);
+       err = pthread_mutex_lock(&sched->start_work_mutex);
        BUG_ON(err);
-       err = pthread_mutex_lock(&work_done_wait_mutex);
+       err = pthread_mutex_lock(&sched->work_done_wait_mutex);
        BUG_ON(err);
-       for (i = 0; i < nr_tasks; i++) {
-               task = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               struct sched_thread_parms *parms = malloc(sizeof(*parms));
+               BUG_ON(parms == NULL);
+               parms->task = task = sched->tasks[i];
+               parms->sched = sched;
                sem_init(&task->sleep_sem, 0, 0);
                sem_init(&task->ready_for_work, 0, 0);
                sem_init(&task->work_done_sem, 0, 0);
                task->curr_event = 0;
-               err = pthread_create(&task->thread, &attr, thread_func, task);
+               err = pthread_create(&task->thread, &attr, thread_func, parms);
                BUG_ON(err);
        }
 }
 
-static void wait_for_tasks(void)
+static void wait_for_tasks(struct perf_sched *sched)
 {
        u64 cpu_usage_0, cpu_usage_1;
        struct task_desc *task;
        unsigned long i, ret;
 
-       start_time = get_nsecs();
-       cpu_usage = 0;
-       pthread_mutex_unlock(&work_done_wait_mutex);
+       sched->start_time = get_nsecs();
+       sched->cpu_usage = 0;
+       pthread_mutex_unlock(&sched->work_done_wait_mutex);
 
-       for (i = 0; i < nr_tasks; i++) {
-               task = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               task = sched->tasks[i];
                ret = sem_wait(&task->ready_for_work);
                BUG_ON(ret);
                sem_init(&task->ready_for_work, 0, 0);
        }
-       ret = pthread_mutex_lock(&work_done_wait_mutex);
+       ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
        BUG_ON(ret);
 
        cpu_usage_0 = get_cpu_usage_nsec_parent();
 
-       pthread_mutex_unlock(&start_work_mutex);
+       pthread_mutex_unlock(&sched->start_work_mutex);
 
-       for (i = 0; i < nr_tasks; i++) {
-               task = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               task = sched->tasks[i];
                ret = sem_wait(&task->work_done_sem);
                BUG_ON(ret);
                sem_init(&task->work_done_sem, 0, 0);
-               cpu_usage += task->cpu_usage;
+               sched->cpu_usage += task->cpu_usage;
                task->cpu_usage = 0;
        }
 
        cpu_usage_1 = get_cpu_usage_nsec_parent();
-       if (!runavg_cpu_usage)
-               runavg_cpu_usage = cpu_usage;
-       runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10;
+       if (!sched->runavg_cpu_usage)
+               sched->runavg_cpu_usage = sched->cpu_usage;
+       sched->runavg_cpu_usage = (sched->runavg_cpu_usage * 9 + sched->cpu_usage) / 10;
 
-       parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
-       if (!runavg_parent_cpu_usage)
-               runavg_parent_cpu_usage = parent_cpu_usage;
-       runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 +
-                                  parent_cpu_usage)/10;
+       sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
+       if (!sched->runavg_parent_cpu_usage)
+               sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
+       sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * 9 +
+                                        sched->parent_cpu_usage)/10;
 
-       ret = pthread_mutex_lock(&start_work_mutex);
+       ret = pthread_mutex_lock(&sched->start_work_mutex);
        BUG_ON(ret);
 
-       for (i = 0; i < nr_tasks; i++) {
-               task = tasks[i];
+       for (i = 0; i < sched->nr_tasks; i++) {
+               task = sched->tasks[i];
                sem_init(&task->sleep_sem, 0, 0);
                task->curr_event = 0;
        }
 }
 
-static void run_one_test(void)
+static void run_one_test(struct perf_sched *sched)
 {
        u64 T0, T1, delta, avg_delta, fluct;
 
        T0 = get_nsecs();
-       wait_for_tasks();
+       wait_for_tasks(sched);
        T1 = get_nsecs();
 
        delta = T1 - T0;
-       sum_runtime += delta;
-       nr_runs++;
+       sched->sum_runtime += delta;
+       sched->nr_runs++;
 
-       avg_delta = sum_runtime / nr_runs;
+       avg_delta = sched->sum_runtime / sched->nr_runs;
        if (delta < avg_delta)
                fluct = avg_delta - delta;
        else
                fluct = delta - avg_delta;
-       sum_fluct += fluct;
-       if (!run_avg)
-               run_avg = delta;
-       run_avg = (run_avg*9 + delta)/10;
+       sched->sum_fluct += fluct;
+       if (!sched->run_avg)
+               sched->run_avg = delta;
+       sched->run_avg = (sched->run_avg * 9 + delta) / 10;
 
-       printf("#%-3ld: %0.3f, ",
-               nr_runs, (double)delta/1000000.0);
+       printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0);
 
-       printf("ravg: %0.2f, ",
-               (double)run_avg/1e6);
+       printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6);
 
        printf("cpu: %0.2f / %0.2f",
-               (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6);
+               (double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6);
 
 #if 0
        /*
         * rusage statistics done by the parent, these are less
-        * accurate than the sum_exec_runtime based statistics:
+        * accurate than the sched->sum_exec_runtime based statistics:
         */
        printf(" [%0.2f / %0.2f]",
-               (double)parent_cpu_usage/1e6,
-               (double)runavg_parent_cpu_usage/1e6);
+               (double)sched->parent_cpu_usage/1e6,
+               (double)sched->runavg_parent_cpu_usage/1e6);
 #endif
 
        printf("\n");
 
-       if (nr_sleep_corrections)
-               printf(" (%ld sleep corrections)\n", nr_sleep_corrections);
-       nr_sleep_corrections = 0;
+       if (sched->nr_sleep_corrections)
+               printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
+       sched->nr_sleep_corrections = 0;
 }
 
-static void test_calibrations(void)
+static void test_calibrations(struct perf_sched *sched)
 {
        u64 T0, T1;
 
        T0 = get_nsecs();
-       burn_nsecs(1e6);
+       burn_nsecs(sched, 1e6);
        T1 = get_nsecs();
 
        printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
@@ -643,115 +758,9 @@ do {                                                              \
        FILL_FIELD(ptr, common_tgid, event, data);              \
 } while (0)
 
-
-
-struct trace_switch_event {
-       u32 size;
-
-       u16 common_type;
-       u8 common_flags;
-       u8 common_preempt_count;
-       u32 common_pid;
-       u32 common_tgid;
-
-       char prev_comm[16];
-       u32 prev_pid;
-       u32 prev_prio;
-       u64 prev_state;
-       char next_comm[16];
-       u32 next_pid;
-       u32 next_prio;
-};
-
-struct trace_runtime_event {
-       u32 size;
-
-       u16 common_type;
-       u8 common_flags;
-       u8 common_preempt_count;
-       u32 common_pid;
-       u32 common_tgid;
-
-       char comm[16];
-       u32 pid;
-       u64 runtime;
-       u64 vruntime;
-};
-
-struct trace_wakeup_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 success;
-       u32 cpu;
-};
-
-struct trace_fork_event {
-       u32 size;
-
-       u16 common_type;
-       u8 common_flags;
-       u8 common_preempt_count;
-       u32 common_pid;
-       u32 common_tgid;
-
-       char parent_comm[16];
-       u32 parent_pid;
-       char child_comm[16];
-       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 {
-       int (*switch_event)(struct trace_switch_event *event,
-                           struct machine *machine,
-                           struct event_format *tp_format,
-                           struct perf_sample *sample);
-
-       int (*runtime_event)(struct trace_runtime_event *event,
-                            struct machine *machine,
-                            struct perf_sample *sample);
-
-       int (*wakeup_event)(struct trace_wakeup_event *event,
-                           struct machine *machine,
-                           struct event_format *tp_format,
-                           struct perf_sample *sample);
-
-       int (*fork_event)(struct trace_fork_event *event,
-                         struct event_format *tp_format);
-
-       int (*migrate_task_event)(struct trace_migrate_task_event *event,
-                                 struct machine *machine,
-                                 struct perf_sample *sample);
-};
-
-
 static int
-replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
+replay_wakeup_event(struct perf_sched *sched,
+                   struct trace_wakeup_event *wakeup_event,
                    struct machine *machine __maybe_unused,
                    struct event_format *event, struct perf_sample *sample)
 {
@@ -761,22 +770,19 @@ replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
                printf("sched_wakeup event %p\n", event);
 
                printf(" ... pid %d woke up %s/%d\n",
-                       wakeup_event->common_pid,
-                       wakeup_event->comm,
-                       wakeup_event->pid);
+                      wakeup_event->common_pid, wakeup_event->comm, wakeup_event->pid);
        }
 
-       waker = register_pid(wakeup_event->common_pid, "<unknown>");
-       wakee = register_pid(wakeup_event->pid, wakeup_event->comm);
+       waker = register_pid(sched, wakeup_event->common_pid, "<unknown>");
+       wakee = register_pid(sched, wakeup_event->pid, wakeup_event->comm);
 
-       add_sched_event_wakeup(waker, sample->time, wakee);
+       add_sched_event_wakeup(sched, waker, sample->time, wakee);
        return 0;
 }
 
-static u64 cpu_last_switched[MAX_CPUS];
-
 static int
-replay_switch_event(struct trace_switch_event *switch_event,
+replay_switch_event(struct perf_sched *sched,
+                   struct trace_switch_event *switch_event,
                    struct machine *machine __maybe_unused,
                    struct event_format *event,
                    struct perf_sample *sample)
@@ -792,7 +798,7 @@ replay_switch_event(struct trace_switch_event *switch_event,
        if (cpu >= MAX_CPUS || cpu < 0)
                return 0;
 
-       timestamp0 = cpu_last_switched[cpu];
+       timestamp0 = sched->cpu_last_switched[cpu];
        if (timestamp0)
                delta = timestamp - timestamp0;
        else
@@ -810,20 +816,19 @@ replay_switch_event(struct trace_switch_event *switch_event,
                        delta);
        }
 
-       prev = register_pid(switch_event->prev_pid, switch_event->prev_comm);
-       next = register_pid(switch_event->next_pid, switch_event->next_comm);
+       prev = register_pid(sched, switch_event->prev_pid, switch_event->prev_comm);
+       next = register_pid(sched, switch_event->next_pid, switch_event->next_comm);
 
-       cpu_last_switched[cpu] = timestamp;
+       sched->cpu_last_switched[cpu] = timestamp;
 
-       add_sched_event_run(prev, timestamp, delta);
-       add_sched_event_sleep(prev, timestamp, switch_event->prev_state);
+       add_sched_event_run(sched, prev, timestamp, delta);
+       add_sched_event_sleep(sched, prev, timestamp, switch_event->prev_state);
 
        return 0;
 }
 
-
 static int
-replay_fork_event(struct trace_fork_event *fork_event,
+replay_fork_event(struct perf_sched *sched, struct trace_fork_event *fork_event,
                  struct event_format *event)
 {
        if (verbose) {
@@ -831,25 +836,17 @@ replay_fork_event(struct trace_fork_event *fork_event,
                printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid);
                printf("...  child: %s/%d\n", fork_event->child_comm, fork_event->child_pid);
        }
-       register_pid(fork_event->parent_pid, fork_event->parent_comm);
-       register_pid(fork_event->child_pid, fork_event->child_comm);
+       register_pid(sched, fork_event->parent_pid, fork_event->parent_comm);
+       register_pid(sched, fork_event->child_pid, fork_event->child_comm);
        return 0;
 }
 
-static struct trace_sched_handler replay_ops  = {
-       .wakeup_event           = replay_wakeup_event,
-       .switch_event           = replay_switch_event,
-       .fork_event             = replay_fork_event,
-};
-
 struct sort_dimension {
        const char              *name;
        sort_fn_t               cmp;
        struct list_head        list;
 };
 
-static LIST_HEAD(cmp_pid);
-
 static int
 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
 {
@@ -918,7 +915,7 @@ __thread_latency_insert(struct rb_root *root, struct work_atoms *data,
        rb_insert_color(&data->node, root);
 }
 
-static int thread_atoms_insert(struct thread *thread)
+static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
 {
        struct work_atoms *atoms = zalloc(sizeof(*atoms));
        if (!atoms) {
@@ -928,11 +925,12 @@ static int thread_atoms_insert(struct thread *thread)
 
        atoms->thread = thread;
        INIT_LIST_HEAD(&atoms->work_list);
-       __thread_latency_insert(&atom_root, atoms, &cmp_pid);
+       __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
        return 0;
 }
 
-static int latency_fork_event(struct trace_fork_event *fork_event __maybe_unused,
+static int latency_fork_event(struct perf_sched *sched __maybe_unused,
+                             struct trace_fork_event *fork_event __maybe_unused,
                              struct event_format *event __maybe_unused)
 {
        /* should insert the newcomer */
@@ -1014,7 +1012,8 @@ add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
 }
 
 static int
-latency_switch_event(struct trace_switch_event *switch_event,
+latency_switch_event(struct perf_sched *sched,
+                    struct trace_switch_event *switch_event,
                     struct machine *machine,
                     struct event_format *event __maybe_unused,
                     struct perf_sample *sample)
@@ -1027,8 +1026,8 @@ latency_switch_event(struct trace_switch_event *switch_event,
 
        BUG_ON(cpu >= MAX_CPUS || cpu < 0);
 
-       timestamp0 = cpu_last_switched[cpu];
-       cpu_last_switched[cpu] = timestamp;
+       timestamp0 = sched->cpu_last_switched[cpu];
+       sched->cpu_last_switched[cpu] = timestamp;
        if (timestamp0)
                delta = timestamp - timestamp0;
        else
@@ -1042,11 +1041,11 @@ latency_switch_event(struct trace_switch_event *switch_event,
        sched_out = machine__findnew_thread(machine, switch_event->prev_pid);
        sched_in = machine__findnew_thread(machine, switch_event->next_pid);
 
-       out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
+       out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
        if (!out_events) {
-               if (thread_atoms_insert(sched_out))
+               if (thread_atoms_insert(sched, sched_out))
                        return -1;
-               out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
+               out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
                if (!out_events) {
                        pr_err("out-event: Internal tree error");
                        return -1;
@@ -1055,11 +1054,11 @@ latency_switch_event(struct trace_switch_event *switch_event,
        if (add_sched_out_event(out_events, sched_out_state(switch_event), timestamp))
                return -1;
 
-       in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
+       in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
        if (!in_events) {
-               if (thread_atoms_insert(sched_in))
+               if (thread_atoms_insert(sched, sched_in))
                        return -1;
-               in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
+               in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
                if (!in_events) {
                        pr_err("in-event: Internal tree error");
                        return -1;
@@ -1077,19 +1076,20 @@ latency_switch_event(struct trace_switch_event *switch_event,
 }
 
 static int
-latency_runtime_event(struct trace_runtime_event *runtime_event,
+latency_runtime_event(struct perf_sched *sched,
+                     struct trace_runtime_event *runtime_event,
                      struct machine *machine, struct perf_sample *sample)
 {
        struct thread *thread = machine__findnew_thread(machine, runtime_event->pid);
-       struct work_atoms *atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
+       struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
        u64 timestamp = sample->time;
        int cpu = sample->cpu;
 
        BUG_ON(cpu >= MAX_CPUS || cpu < 0);
        if (!atoms) {
-               if (thread_atoms_insert(thread))
+               if (thread_atoms_insert(sched, thread))
                        return -1;
-               atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
+               atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
                if (!atoms) {
                        pr_debug("in-event: Internal tree error");
                        return -1;
@@ -1103,7 +1103,8 @@ latency_runtime_event(struct trace_runtime_event *runtime_event,
 }
 
 static int
-latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
+latency_wakeup_event(struct perf_sched *sched,
+                    struct trace_wakeup_event *wakeup_event,
                     struct machine *machine,
                     struct event_format *event __maybe_unused,
                     struct perf_sample *sample)
@@ -1118,11 +1119,11 @@ latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
                return 0;
 
        wakee = machine__findnew_thread(machine, wakeup_event->pid);
-       atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
+       atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
        if (!atoms) {
-               if (thread_atoms_insert(wakee))
+               if (thread_atoms_insert(sched, wakee))
                        return -1;
-               atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
+               atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
                if (!atoms) {
                        pr_debug("wakeup-event: Internal tree error");
                        return -1;
@@ -1140,12 +1141,12 @@ latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
         * 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++;
+       if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
+               sched->nr_state_machine_bugs++;
 
-       nr_timestamps++;
+       sched->nr_timestamps++;
        if (atom->sched_out_time > timestamp) {
-               nr_unordered_timestamps++;
+               sched->nr_unordered_timestamps++;
                return 0;
        }
 
@@ -1155,7 +1156,8 @@ latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
 }
 
 static int
-latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event,
+latency_migrate_task_event(struct perf_sched *sched,
+                          struct trace_migrate_task_event *migrate_task_event,
                           struct machine *machine, struct perf_sample *sample)
 {
        u64 timestamp = sample->time;
@@ -1166,16 +1168,16 @@ latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event,
        /*
         * Only need to worry about migration when profiling one CPU.
         */
-       if (profile_cpu == -1)
+       if (sched->profile_cpu == -1)
                return 0;
 
        migrant = machine__findnew_thread(machine, migrate_task_event->pid);
-       atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
+       atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
        if (!atoms) {
-               if (thread_atoms_insert(migrant))
+               if (thread_atoms_insert(sched, migrant))
                        return -1;
-               register_pid(migrant->pid, migrant->comm);
-               atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
+               register_pid(sched, migrant->pid, migrant->comm);
+               atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
                if (!atoms) {
                        pr_debug("migration-event: Internal tree error");
                        return -1;
@@ -1189,23 +1191,15 @@ latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event,
        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++;
+       sched->nr_timestamps++;
 
        if (atom->sched_out_time > timestamp)
-               nr_unordered_timestamps++;
+               sched->nr_unordered_timestamps++;
 
        return 0;
 }
 
-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 void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
 {
        int i;
        int ret;
@@ -1219,8 +1213,8 @@ static void output_lat_thread(struct work_atoms *work_list)
        if (!strcmp(work_list->thread->comm, "swapper"))
                return;
 
-       all_runtime += work_list->total_runtime;
-       all_count += work_list->nb_atoms;
+       sched->all_runtime += work_list->total_runtime;
+       sched->all_count   += work_list->nb_atoms;
 
        ret = printf("  %s:%d ", work_list->thread->comm, work_list->thread->pid);
 
@@ -1246,11 +1240,6 @@ static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
        return 0;
 }
 
-static struct sort_dimension pid_sort_dimension = {
-       .name                   = "pid",
-       .cmp                    = pid_cmp,
-};
-
 static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
 {
        u64 avgl, avgr;
@@ -1272,11 +1261,6 @@ static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
        return 0;
 }
 
-static struct sort_dimension avg_sort_dimension = {
-       .name                   = "avg",
-       .cmp                    = avg_cmp,
-};
-
 static int max_cmp(struct work_atoms *l, struct work_atoms *r)
 {
        if (l->max_lat < r->max_lat)
@@ -1287,11 +1271,6 @@ static int max_cmp(struct work_atoms *l, struct work_atoms *r)
        return 0;
 }
 
-static struct sort_dimension max_sort_dimension = {
-       .name                   = "max",
-       .cmp                    = max_cmp,
-};
-
 static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
 {
        if (l->nb_atoms < r->nb_atoms)
@@ -1302,11 +1281,6 @@ static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
        return 0;
 }
 
-static struct sort_dimension switch_sort_dimension = {
-       .name                   = "switch",
-       .cmp                    = switch_cmp,
-};
-
 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
 {
        if (l->total_runtime < r->total_runtime)
@@ -1317,28 +1291,38 @@ static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
        return 0;
 }
 
-static struct sort_dimension runtime_sort_dimension = {
-       .name                   = "runtime",
-       .cmp                    = runtime_cmp,
-};
-
-static struct sort_dimension *available_sorts[] = {
-       &pid_sort_dimension,
-       &avg_sort_dimension,
-       &max_sort_dimension,
-       &switch_sort_dimension,
-       &runtime_sort_dimension,
-};
-
-#define NB_AVAILABLE_SORTS     (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *))
-
-static LIST_HEAD(sort_list);
-
 static int sort_dimension__add(const char *tok, struct list_head *list)
 {
-       int i;
+       size_t i;
+       static struct sort_dimension avg_sort_dimension = {
+               .name = "avg",
+               .cmp  = avg_cmp,
+       };
+       static struct sort_dimension max_sort_dimension = {
+               .name = "max",
+               .cmp  = max_cmp,
+       };
+       static struct sort_dimension pid_sort_dimension = {
+               .name = "pid",
+               .cmp  = pid_cmp,
+       };
+       static struct sort_dimension runtime_sort_dimension = {
+               .name = "runtime",
+               .cmp  = runtime_cmp,
+       };
+       static struct sort_dimension switch_sort_dimension = {
+               .name = "switch",
+               .cmp  = switch_cmp,
+       };
+       struct sort_dimension *available_sorts[] = {
+               &pid_sort_dimension,
+               &avg_sort_dimension,
+               &max_sort_dimension,
+               &switch_sort_dimension,
+               &runtime_sort_dimension,
+       };
 
-       for (i = 0; i < NB_AVAILABLE_SORTS; i++) {
+       for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
                if (!strcmp(available_sorts[i]->name, tok)) {
                        list_add_tail(&available_sorts[i]->list, list);
 
@@ -1349,31 +1333,28 @@ static int sort_dimension__add(const char *tok, struct list_head *list)
        return -1;
 }
 
-static void setup_sorting(void);
-
-static void sort_lat(void)
+static void perf_sched__sort_lat(struct perf_sched *sched)
 {
        struct rb_node *node;
 
        for (;;) {
                struct work_atoms *data;
-               node = rb_first(&atom_root);
+               node = rb_first(&sched->atom_root);
                if (!node)
                        break;
 
-               rb_erase(node, &atom_root);
+               rb_erase(node, &sched->atom_root);
                data = rb_entry(node, struct work_atoms, node);
-               __thread_latency_insert(&sorted_atom_root, data, &sort_list);
+               __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
        }
 }
 
-static struct trace_sched_handler *trace_handler;
-
-static int process_sched_wakeup_event(struct perf_tool *tool __maybe_unused,
+static int process_sched_wakeup_event(struct perf_tool *tool,
                                      struct event_format *event,
                                      struct perf_sample *sample,
                                      struct machine *machine)
 {
+       struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
        void *data = sample->raw_data;
        struct trace_wakeup_event wakeup_event;
        int err = 0;
@@ -1386,27 +1367,15 @@ static int process_sched_wakeup_event(struct perf_tool *tool __maybe_unused,
        FILL_FIELD(wakeup_event, success, event, data);
        FILL_FIELD(wakeup_event, cpu, event, data);
 
-       if (trace_handler->wakeup_event)
-               err = trace_handler->wakeup_event(&wakeup_event, machine, event, sample);
+       if (sched->tp_handler->wakeup_event)
+               err = sched->tp_handler->wakeup_event(sched, &wakeup_event, machine, event, sample);
 
        return err;
 }
 
-/*
- * Track the current task - that way we can know whether there's any
- * weird events, such as a task being switched away that is not current.
- */
-static int max_cpu;
-
-static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
-
-static struct thread *curr_thread[MAX_CPUS];
-
-static char next_shortname1 = 'A';
-static char next_shortname2 = '0';
-
 static int
-map_switch_event(struct trace_switch_event *switch_event,
+map_switch_event(struct perf_sched *sched,
+                struct trace_switch_event *switch_event,
                 struct machine *machine,
                 struct event_format *event __maybe_unused,
                 struct perf_sample *sample)
@@ -1419,11 +1388,11 @@ map_switch_event(struct trace_switch_event *switch_event,
 
        BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
 
-       if (this_cpu > max_cpu)
-               max_cpu = this_cpu;
+       if (this_cpu > sched->max_cpu)
+               sched->max_cpu = this_cpu;
 
-       timestamp0 = cpu_last_switched[this_cpu];
-       cpu_last_switched[this_cpu] = timestamp;
+       timestamp0 = sched->cpu_last_switched[this_cpu];
+       sched->cpu_last_switched[this_cpu] = timestamp;
        if (timestamp0)
                delta = timestamp - timestamp0;
        else
@@ -1437,37 +1406,37 @@ map_switch_event(struct trace_switch_event *switch_event,
        sched_out = machine__findnew_thread(machine, switch_event->prev_pid);
        sched_in = machine__findnew_thread(machine, switch_event->next_pid);
 
-       curr_thread[this_cpu] = sched_in;
+       sched->curr_thread[this_cpu] = sched_in;
 
        printf("  ");
 
        new_shortname = 0;
        if (!sched_in->shortname[0]) {
-               sched_in->shortname[0] = next_shortname1;
-               sched_in->shortname[1] = next_shortname2;
+               sched_in->shortname[0] = sched->next_shortname1;
+               sched_in->shortname[1] = sched->next_shortname2;
 
-               if (next_shortname1 < 'Z') {
-                       next_shortname1++;
+               if (sched->next_shortname1 < 'Z') {
+                       sched->next_shortname1++;
                } else {
-                       next_shortname1='A';
-                       if (next_shortname2 < '9') {
-                               next_shortname2++;
+                       sched->next_shortname1='A';
+                       if (sched->next_shortname2 < '9') {
+                               sched->next_shortname2++;
                        } else {
-                               next_shortname2='0';
+                               sched->next_shortname2='0';
                        }
                }
                new_shortname = 1;
        }
 
-       for (cpu = 0; cpu <= max_cpu; cpu++) {
+       for (cpu = 0; cpu <= sched->max_cpu; cpu++) {
                if (cpu != this_cpu)
                        printf(" ");
                else
                        printf("*");
 
-               if (curr_thread[cpu]) {
-                       if (curr_thread[cpu]->pid)
-                               printf("%2s ", curr_thread[cpu]->shortname);
+               if (sched->curr_thread[cpu]) {
+                       if (sched->curr_thread[cpu]->pid)
+                               printf("%2s ", sched->curr_thread[cpu]->shortname);
                        else
                                printf(".  ");
                } else
@@ -1485,11 +1454,12 @@ map_switch_event(struct trace_switch_event *switch_event,
        return 0;
 }
 
-static int process_sched_switch_event(struct perf_tool *tool __maybe_unused,
+static int process_sched_switch_event(struct perf_tool *tool,
                                      struct event_format *event,
                                      struct perf_sample *sample,
                                      struct machine *machine)
 {
+       struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
        int this_cpu = sample->cpu, err = 0;
        void *data = sample->raw_data;
        struct trace_switch_event switch_event;
@@ -1504,26 +1474,27 @@ static int process_sched_switch_event(struct perf_tool *tool __maybe_unused,
        FILL_FIELD(switch_event, next_pid, event, data);
        FILL_FIELD(switch_event, next_prio, event, data);
 
-       if (curr_pid[this_cpu] != (u32)-1) {
+       if (sched->curr_pid[this_cpu] != (u32)-1) {
                /*
                 * Are we trying to switch away a PID that is
                 * not current?
                 */
-               if (curr_pid[this_cpu] != switch_event.prev_pid)
-                       nr_context_switch_bugs++;
+               if (sched->curr_pid[this_cpu] != switch_event.prev_pid)
+                       sched->nr_context_switch_bugs++;
        }
-       if (trace_handler->switch_event)
-               err = trace_handler->switch_event(&switch_event, machine, event, sample);
+       if (sched->tp_handler->switch_event)
+               err = sched->tp_handler->switch_event(sched, &switch_event, machine, event, sample);
 
-       curr_pid[this_cpu] = switch_event.next_pid;
+       sched->curr_pid[this_cpu] = switch_event.next_pid;
        return err;
 }
 
-static int process_sched_runtime_event(struct perf_tool *tool __maybe_unused,
+static int process_sched_runtime_event(struct perf_tool *tool,
                                       struct event_format *event,
                                       struct perf_sample *sample,
                                       struct machine *machine)
 {
+       struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
        void *data = sample->raw_data;
        struct trace_runtime_event runtime_event;
        int err = 0;
@@ -1533,17 +1504,18 @@ static int process_sched_runtime_event(struct perf_tool *tool __maybe_unused,
        FILL_FIELD(runtime_event, runtime, event, data);
        FILL_FIELD(runtime_event, vruntime, event, data);
 
-       if (trace_handler->runtime_event)
-               err = trace_handler->runtime_event(&runtime_event, machine, sample);
+       if (sched->tp_handler->runtime_event)
+               err = sched->tp_handler->runtime_event(sched, &runtime_event, machine, sample);
 
        return err;
 }
 
-static int process_sched_fork_event(struct perf_tool *tool __maybe_unused,
+static int process_sched_fork_event(struct perf_tool *tool,
                                    struct event_format *event,
                                    struct perf_sample *sample,
                                    struct machine *machine __maybe_unused)
 {
+       struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
        void *data = sample->raw_data;
        struct trace_fork_event fork_event;
        int err = 0;
@@ -1555,8 +1527,8 @@ static int process_sched_fork_event(struct perf_tool *tool __maybe_unused,
        FILL_ARRAY(fork_event, child_comm, event, data);
        FILL_FIELD(fork_event, child_pid, event, data);
 
-       if (trace_handler->fork_event)
-               err = trace_handler->fork_event(&fork_event, event);
+       if (sched->tp_handler->fork_event)
+               err = sched->tp_handler->fork_event(sched, &fork_event, event);
 
        return err;
 }
@@ -1572,11 +1544,12 @@ static int process_sched_exit_event(struct perf_tool *tool __maybe_unused,
        return 0;
 }
 
-static int process_sched_migrate_task_event(struct perf_tool *tool __maybe_unused,
+static int process_sched_migrate_task_event(struct perf_tool *tool,
                                            struct event_format *event,
                                            struct perf_sample *sample,
                                            struct machine *machine)
 {
+       struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
        void *data = sample->raw_data;
        struct trace_migrate_task_event migrate_task_event;
        int err = 0;
@@ -1588,8 +1561,8 @@ static int process_sched_migrate_task_event(struct perf_tool *tool __maybe_unuse
        FILL_FIELD(migrate_task_event, prio, event, data);
        FILL_FIELD(migrate_task_event, cpu, event, data);
 
-       if (trace_handler->migrate_task_event)
-               err = trace_handler->migrate_task_event(&migrate_task_event, machine, sample);
+       if (sched->tp_handler->migrate_task_event)
+               err = sched->tp_handler->migrate_task_event(sched, &migrate_task_event, machine, sample);
 
        return err;
 }
@@ -1625,15 +1598,8 @@ static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_
        return err;
 }
 
-static struct perf_tool perf_sched = {
-       .sample          = perf_sched__process_tracepoint_sample,
-       .comm            = perf_event__process_comm,
-       .lost            = perf_event__process_lost,
-       .fork            = perf_event__process_task,
-       .ordered_samples = true,
-};
-
-static int read_events(bool destroy, struct perf_session **psession)
+static int perf_sched__read_events(struct perf_sched *sched, bool destroy,
+                                  struct perf_session **psession)
 {
        const struct perf_evsel_str_handler handlers[] = {
                { "sched:sched_switch",       process_sched_switch_event, },
@@ -1646,7 +1612,7 @@ static int read_events(bool destroy, struct perf_session **psession)
        };
        struct perf_session *session;
 
-       session = perf_session__new(input_name, O_RDONLY, 0, false, &perf_sched);
+       session = perf_session__new(sched->input_name, O_RDONLY, 0, false, &sched->tool);
        if (session == NULL) {
                pr_debug("No Memory for session\n");
                return -1;
@@ -1656,15 +1622,15 @@ static int read_events(bool destroy, struct perf_session **psession)
                goto out_delete;
 
        if (perf_session__has_traces(session, "record -R")) {
-               int err = perf_session__process_events(session, &perf_sched);
+               int err = perf_session__process_events(session, &sched->tool);
                if (err) {
                        pr_err("Failed to process events, error %d", err);
                        goto out_delete;
                }
 
-               nr_events      = session->hists.stats.nr_events[0];
-               nr_lost_events = session->hists.stats.total_lost;
-               nr_lost_chunks = session->hists.stats.nr_events[PERF_RECORD_LOST];
+               sched->nr_events      = session->hists.stats.nr_events[0];
+               sched->nr_lost_events = session->hists.stats.total_lost;
+               sched->nr_lost_chunks = session->hists.stats.nr_events[PERF_RECORD_LOST];
        }
 
        if (destroy)
@@ -1680,213 +1646,158 @@ out_delete:
        return -1;
 }
 
-static void print_bad_events(void)
+static void print_bad_events(struct perf_sched *sched)
 {
-       if (nr_unordered_timestamps && nr_timestamps) {
+       if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
                printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
-                       (double)nr_unordered_timestamps/(double)nr_timestamps*100.0,
-                       nr_unordered_timestamps, nr_timestamps);
+                       (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
+                       sched->nr_unordered_timestamps, sched->nr_timestamps);
        }
-       if (nr_lost_events && nr_events) {
+       if (sched->nr_lost_events && sched->nr_events) {
                printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
-                       (double)nr_lost_events/(double)nr_events*100.0,
-                       nr_lost_events, nr_events, nr_lost_chunks);
+                       (double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
+                       sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
        }
-       if (nr_state_machine_bugs && nr_timestamps) {
+       if (sched->nr_state_machine_bugs && sched->nr_timestamps) {
                printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
-                       (double)nr_state_machine_bugs/(double)nr_timestamps*100.0,
-                       nr_state_machine_bugs, nr_timestamps);
-               if (nr_lost_events)
+                       (double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0,
+                       sched->nr_state_machine_bugs, sched->nr_timestamps);
+               if (sched->nr_lost_events)
                        printf(" (due to lost events?)");
                printf("\n");
        }
-       if (nr_context_switch_bugs && nr_timestamps) {
+       if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
                printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
-                       (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
-                       nr_context_switch_bugs, nr_timestamps);
-               if (nr_lost_events)
+                       (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
+                       sched->nr_context_switch_bugs, sched->nr_timestamps);
+               if (sched->nr_lost_events)
                        printf(" (due to lost events?)");
                printf("\n");
        }
 }
 
-static int __cmd_lat(void)
+static int perf_sched__lat(struct perf_sched *sched)
 {
        struct rb_node *next;
        struct perf_session *session;
 
        setup_pager();
-       if (read_events(false, &session))
+       if (perf_sched__read_events(sched, false, &session))
                return -1;
-       sort_lat();
+       perf_sched__sort_lat(sched);
 
        printf("\n ---------------------------------------------------------------------------------------------------------------\n");
        printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms | Maximum delay at     |\n");
        printf(" ---------------------------------------------------------------------------------------------------------------\n");
 
-       next = rb_first(&sorted_atom_root);
+       next = rb_first(&sched->sorted_atom_root);
 
        while (next) {
                struct work_atoms *work_list;
 
                work_list = rb_entry(next, struct work_atoms, node);
-               output_lat_thread(work_list);
+               output_lat_thread(sched, work_list);
                next = rb_next(next);
        }
 
        printf(" -----------------------------------------------------------------------------------------\n");
        printf("  TOTAL:                |%11.3f ms |%9" PRIu64 " |\n",
-               (double)all_runtime/1e6, all_count);
+               (double)sched->all_runtime / 1e6, sched->all_count);
 
        printf(" ---------------------------------------------------\n");
 
-       print_bad_events();
+       print_bad_events(sched);
        printf("\n");
 
        perf_session__delete(session);
        return 0;
 }
 
-static struct trace_sched_handler map_ops  = {
-       .wakeup_event           = NULL,
-       .switch_event           = map_switch_event,
-       .runtime_event          = NULL,
-       .fork_event             = NULL,
-};
-
-static int __cmd_map(void)
+static int perf_sched__map(struct perf_sched *sched)
 {
-       max_cpu = sysconf(_SC_NPROCESSORS_CONF);
+       sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF);
 
        setup_pager();
-       if (read_events(true, NULL))
+       if (perf_sched__read_events(sched, true, NULL))
                return -1;
-       print_bad_events();
+       print_bad_events(sched);
        return 0;
 }
 
-static int __cmd_replay(void)
+static int perf_sched__replay(struct perf_sched *sched)
 {
        unsigned long i;
 
-       calibrate_run_measurement_overhead();
-       calibrate_sleep_measurement_overhead();
+       calibrate_run_measurement_overhead(sched);
+       calibrate_sleep_measurement_overhead(sched);
 
-       test_calibrations();
+       test_calibrations(sched);
 
-       if (read_events(true, NULL))
+       if (perf_sched__read_events(sched, true, NULL))
                return -1;
 
-       printf("nr_run_events:        %ld\n", nr_run_events);
-       printf("nr_sleep_events:      %ld\n", nr_sleep_events);
-       printf("nr_wakeup_events:     %ld\n", nr_wakeup_events);
+       printf("nr_run_events:        %ld\n", sched->nr_run_events);
+       printf("nr_sleep_events:      %ld\n", sched->nr_sleep_events);
+       printf("nr_wakeup_events:     %ld\n", sched->nr_wakeup_events);
 
-       if (targetless_wakeups)
-               printf("target-less wakeups:  %ld\n", targetless_wakeups);
-       if (multitarget_wakeups)
-               printf("multi-target wakeups: %ld\n", multitarget_wakeups);
-       if (nr_run_events_optimized)
+       if (sched->targetless_wakeups)
+               printf("target-less wakeups:  %ld\n", sched->targetless_wakeups);
+       if (sched->multitarget_wakeups)
+               printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
+       if (sched->nr_run_events_optimized)
                printf("run atoms optimized: %ld\n",
-                       nr_run_events_optimized);
+                       sched->nr_run_events_optimized);
 
-       print_task_traces();
-       add_cross_task_wakeups();
+       print_task_traces(sched);
+       add_cross_task_wakeups(sched);
 
-       create_tasks();
+       create_tasks(sched);
        printf("------------------------------------------------------------\n");
-       for (i = 0; i < replay_repeat; i++)
-               run_one_test();
+       for (i = 0; i < sched->replay_repeat; i++)
+               run_one_test(sched);
 
        return 0;
 }
 
-
-static const char * const sched_usage[] = {
-       "perf sched [<options>] {record|latency|map|replay|script}",
-       NULL
-};
-
-static const struct option sched_options[] = {
-       OPT_STRING('i', "input", &input_name, "file",
-                   "input file name"),
-       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 * const latency_usage[] = {
-       "perf sched latency [<options>]",
-       NULL
-};
-
-static const struct option latency_options[] = {
-       OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
-                  "sort by key(s): runtime, switch, avg, max"),
-       OPT_INCR('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()
-};
-
-static const char * const replay_usage[] = {
-       "perf sched replay [<options>]",
-       NULL
-};
-
-static const struct option replay_options[] = {
-       OPT_UINTEGER('r', "repeat", &replay_repeat,
-                    "repeat the workload replay N times (-1: infinite)"),
-       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 void setup_sorting(void)
+static void setup_sorting(struct perf_sched *sched, const struct option *options,
+                         const char * const usage_msg[])
 {
-       char *tmp, *tok, *str = strdup(sort_order);
+       char *tmp, *tok, *str = strdup(sched->sort_order);
 
        for (tok = strtok_r(str, ", ", &tmp);
                        tok; tok = strtok_r(NULL, ", ", &tmp)) {
-               if (sort_dimension__add(tok, &sort_list) < 0) {
+               if (sort_dimension__add(tok, &sched->sort_list) < 0) {
                        error("Unknown --sort key: `%s'", tok);
-                       usage_with_options(latency_usage, latency_options);
+                       usage_with_options(usage_msg, options);
                }
        }
 
        free(str);
 
-       sort_dimension__add("pid", &cmp_pid);
+       sort_dimension__add("pid", &sched->cmp_pid);
 }
 
-static const char *record_args[] = {
-       "record",
-       "-a",
-       "-R",
-       "-f",
-       "-m", "1024",
-       "-c", "1",
-       "-e", "sched:sched_switch",
-       "-e", "sched:sched_stat_wait",
-       "-e", "sched:sched_stat_sleep",
-       "-e", "sched:sched_stat_iowait",
-       "-e", "sched:sched_stat_runtime",
-       "-e", "sched:sched_process_exit",
-       "-e", "sched:sched_process_fork",
-       "-e", "sched:sched_wakeup",
-       "-e", "sched:sched_migrate_task",
-};
-
 static int __cmd_record(int argc, const char **argv)
 {
        unsigned int rec_argc, i, j;
        const char **rec_argv;
+       const char * const record_args[] = {
+               "record",
+               "-a",
+               "-R",
+               "-f",
+               "-m", "1024",
+               "-c", "1",
+               "-e", "sched:sched_switch",
+               "-e", "sched:sched_stat_wait",
+               "-e", "sched:sched_stat_sleep",
+               "-e", "sched:sched_stat_iowait",
+               "-e", "sched:sched_stat_runtime",
+               "-e", "sched:sched_process_exit",
+               "-e", "sched:sched_process_fork",
+               "-e", "sched:sched_wakeup",
+               "-e", "sched:sched_migrate_task",
+       };
 
        rec_argc = ARRAY_SIZE(record_args) + argc - 1;
        rec_argv = calloc(rec_argc + 1, sizeof(char *));
@@ -1907,6 +1818,83 @@ static int __cmd_record(int argc, const char **argv)
 
 int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused)
 {
+       const char default_sort_order[] = "avg, max, switch, runtime";
+       struct perf_sched sched = {
+               .tool = {
+                       .sample          = perf_sched__process_tracepoint_sample,
+                       .comm            = perf_event__process_comm,
+                       .lost            = perf_event__process_lost,
+                       .fork            = perf_event__process_task,
+                       .ordered_samples = true,
+               },
+               .cmp_pid              = LIST_HEAD_INIT(sched.cmp_pid),
+               .sort_list            = LIST_HEAD_INIT(sched.sort_list),
+               .start_work_mutex     = PTHREAD_MUTEX_INITIALIZER,
+               .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER,
+               .curr_pid             = { [0 ... MAX_CPUS - 1] = -1 },
+               .sort_order           = default_sort_order,
+               .replay_repeat        = 10,
+               .profile_cpu          = -1,
+               .next_shortname1      = 'A',
+               .next_shortname2      = '0',
+       };
+       const struct option latency_options[] = {
+       OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
+                  "sort by key(s): runtime, switch, avg, max"),
+       OPT_INCR('v', "verbose", &verbose,
+                   "be more verbose (show symbol address, etc)"),
+       OPT_INTEGER('C', "CPU", &sched.profile_cpu,
+                   "CPU to profile on"),
+       OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
+                   "dump raw trace in ASCII"),
+       OPT_END()
+       };
+       const struct option replay_options[] = {
+       OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
+                    "repeat the workload replay N times (-1: infinite)"),
+       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()
+       };
+       const struct option sched_options[] = {
+       OPT_STRING('i', "input", &sched.input_name, "file",
+                   "input file name"),
+       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()
+       };
+       const char * const latency_usage[] = {
+               "perf sched latency [<options>]",
+               NULL
+       };
+       const char * const replay_usage[] = {
+               "perf sched replay [<options>]",
+               NULL
+       };
+       const char * const sched_usage[] = {
+               "perf sched [<options>] {record|latency|map|replay|script}",
+               NULL
+       };
+       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,
+       };
+       struct trace_sched_handler map_ops  = {
+               .switch_event       = map_switch_event,
+       };
+       struct trace_sched_handler replay_ops  = {
+               .wakeup_event       = replay_wakeup_event,
+               .switch_event       = replay_switch_event,
+               .fork_event         = replay_fork_event,
+       };
+
        argc = parse_options(argc, argv, sched_options, sched_usage,
                             PARSE_OPT_STOP_AT_NON_OPTION);
        if (!argc)
@@ -1922,26 +1910,26 @@ int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused)
        if (!strncmp(argv[0], "rec", 3)) {
                return __cmd_record(argc, argv);
        } else if (!strncmp(argv[0], "lat", 3)) {
-               trace_handler = &lat_ops;
+               sched.tp_handler = &lat_ops;
                if (argc > 1) {
                        argc = parse_options(argc, argv, latency_options, latency_usage, 0);
                        if (argc)
                                usage_with_options(latency_usage, latency_options);
                }
-               setup_sorting();
-               return __cmd_lat();
+               setup_sorting(&sched, latency_options, latency_usage);
+               return perf_sched__lat(&sched);
        } else if (!strcmp(argv[0], "map")) {
-               trace_handler = &map_ops;
-               setup_sorting();
-               return __cmd_map();
+               sched.tp_handler = &map_ops;
+               setup_sorting(&sched, latency_options, latency_usage);
+               return perf_sched__map(&sched);
        } else if (!strncmp(argv[0], "rep", 3)) {
-               trace_handler = &replay_ops;
+               sched.tp_handler = &replay_ops;
                if (argc) {
                        argc = parse_options(argc, argv, replay_options, replay_usage, 0);
                        if (argc)
                                usage_with_options(replay_usage, replay_options);
                }
-               return __cmd_replay();
+               return perf_sched__replay(&sched);
        } else {
                usage_with_options(sched_usage, sched_options);
        }