perf: Add the timechart tool
Arjan van de Ven [Sat, 12 Sep 2009 05:53:05 +0000 (07:53 +0200)]
timechart is a tool to visualize what is going on in the system.

The user makes a trace of what is going on with

 > perf record --timechart /usr/bin/some_command

and then can turn the output of this into an svg file

 > perf timechart

which then can be viewed with any SVG view; inkscape works well
enough for me.

The idea behind timechart is to create a "infinitely zoomable"
picture; something that has high level information on a 1:1 zoom
level, but which exposes more details every time you zoom into a
specific area.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20090912130713.6a77bbc0@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tools/perf/Makefile
tools/perf/builtin-timechart.c [new file with mode: 0644]
tools/perf/builtin.h
tools/perf/perf.c

index 0388e36..0aba8b6 100644 (file)
@@ -382,6 +382,7 @@ BUILTIN_OBJS += builtin-list.o
 BUILTIN_OBJS += builtin-record.o
 BUILTIN_OBJS += builtin-report.o
 BUILTIN_OBJS += builtin-stat.o
+BUILTIN_OBJS += builtin-timechart.o
 BUILTIN_OBJS += builtin-top.o
 BUILTIN_OBJS += builtin-trace.o
 
@@ -712,6 +713,12 @@ builtin-help.o: builtin-help.c common-cmds.h PERF-CFLAGS
                '-DPERF_MAN_PATH="$(mandir_SQ)"' \
                '-DPERF_INFO_PATH="$(infodir_SQ)"' $<
 
+builtin-timechart.o: builtin-timechart.c common-cmds.h PERF-CFLAGS
+       $(QUIET_CC)$(CC) -o $*.o -c $(ALL_CFLAGS) \
+               '-DPERF_HTML_PATH="$(htmldir_SQ)"' \
+               '-DPERF_MAN_PATH="$(mandir_SQ)"' \
+               '-DPERF_INFO_PATH="$(infodir_SQ)"' $<
+
 $(BUILT_INS): perf$X
        $(QUIET_BUILT_IN)$(RM) $@ && \
        ln perf$X $@ 2>/dev/null || \
diff --git a/tools/perf/builtin-timechart.c b/tools/perf/builtin-timechart.c
new file mode 100644 (file)
index 0000000..00fac1b
--- /dev/null
@@ -0,0 +1,1120 @@
+/*
+ * builtin-timechart.c - make an svg timechart of system activity
+ *
+ * (C) Copyright 2009 Intel Corporation
+ *
+ * Authors:
+ *     Arjan van de Ven <arjan@linux.intel.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+
+#include "builtin.h"
+
+#include "util/util.h"
+
+#include "util/color.h"
+#include <linux/list.h>
+#include "util/cache.h"
+#include <linux/rbtree.h>
+#include "util/symbol.h"
+#include "util/string.h"
+#include "util/callchain.h"
+#include "util/strlist.h"
+
+#include "perf.h"
+#include "util/header.h"
+#include "util/parse-options.h"
+#include "util/parse-events.h"
+#include "util/svghelper.h"
+
+static char            const *input_name = "perf.data";
+static char            const *output_name = "output.svg";
+
+
+static unsigned long   page_size;
+static unsigned long   mmap_window = 32;
+static u64             sample_type;
+
+static unsigned int    numcpus;
+static u64             min_freq;       /* Lowest CPU frequency seen */
+static u64             max_freq;       /* Highest CPU frequency seen */
+static u64             turbo_frequency;
+
+static u64             first_time, last_time;
+
+
+static struct perf_header      *header;
+
+struct per_pid;
+struct per_pidcomm;
+
+struct cpu_sample;
+struct power_event;
+struct wake_event;
+
+struct sample_wrapper;
+
+/*
+ * Datastructure layout:
+ * We keep an list of "pid"s, matching the kernels notion of a task struct.
+ * Each "pid" entry, has a list of "comm"s.
+ *     this is because we want to track different programs different, while
+ *     exec will reuse the original pid (by design).
+ * Each comm has a list of samples that will be used to draw
+ * final graph.
+ */
+
+struct per_pid {
+       struct per_pid *next;
+
+       int             pid;
+       int             ppid;
+
+       u64             start_time;
+       u64             end_time;
+       u64             total_time;
+       int             display;
+
+       struct per_pidcomm *all;
+       struct per_pidcomm *current;
+
+       int painted;
+};
+
+
+struct per_pidcomm {
+       struct per_pidcomm *next;
+
+       u64             start_time;
+       u64             end_time;
+       u64             total_time;
+
+       int             Y;
+       int             display;
+
+       long            state;
+       u64             state_since;
+
+       char            *comm;
+
+       struct cpu_sample *samples;
+};
+
+struct sample_wrapper {
+       struct sample_wrapper *next;
+
+       u64             timestamp;
+       unsigned char   data[0];
+};
+
+#define TYPE_NONE      0
+#define TYPE_RUNNING   1
+#define TYPE_WAITING   2
+#define TYPE_BLOCKED   3
+
+struct cpu_sample {
+       struct cpu_sample *next;
+
+       u64 start_time;
+       u64 end_time;
+       int type;
+       int cpu;
+};
+
+static struct per_pid *all_data;
+
+#define CSTATE 1
+#define PSTATE 2
+
+struct power_event {
+       struct power_event *next;
+       int type;
+       int state;
+       u64 start_time;
+       u64 end_time;
+       int cpu;
+};
+
+struct wake_event {
+       struct wake_event *next;
+       int waker;
+       int wakee;
+       u64 time;
+};
+
+static struct power_event    *power_events;
+static struct wake_event     *wake_events;
+
+struct sample_wrapper *all_samples;
+
+static struct per_pid *find_create_pid(int pid)
+{
+       struct per_pid *cursor = all_data;
+
+       while (cursor) {
+               if (cursor->pid == pid)
+                       return cursor;
+               cursor = cursor->next;
+       }
+       cursor = malloc(sizeof(struct per_pid));
+       assert(cursor != NULL);
+       memset(cursor, 0, sizeof(struct per_pid));
+       cursor->pid = pid;
+       cursor->next = all_data;
+       all_data = cursor;
+       return cursor;
+}
+
+static void pid_set_comm(int pid, char *comm)
+{
+       struct per_pid *p;
+       struct per_pidcomm *c;
+       p = find_create_pid(pid);
+       c = p->all;
+       while (c) {
+               if (c->comm && strcmp(c->comm, comm) == 0) {
+                       p->current = c;
+                       return;
+               }
+               if (!c->comm) {
+                       c->comm = strdup(comm);
+                       p->current = c;
+                       return;
+               }
+               c = c->next;
+       }
+       c = malloc(sizeof(struct per_pidcomm));
+       assert(c != NULL);
+       memset(c, 0, sizeof(struct per_pidcomm));
+       c->comm = strdup(comm);
+       p->current = c;
+       c->next = p->all;
+       p->all = c;
+}
+
+static void pid_fork(int pid, int ppid, u64 timestamp)
+{
+       struct per_pid *p, *pp;
+       p = find_create_pid(pid);
+       pp = find_create_pid(ppid);
+       p->ppid = ppid;
+       if (pp->current && pp->current->comm && !p->current)
+               pid_set_comm(pid, pp->current->comm);
+
+       p->start_time = timestamp;
+       if (p->current) {
+               p->current->start_time = timestamp;
+               p->current->state_since = timestamp;
+       }
+}
+
+static void pid_exit(int pid, u64 timestamp)
+{
+       struct per_pid *p;
+       p = find_create_pid(pid);
+       p->end_time = timestamp;
+       if (p->current)
+               p->current->end_time = timestamp;
+}
+
+static void
+pid_put_sample(int pid, int type, unsigned int cpu, u64 start, u64 end)
+{
+       struct per_pid *p;
+       struct per_pidcomm *c;
+       struct cpu_sample *sample;
+
+       p = find_create_pid(pid);
+       c = p->current;
+       if (!c) {
+               c = malloc(sizeof(struct per_pidcomm));
+               assert(c != NULL);
+               memset(c, 0, sizeof(struct per_pidcomm));
+               p->current = c;
+               c->next = p->all;
+               p->all = c;
+       }
+
+       sample = malloc(sizeof(struct cpu_sample));
+       assert(sample != NULL);
+       memset(sample, 0, sizeof(struct cpu_sample));
+       sample->start_time = start;
+       sample->end_time = end;
+       sample->type = type;
+       sample->next = c->samples;
+       sample->cpu = cpu;
+       c->samples = sample;
+
+       if (sample->type == TYPE_RUNNING && end > start && start > 0) {
+               c->total_time += (end-start);
+               p->total_time += (end-start);
+       }
+
+       if (c->start_time == 0 || c->start_time > start)
+               c->start_time = start;
+       if (p->start_time == 0 || p->start_time > start)
+               p->start_time = start;
+
+       if (cpu > numcpus)
+               numcpus = cpu;
+}
+
+#define MAX_CPUS 4096
+
+static u64 cpus_cstate_start_times[MAX_CPUS];
+static int cpus_cstate_state[MAX_CPUS];
+static u64 cpus_pstate_start_times[MAX_CPUS];
+static u64 cpus_pstate_state[MAX_CPUS];
+
+static int
+process_comm_event(event_t *event)
+{
+       pid_set_comm(event->comm.pid, event->comm.comm);
+       return 0;
+}
+static int
+process_fork_event(event_t *event)
+{
+       pid_fork(event->fork.pid, event->fork.ppid, event->fork.time);
+       return 0;
+}
+
+static int
+process_exit_event(event_t *event)
+{
+       pid_exit(event->fork.pid, event->fork.time);
+       return 0;
+}
+
+struct trace_entry {
+       u32                     size;
+       unsigned short          type;
+       unsigned char           flags;
+       unsigned char           preempt_count;
+       int                     pid;
+       int                     tgid;
+};
+
+struct power_entry {
+       struct trace_entry te;
+       s64     type;
+       s64     value;
+};
+
+#define TASK_COMM_LEN 16
+struct wakeup_entry {
+       struct trace_entry te;
+       char comm[TASK_COMM_LEN];
+       int   pid;
+       int   prio;
+       int   success;
+};
+
+/*
+ * trace_flag_type is an enumeration that holds different
+ * states when a trace occurs. These are:
+ *  IRQS_OFF            - interrupts were disabled
+ *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
+ *  NEED_RESCED         - reschedule is requested
+ *  HARDIRQ             - inside an interrupt handler
+ *  SOFTIRQ             - inside a softirq handler
+ */
+enum trace_flag_type {
+       TRACE_FLAG_IRQS_OFF             = 0x01,
+       TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
+       TRACE_FLAG_NEED_RESCHED         = 0x04,
+       TRACE_FLAG_HARDIRQ              = 0x08,
+       TRACE_FLAG_SOFTIRQ              = 0x10,
+};
+
+
+
+struct sched_switch {
+       struct trace_entry te;
+       char prev_comm[TASK_COMM_LEN];
+       int  prev_pid;
+       int  prev_prio;
+       long prev_state; /* Arjan weeps. */
+       char next_comm[TASK_COMM_LEN];
+       int  next_pid;
+       int  next_prio;
+};
+
+static void c_state_start(int cpu, u64 timestamp, int state)
+{
+       cpus_cstate_start_times[cpu] = timestamp;
+       cpus_cstate_state[cpu] = state;
+}
+
+static void c_state_end(int cpu, u64 timestamp)
+{
+       struct power_event *pwr;
+       pwr = malloc(sizeof(struct power_event));
+       if (!pwr)
+               return;
+       memset(pwr, 0, sizeof(struct power_event));
+
+       pwr->state = cpus_cstate_state[cpu];
+       pwr->start_time = cpus_cstate_start_times[cpu];
+       pwr->end_time = timestamp;
+       pwr->cpu = cpu;
+       pwr->type = CSTATE;
+       pwr->next = power_events;
+
+       power_events = pwr;
+}
+
+static void p_state_change(int cpu, u64 timestamp, u64 new_freq)
+{
+       struct power_event *pwr;
+       pwr = malloc(sizeof(struct power_event));
+
+       if (new_freq > 8000000) /* detect invalid data */
+               return;
+
+       if (!pwr)
+               return;
+       memset(pwr, 0, sizeof(struct power_event));
+
+       pwr->state = cpus_pstate_state[cpu];
+       pwr->start_time = cpus_pstate_start_times[cpu];
+       pwr->end_time = timestamp;
+       pwr->cpu = cpu;
+       pwr->type = PSTATE;
+       pwr->next = power_events;
+
+       if (!pwr->start_time)
+               pwr->start_time = first_time;
+
+       power_events = pwr;
+
+       cpus_pstate_state[cpu] = new_freq;
+       cpus_pstate_start_times[cpu] = timestamp;
+
+       if ((u64)new_freq > max_freq)
+               max_freq = new_freq;
+
+       if (new_freq < min_freq || min_freq == 0)
+               min_freq = new_freq;
+
+       if (new_freq == max_freq - 1000)
+                       turbo_frequency = max_freq;
+}
+
+static void
+sched_wakeup(int cpu, u64 timestamp, int pid, struct trace_entry *te)
+{
+       struct wake_event *we;
+       struct per_pid *p;
+       struct wakeup_entry *wake = (void *)te;
+
+       we = malloc(sizeof(struct wake_event));
+       if (!we)
+               return;
+
+       memset(we, 0, sizeof(struct wake_event));
+       we->time = timestamp;
+       we->waker = pid;
+
+       if ((te->flags & TRACE_FLAG_HARDIRQ) || (te->flags & TRACE_FLAG_SOFTIRQ))
+               we->waker = -1;
+
+       we->wakee = wake->pid;
+       we->next = wake_events;
+       wake_events = we;
+       p = find_create_pid(we->wakee);
+
+       if (p && p->current && p->current->state == TYPE_NONE) {
+               p->current->state_since = timestamp;
+               p->current->state = TYPE_WAITING;
+       }
+       if (p && p->current && p->current->state == TYPE_BLOCKED) {
+               pid_put_sample(p->pid, p->current->state, cpu, p->current->state_since, timestamp);
+               p->current->state_since = timestamp;
+               p->current->state = TYPE_WAITING;
+       }
+}
+
+static void sched_switch(int cpu, u64 timestamp, struct trace_entry *te)
+{
+       struct per_pid *p = NULL, *prev_p;
+       struct sched_switch *sw = (void *)te;
+
+
+       prev_p = find_create_pid(sw->prev_pid);
+
+       p = find_create_pid(sw->next_pid);
+
+       if (prev_p->current && prev_p->current->state != TYPE_NONE)
+               pid_put_sample(sw->prev_pid, TYPE_RUNNING, cpu, prev_p->current->state_since, timestamp);
+       if (p && p->current) {
+               if (p->current->state != TYPE_NONE)
+                       pid_put_sample(sw->next_pid, p->current->state, cpu, p->current->state_since, timestamp);
+
+                       p->current->state_since = timestamp;
+                       p->current->state = TYPE_RUNNING;
+       }
+
+       if (prev_p->current) {
+               prev_p->current->state = TYPE_NONE;
+               prev_p->current->state_since = timestamp;
+               if (sw->prev_state & 2)
+                       prev_p->current->state = TYPE_BLOCKED;
+               if (sw->prev_state == 0)
+                       prev_p->current->state = TYPE_WAITING;
+       }
+}
+
+
+static int
+process_sample_event(event_t *event)
+{
+       int cursor = 0;
+       u64 addr = 0;
+       u64 stamp = 0;
+       u32 cpu = 0;
+       u32 pid = 0;
+       struct trace_entry *te;
+
+       if (sample_type & PERF_SAMPLE_IP)
+               cursor++;
+
+       if (sample_type & PERF_SAMPLE_TID) {
+               pid = event->sample.array[cursor]>>32;
+               cursor++;
+       }
+       if (sample_type & PERF_SAMPLE_TIME) {
+               stamp = event->sample.array[cursor++];
+
+               if (!first_time || first_time > stamp)
+                       first_time = stamp;
+               if (last_time < stamp)
+                       last_time = stamp;
+
+       }
+       if (sample_type & PERF_SAMPLE_ADDR)
+               addr = event->sample.array[cursor++];
+       if (sample_type & PERF_SAMPLE_ID)
+               cursor++;
+       if (sample_type & PERF_SAMPLE_STREAM_ID)
+               cursor++;
+       if (sample_type & PERF_SAMPLE_CPU)
+               cpu = event->sample.array[cursor++] & 0xFFFFFFFF;
+       if (sample_type & PERF_SAMPLE_PERIOD)
+               cursor++;
+
+       te = (void *)&event->sample.array[cursor];
+
+       if (sample_type & PERF_SAMPLE_RAW && te->size > 0) {
+               char *event_str;
+               struct power_entry *pe;
+
+               pe = (void *)te;
+
+               event_str = perf_header__find_event(te->type);
+
+               if (!event_str)
+                       return 0;
+
+               if (strcmp(event_str, "power:power_start") == 0)
+                       c_state_start(cpu, stamp, pe->value);
+
+               if (strcmp(event_str, "power:power_end") == 0)
+                       c_state_end(cpu, stamp);
+
+               if (strcmp(event_str, "power:power_frequency") == 0)
+                       p_state_change(cpu, stamp, pe->value);
+
+               if (strcmp(event_str, "sched:sched_wakeup") == 0)
+                       sched_wakeup(cpu, stamp, pid, te);
+
+               if (strcmp(event_str, "sched:sched_switch") == 0)
+                       sched_switch(cpu, stamp, te);
+       }
+       return 0;
+}
+
+/*
+ * After the last sample we need to wrap up the current C/P state
+ * and close out each CPU for these.
+ */
+static void end_sample_processing(void)
+{
+       u64 cpu;
+       struct power_event *pwr;
+
+       for (cpu = 0; cpu < numcpus; cpu++) {
+               pwr = malloc(sizeof(struct power_event));
+               if (!pwr)
+                       return;
+               memset(pwr, 0, sizeof(struct power_event));
+
+               /* C state */
+#if 0
+               pwr->state = cpus_cstate_state[cpu];
+               pwr->start_time = cpus_cstate_start_times[cpu];
+               pwr->end_time = last_time;
+               pwr->cpu = cpu;
+               pwr->type = CSTATE;
+               pwr->next = power_events;
+
+               power_events = pwr;
+#endif
+               /* P state */
+
+               pwr = malloc(sizeof(struct power_event));
+               if (!pwr)
+                       return;
+               memset(pwr, 0, sizeof(struct power_event));
+
+               pwr->state = cpus_pstate_state[cpu];
+               pwr->start_time = cpus_pstate_start_times[cpu];
+               pwr->end_time = last_time;
+               pwr->cpu = cpu;
+               pwr->type = PSTATE;
+               pwr->next = power_events;
+
+               if (!pwr->start_time)
+                       pwr->start_time = first_time;
+               if (!pwr->state)
+                       pwr->state = min_freq;
+               power_events = pwr;
+       }
+}
+
+static u64 sample_time(event_t *event)
+{
+       int cursor;
+
+       cursor = 0;
+       if (sample_type & PERF_SAMPLE_IP)
+               cursor++;
+       if (sample_type & PERF_SAMPLE_TID)
+               cursor++;
+       if (sample_type & PERF_SAMPLE_TIME)
+               return event->sample.array[cursor];
+       return 0;
+}
+
+
+/*
+ * We first queue all events, sorted backwards by insertion.
+ * The order will get flipped later.
+ */
+static int
+queue_sample_event(event_t *event)
+{
+       struct sample_wrapper *copy, *prev;
+       int size;
+
+       size = event->sample.header.size + sizeof(struct sample_wrapper) + 8;
+
+       copy = malloc(size);
+       if (!copy)
+               return 1;
+
+       memset(copy, 0, size);
+
+       copy->next = NULL;
+       copy->timestamp = sample_time(event);
+
+       memcpy(&copy->data, event, event->sample.header.size);
+
+       /* insert in the right place in the list */
+
+       if (!all_samples) {
+               /* first sample ever */
+               all_samples = copy;
+               return 0;
+       }
+
+       if (all_samples->timestamp < copy->timestamp) {
+               /* insert at the head of the list */
+               copy->next = all_samples;
+               all_samples = copy;
+               return 0;
+       }
+
+       prev = all_samples;
+       while (prev->next) {
+               if (prev->next->timestamp < copy->timestamp) {
+                       copy->next = prev->next;
+                       prev->next = copy;
+                       return 0;
+               }
+               prev = prev->next;
+       }
+       /* insert at the end of the list */
+       prev->next = copy;
+
+       return 0;
+}
+
+static void sort_queued_samples(void)
+{
+       struct sample_wrapper *cursor, *next;
+
+       cursor = all_samples;
+       all_samples = NULL;
+
+       while (cursor) {
+               next = cursor->next;
+               cursor->next = all_samples;
+               all_samples = cursor;
+               cursor = next;
+       }
+}
+
+/*
+ * Sort the pid datastructure
+ */
+static void sort_pids(void)
+{
+       struct per_pid *new_list, *p, *cursor, *prev;
+       /* sort by ppid first, then by pid, lowest to highest */
+
+       new_list = NULL;
+
+       while (all_data) {
+               p = all_data;
+               all_data = p->next;
+               p->next = NULL;
+
+               if (new_list == NULL) {
+                       new_list = p;
+                       p->next = NULL;
+                       continue;
+               }
+               prev = NULL;
+               cursor = new_list;
+               while (cursor) {
+                       if (cursor->ppid > p->ppid ||
+                               (cursor->ppid == p->ppid && cursor->pid > p->pid)) {
+                               /* must insert before */
+                               if (prev) {
+                                       p->next = prev->next;
+                                       prev->next = p;
+                                       cursor = NULL;
+                                       continue;
+                               } else {
+                                       p->next = new_list;
+                                       new_list = p;
+                                       cursor = NULL;
+                                       continue;
+                               }
+                       }
+
+                       prev = cursor;
+                       cursor = cursor->next;
+                       if (!cursor)
+                               prev->next = p;
+               }
+       }
+       all_data = new_list;
+}
+
+
+static void draw_c_p_states(void)
+{
+       struct power_event *pwr;
+       pwr = power_events;
+
+       /*
+        * two pass drawing so that the P state bars are on top of the C state blocks
+        */
+       while (pwr) {
+               if (pwr->type == CSTATE)
+                       svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
+               pwr = pwr->next;
+       }
+
+       pwr = power_events;
+       while (pwr) {
+               if (pwr->type == PSTATE) {
+                       if (!pwr->state)
+                               pwr->state = min_freq;
+                       svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
+               }
+               pwr = pwr->next;
+       }
+}
+
+static void draw_wakeups(void)
+{
+       struct wake_event *we;
+       struct per_pid *p;
+       struct per_pidcomm *c;
+
+       we = wake_events;
+       while (we) {
+               int from = 0, to = 0;
+
+               /* locate the column of the waker and wakee */
+               p = all_data;
+               while (p) {
+                       if (p->pid == we->waker || p->pid == we->wakee) {
+                               c = p->all;
+                               while (c) {
+                                       if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
+                                               if (p->pid == we->waker)
+                                                       from = c->Y;
+                                               if (p->pid == we->wakee)
+                                                       to = c->Y;
+                                       }
+                                       c = c->next;
+                               }
+                       }
+                       p = p->next;
+               }
+
+               if (we->waker == -1)
+                       svg_interrupt(we->time, to);
+               else if (from && to && abs(from - to) == 1)
+                       svg_wakeline(we->time, from, to);
+               else
+                       svg_partial_wakeline(we->time, from, to);
+               we = we->next;
+       }
+}
+
+static void draw_cpu_usage(void)
+{
+       struct per_pid *p;
+       struct per_pidcomm *c;
+       struct cpu_sample *sample;
+       p = all_data;
+       while (p) {
+               c = p->all;
+               while (c) {
+                       sample = c->samples;
+                       while (sample) {
+                               if (sample->type == TYPE_RUNNING)
+                                       svg_process(sample->cpu, sample->start_time, sample->end_time, "sample", c->comm);
+
+                               sample = sample->next;
+                       }
+                       c = c->next;
+               }
+               p = p->next;
+       }
+}
+
+static void draw_process_bars(void)
+{
+       struct per_pid *p;
+       struct per_pidcomm *c;
+       struct cpu_sample *sample;
+       int Y = 0;
+
+       Y = 2 * numcpus + 2;
+
+       p = all_data;
+       while (p) {
+               c = p->all;
+               while (c) {
+                       if (!c->display) {
+                               c->Y = 0;
+                               c = c->next;
+                               continue;
+                       }
+
+                       svg_box(Y, p->start_time, p->end_time, "process");
+                       sample = c->samples;
+                       while (sample) {
+                               if (sample->type == TYPE_RUNNING)
+                                       svg_sample(Y, sample->cpu, sample->start_time, sample->end_time, "sample");
+                               if (sample->type == TYPE_BLOCKED)
+                                       svg_box(Y, sample->start_time, sample->end_time, "blocked");
+                               if (sample->type == TYPE_WAITING)
+                                       svg_box(Y, sample->start_time, sample->end_time, "waiting");
+                               sample = sample->next;
+                       }
+
+                       if (c->comm) {
+                               char comm[256];
+                               if (c->total_time > 5000000000) /* 5 seconds */
+                                       sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / 1000000000.0);
+                               else
+                                       sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / 1000000.0);
+
+                               svg_text(Y, c->start_time, comm);
+                       }
+                       c->Y = Y;
+                       Y++;
+                       c = c->next;
+               }
+               p = p->next;
+       }
+}
+
+static int determine_display_tasks(u64 threshold)
+{
+       struct per_pid *p;
+       struct per_pidcomm *c;
+       int count = 0;
+
+       p = all_data;
+       while (p) {
+               p->display = 0;
+               if (p->start_time == 1)
+                       p->start_time = first_time;
+
+               /* no exit marker, task kept running to the end */
+               if (p->end_time == 0)
+                       p->end_time = last_time;
+               if (p->total_time >= threshold)
+                       p->display = 1;
+
+               c = p->all;
+
+               while (c) {
+                       c->display = 0;
+
+                       if (c->start_time == 1)
+                               c->start_time = first_time;
+
+                       if (c->total_time >= threshold) {
+                               c->display = 1;
+                               count++;
+                       }
+
+                       if (c->end_time == 0)
+                               c->end_time = last_time;
+
+                       c = c->next;
+               }
+               p = p->next;
+       }
+       return count;
+}
+
+
+
+#define TIME_THRESH 10000000
+
+static void write_svg_file(const char *filename)
+{
+       u64 i;
+       int count;
+
+       numcpus++;
+
+
+       count = determine_display_tasks(TIME_THRESH);
+
+       /* We'd like to show at least 15 tasks; be less picky if we have fewer */
+       if (count < 15)
+               count = determine_display_tasks(TIME_THRESH / 10);
+
+       open_svg(filename, numcpus, count);
+
+       svg_time_grid(first_time, last_time);
+       svg_legenda();
+
+       for (i = 0; i < numcpus; i++)
+               svg_cpu_box(i, max_freq, turbo_frequency);
+
+       draw_cpu_usage();
+       draw_process_bars();
+       draw_c_p_states();
+       draw_wakeups();
+
+       svg_close();
+}
+
+static int
+process_event(event_t *event)
+{
+
+       switch (event->header.type) {
+
+       case PERF_EVENT_COMM:
+               return process_comm_event(event);
+       case PERF_EVENT_FORK:
+               return process_fork_event(event);
+       case PERF_EVENT_EXIT:
+               return process_exit_event(event);
+       case PERF_EVENT_SAMPLE:
+               return queue_sample_event(event);
+
+       /*
+        * We dont process them right now but they are fine:
+        */
+       case PERF_EVENT_MMAP:
+       case PERF_EVENT_THROTTLE:
+       case PERF_EVENT_UNTHROTTLE:
+               return 0;
+
+       default:
+               return -1;
+       }
+
+       return 0;
+}
+
+static void process_samples(void)
+{
+       struct sample_wrapper *cursor;
+       event_t *event;
+
+       sort_queued_samples();
+
+       cursor = all_samples;
+       while (cursor) {
+               event = (void *)&cursor->data;
+               cursor = cursor->next;
+               process_sample_event(event);
+       }
+}
+
+
+static int __cmd_timechart(void)
+{
+       int ret, rc = EXIT_FAILURE;
+       unsigned long offset = 0;
+       unsigned long head, shift;
+       struct stat statbuf;
+       event_t *event;
+       uint32_t size;
+       char *buf;
+       int input;
+
+       input = open(input_name, O_RDONLY);
+       if (input < 0) {
+               fprintf(stderr, " failed to open file: %s", input_name);
+               if (!strcmp(input_name, "perf.data"))
+                       fprintf(stderr, "  (try 'perf record' first)");
+               fprintf(stderr, "\n");
+               exit(-1);
+       }
+
+       ret = fstat(input, &statbuf);
+       if (ret < 0) {
+               perror("failed to stat file");
+               exit(-1);
+       }
+
+       if (!statbuf.st_size) {
+               fprintf(stderr, "zero-sized file, nothing to do!\n");
+               exit(0);
+       }
+
+       header = perf_header__read(input);
+       head = header->data_offset;
+
+       sample_type = perf_header__sample_type(header);
+
+       shift = page_size * (head / page_size);
+       offset += shift;
+       head -= shift;
+
+remap:
+       buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
+                          MAP_SHARED, input, offset);
+       if (buf == MAP_FAILED) {
+               perror("failed to mmap file");
+               exit(-1);
+       }
+
+more:
+       event = (event_t *)(buf + head);
+
+       size = event->header.size;
+       if (!size)
+               size = 8;
+
+       if (head + event->header.size >= page_size * mmap_window) {
+               int ret2;
+
+               shift = page_size * (head / page_size);
+
+               ret2 = munmap(buf, page_size * mmap_window);
+               assert(ret2 == 0);
+
+               offset += shift;
+               head -= shift;
+               goto remap;
+       }
+
+       size = event->header.size;
+
+       if (!size || process_event(event) < 0) {
+
+               printf("%p [%p]: skipping unknown header type: %d\n",
+                       (void *)(offset + head),
+                       (void *)(long)(event->header.size),
+                       event->header.type);
+
+               /*
+                * assume we lost track of the stream, check alignment, and
+                * increment a single u64 in the hope to catch on again 'soon'.
+                */
+
+               if (unlikely(head & 7))
+                       head &= ~7ULL;
+
+               size = 8;
+       }
+
+       head += size;
+
+       if (offset + head >= header->data_offset + header->data_size)
+               goto done;
+
+       if (offset + head < (unsigned long)statbuf.st_size)
+               goto more;
+
+done:
+       rc = EXIT_SUCCESS;
+       close(input);
+
+
+       process_samples();
+
+       end_sample_processing();
+
+       sort_pids();
+
+       write_svg_file(output_name);
+
+       printf("Written %2.1f seconds of trace to %s.\n", (last_time - first_time) / 1000000000.0, output_name);
+
+       return rc;
+}
+
+static const char * const report_usage[] = {
+       "perf report [<options>] <command>",
+       NULL
+};
+
+static const struct option options[] = {
+       OPT_STRING('i', "input", &input_name, "file",
+                   "input file name"),
+       OPT_STRING('o', "output", &output_name, "file",
+                   "output file name"),
+       OPT_END()
+};
+
+
+int cmd_timechart(int argc, const char **argv, const char *prefix __used)
+{
+       symbol__init();
+
+       page_size = getpagesize();
+
+       argc = parse_options(argc, argv, options, report_usage, 0);
+
+       /*
+        * Any (unrecognized) arguments left?
+        */
+       if (argc)
+               usage_with_options(report_usage, options);
+
+       setup_pager();
+
+       return __cmd_timechart();
+}
index b09cadb..e11d8d2 100644 (file)
@@ -21,6 +21,7 @@ extern int cmd_list(int argc, const char **argv, const char *prefix);
 extern int cmd_record(int argc, const char **argv, const char *prefix);
 extern int cmd_report(int argc, const char **argv, const char *prefix);
 extern int cmd_stat(int argc, const char **argv, const char *prefix);
+extern int cmd_timechart(int argc, const char **argv, const char *prefix);
 extern int cmd_top(int argc, const char **argv, const char *prefix);
 extern int cmd_trace(int argc, const char **argv, const char *prefix);
 extern int cmd_version(int argc, const char **argv, const char *prefix);
index c972d1c..19fc7fe 100644 (file)
@@ -289,6 +289,7 @@ static void handle_internal_command(int argc, const char **argv)
                { "record", cmd_record, 0 },
                { "report", cmd_report, 0 },
                { "stat", cmd_stat, 0 },
+               { "timechart", cmd_timechart, 0 },
                { "top", cmd_top, 0 },
                { "annotate", cmd_annotate, 0 },
                { "version", cmd_version, 0 },