[PATCH 1/8] perf: Add a timestamp to fork events

[PATCH 1/8] perf: Add a timestamp to fork events

[Arjan van de Ven]

>From 52d6521b9c81230e3a318dfc741274d6601374ca Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:52:47 +0200
Subject: [PATCH] perf: Add a timestamp to fork events

timechart needs to know when a process forked, in order to be able
to visualize properly when tasks start.

This patch adds a time field to the event structure, and fills it
in appropriately.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 include/linux/perf_counter.h |    2 ++
 kernel/perf_counter.c        |   11 +++++++++--
 tools/perf/util/event.h      |    1 +
 3 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 972f90d..5fe6898 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -332,6 +332,7 @@ enum perf_event_type {
 	 *	struct perf_event_header	header;
 	 *	u32				pid, ppid;
 	 *	u32				tid, ptid;
+	 *	u64				time;
 	 * };
 	 */
 	PERF_EVENT_EXIT			= 4,
@@ -352,6 +353,7 @@ enum perf_event_type {
 	 *	struct perf_event_header	header;
 	 *	u32				pid, ppid;
 	 *	u32				tid, ptid;
+	 *	{ u64				time;     } && PERF_SAMPLE_TIME
 	 * };
 	 */
 	PERF_EVENT_FORK			= 7,
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index e0d91fd..25490db 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -3071,6 +3071,7 @@ struct perf_task_event {
 		u32				ppid;
 		u32				tid;
 		u32				ptid;
+		u64				time;
 	} event;
 };
 
@@ -3078,9 +3079,12 @@ static void perf_counter_task_output(struct perf_counter *counter,
 				     struct perf_task_event *task_event)
 {
 	struct perf_output_handle handle;
-	int size = task_event->event.header.size;
+	int size;
 	struct task_struct *task = task_event->task;
-	int ret = perf_output_begin(&handle, counter, size, 0, 0);
+	int ret;
+
+	size  = task_event->event.header.size;
+	ret = perf_output_begin(&handle, counter, size, 0, 0);
 
 	if (ret)
 		return;
@@ -3091,7 +3095,10 @@ static void perf_counter_task_output(struct perf_counter *counter,
 	task_event->event.tid = perf_counter_tid(counter, task);
 	task_event->event.ptid = perf_counter_tid(counter, current);
 
+	task_event->event.time = sched_clock();
+
 	perf_output_put(&handle, task_event->event);
+
 	perf_output_end(&handle);
 }
 
diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index fa2d4e9..1545349 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -39,6 +39,7 @@ struct fork_event {
 	struct perf_event_header header;
 	u32 pid, ppid;
 	u32 tid, ptid;
+	u64 time;
 };
 
 struct lost_event {
-- 
1.6.0.6



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 2/8] perf: Store trace event name/id pairs in perf.data

[Arjan van de Ven]

>From c9d9dfc617df7a6a6029feb486f2433b7d825ef4 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:52:51 +0200
Subject: [PATCH] perf: Store trace event name/id pairs in perf.data

The trace event name<->id mapping is dynamic for each kernel compile.
In order for perf.data to be useable outside the actual system, we thus
need to store a table of this mapping for later use.

This patch adds this table to perf.data, and provides helper functions
for lookup up fields from this table.

To avoid mistakes, lookup-from-table is kept completely seprate from
lookup-from-local-debugfs.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/util/header.c       |   59 ++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/header.h       |    6 ++++
 tools/perf/util/parse-events.c |   25 +++++++++++++++++
 3 files changed, 90 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index ec4d4c2..ef91145 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -86,6 +86,44 @@ void perf_header__add_attr(struct perf_header *self,
 	self->attr[pos] = attr;
 }
 
+struct perf_trace_event_type {
+	u64	event_id;
+	char	name[64];
+};
+
+static int event_count;
+static struct perf_trace_event_type *events;
+
+void perf_header__push_event(u64 id, const char *name)
+{
+	if (strlen(name) > 64)
+		printf("Event %s will be truncated\n", name);
+
+	if (!events) {
+		events = malloc(sizeof(struct perf_trace_event_type));
+		if (!events)
+			die("nomem");
+	} else {
+		events = realloc(events, (event_count + 1) * sizeof(struct perf_trace_event_type));
+		if (!events)
+			die("nomem");
+	}
+	memset(&events[event_count], 0, sizeof(struct perf_trace_event_type));
+	events[event_count].event_id = id;
+	strncpy(events[event_count].name, name, 63);
+	event_count++;
+}
+
+char *perf_header__find_event(u64 id)
+{
+	int i;
+	for (i = 0 ; i < event_count; i++) {
+		if (events[i].event_id == id)
+			return events[i].name;
+	}
+	return NULL;
+}
+
 static const char *__perf_magic = "PERFFILE";
 
 #define PERF_MAGIC	(*(u64 *)__perf_magic)
@@ -106,6 +144,7 @@ struct perf_file_header {
 	u64				attr_size;
 	struct perf_file_section	attrs;
 	struct perf_file_section	data;
+	struct perf_file_section	event_types;
 };
 
 static void do_write(int fd, void *buf, size_t size)
@@ -154,6 +193,11 @@ void perf_header__write(struct perf_header *self, int fd)
 		do_write(fd, &f_attr, sizeof(f_attr));
 	}
 
+	self->event_offset = lseek(fd, 0, SEEK_CUR);
+	self->event_size = event_count * sizeof(struct perf_trace_event_type);
+	if (events)
+		do_write(fd, events, self->event_size);
+
 
 	self->data_offset = lseek(fd, 0, SEEK_CUR);
 
@@ -169,6 +213,10 @@ void perf_header__write(struct perf_header *self, int fd)
 			.offset = self->data_offset,
 			.size	= self->data_size,
 		},
+		.event_types = {
+			.offset = self->event_offset,
+			.size	= self->event_size,
+		},
 	};
 
 	lseek(fd, 0, SEEK_SET);
@@ -234,6 +282,17 @@ struct perf_header *perf_header__read(int fd)
 		lseek(fd, tmp, SEEK_SET);
 	}
 
+	if (f_header.event_types.size) {
+		lseek(fd, f_header.event_types.offset, SEEK_SET);
+		events = malloc(f_header.event_types.size);
+		if (!events)
+			die("nomem");
+		do_read(fd, events, f_header.event_types.size);
+		event_count =  f_header.event_types.size / sizeof(struct perf_trace_event_type);
+	}
+	self->event_offset = f_header.event_types.offset;
+	self->event_size   = f_header.event_types.size;
+
 	self->data_offset = f_header.data.offset;
 	self->data_size   = f_header.data.size;
 
diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
index 5d0a72e..7b0e84a 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -19,6 +19,8 @@ struct perf_header {
 	s64 attr_offset;
 	u64 data_offset;
 	u64 data_size;
+	u64 event_offset;
+	u64 event_size;
 };
 
 struct perf_header *perf_header__read(int fd);
@@ -27,6 +29,10 @@ void perf_header__write(struct perf_header *self, int fd);
 void perf_header__add_attr(struct perf_header *self,
 			   struct perf_header_attr *attr);
 
+void perf_header__push_event(u64 id, const char *name);
+char *perf_header__find_event(u64 id);
+
+
 struct perf_header_attr *
 perf_header_attr__new(struct perf_counter_attr *attr);
 void perf_header_attr__add_id(struct perf_header_attr *self, u64 id);
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index a587d41..37f1357 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -6,6 +6,7 @@
 #include "exec_cmd.h"
 #include "string.h"
 #include "cache.h"
+#include "header.h"
 
 int					nr_counters;
 
@@ -583,10 +584,34 @@ static int parse_event_symbols(const char **str, struct perf_counter_attr *attr)
 	return 1;
 }
 
+static void store_event_type(const char *orgname)
+{
+	char filename[PATH_MAX], *c;
+	FILE *file;
+	int id;
+
+	sprintf(filename, "/sys/kernel/debug/tracing/events/%s/id", orgname);
+	c = strchr(filename, ':');
+	if (c)
+		*c = '/';
+
+	file = fopen(filename, "r");
+	if (!file)
+		return;
+	if (fscanf(file, "%i", &id) < 1)
+		die("cannot store event ID");
+	fclose(file);
+	perf_header__push_event(id, orgname);
+}
+
+
 int parse_events(const struct option *opt __used, const char *str, int unset __used)
 {
 	struct perf_counter_attr attr;
 
+	if (strchr(str, ':'))
+		store_event_type(str);
+
 	for (;;) {
 		if (nr_counters == MAX_COUNTERS)
 			return -1;
-- 
1.6.0.6




-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 3/8] perf: Allow perf utilities to have "callback" options without arguments

[Arjan van de Ven]

>From b37c714203e12c2104b2fae41c6f4ea4fa589528 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:52:54 +0200
Subject: [PATCH] perf: Allow perf utilities to have "callback" options without arguments

timechart needs to add a "callback" type command line argument that does
not take arguments.

This patch adds the parse-options.h infrastructure to make this possible.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/util/parse-options.h |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/parse-options.h b/tools/perf/util/parse-options.h
index 8aa3464..2ee248f 100644
--- a/tools/perf/util/parse-options.h
+++ b/tools/perf/util/parse-options.h
@@ -104,6 +104,8 @@ struct option {
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), .argh = "time", .help = (h), .callback = parse_opt_approxidate_cb }
 #define OPT_CALLBACK(s, l, v, a, h, f) \
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f) }
+#define OPT_CALLBACK_NOOPT(s, l, v, a, h, f) \
+	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f), .flags = PARSE_OPT_NOARG }
 #define OPT_CALLBACK_DEFAULT(s, l, v, a, h, f, d) \
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f), .defval = (intptr_t)d, .flags = PARSE_OPT_LASTARG_DEFAULT }
 
-- 
1.6.0.6



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 4/8] perf: Add a sample_event type to the event_union

[Arjan van de Ven]

>From 4837fa586ab440d6a79f542b90994c1dde9be8ef Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:53:00 +0200
Subject: [PATCH] perf: Add a sample_event type to the event_union

Add a sample_event type to the event_union so that raw samples
can be processed easily.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/util/event.h |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index 1545349..33ced5a 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -60,6 +60,12 @@ struct read_event {
 	u64 id;
 };
 
+struct sample_event{
+	struct perf_event_header        header;
+	u64 array[];
+};
+
+
 typedef union event_union {
 	struct perf_event_header	header;
 	struct ip_event			ip;
@@ -68,6 +74,7 @@ typedef union event_union {
 	struct fork_event		fork;
 	struct lost_event		lost;
 	struct read_event		read;
+	struct sample_event		sample;
 } event_t;
 
 struct map {
-- 
1.6.0.6



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 5/8] trace: Convert the power tracer into an event tracer

[Arjan van de Ven]

>From f591a0f126263733d606838a5d113b9807b12c84 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:53:03 +0200
Subject: [PATCH] trace: Convert the power tracer into an event tracer

This patch converts the existing power tracer into an event tracer,
so that power events (C states and frequency changes) can be tracked
via "perf".

This also removes the perl script that was used to demo the tracer;
its functionality is being replaced entirely with timechart.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    7 +-
 arch/x86/kernel/process.c                  |   28 +---
 include/trace/events/power.h               |   81 ++++++++++
 include/trace/power.h                      |   32 ----
 kernel/trace/Makefile                      |    2 +-
 kernel/trace/power-traces.c                |   18 +++
 kernel/trace/trace.h                       |    8 -
 kernel/trace/trace_event_types.h           |   10 --
 kernel/trace/trace_power.c                 |  218 ----------------------------
 scripts/tracing/power.pl                   |  108 --------------
 10 files changed, 111 insertions(+), 401 deletions(-)
 create mode 100644 include/trace/events/power.h
 delete mode 100644 include/trace/power.h
 create mode 100644 kernel/trace/power-traces.c
 delete mode 100644 kernel/trace/trace_power.c
 delete mode 100644 scripts/tracing/power.pl

diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index ae9b503..dc23bf7 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -33,7 +33,7 @@
 #include <linux/cpufreq.h>
 #include <linux/compiler.h>
 #include <linux/dmi.h>
-#include <trace/power.h>
+#include <trace/events/power.h>
 
 #include <linux/acpi.h>
 #include <linux/io.h>
@@ -77,8 +77,6 @@ struct acpi_msr_data {
 
 static DEFINE_PER_CPU(struct acpi_msr_data, msr_data);
 
-DEFINE_TRACE(power_mark);
-
 /* acpi_perf_data is a pointer to percpu data. */
 static struct acpi_processor_performance *acpi_perf_data;
 
@@ -394,7 +392,6 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 	unsigned int next_perf_state = 0; /* Index into perf table */
 	unsigned int i;
 	int result = 0;
-	struct power_trace it;
 
 	dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 
@@ -426,7 +423,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 		}
 	}
 
-	trace_power_mark(&it, POWER_PSTATE, next_perf_state);
+	trace_power_frequency(POWER_PSTATE, data->freq_table[next_state].frequency);
 
 	switch (data->cpu_feature) {
 	case SYSTEM_INTEL_MSR_CAPABLE:
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 1092a1a..98e9cac 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -9,7 +9,7 @@
 #include <linux/pm.h>
 #include <linux/clockchips.h>
 #include <linux/random.h>
-#include <trace/power.h>
+#include <trace/events/power.h>
 #include <asm/system.h>
 #include <asm/apic.h>
 #include <asm/syscalls.h>
@@ -27,9 +27,6 @@ EXPORT_SYMBOL(idle_nomwait);
 
 struct kmem_cache *task_xstate_cachep;
 
-DEFINE_TRACE(power_start);
-DEFINE_TRACE(power_end);
-
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
 	*dst = *src;
@@ -289,9 +286,7 @@ static inline int hlt_use_halt(void)
 void default_idle(void)
 {
 	if (hlt_use_halt()) {
-		struct power_trace it;
-
-		trace_power_start(&it, POWER_CSTATE, 1);
+		trace_power_start(POWER_CSTATE, 1);
 		current_thread_info()->status &= ~TS_POLLING;
 		/*
 		 * TS_POLLING-cleared state must be visible before we
@@ -304,7 +299,7 @@ void default_idle(void)
 		else
 			local_irq_enable();
 		current_thread_info()->status |= TS_POLLING;
-		trace_power_end(&it);
+		trace_power_end(0);
 	} else {
 		local_irq_enable();
 		/* loop is done by the caller */
@@ -362,9 +357,7 @@ EXPORT_SYMBOL_GPL(cpu_idle_wait);
  */
 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 {
-	struct power_trace it;
-
-	trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
+	trace_power_start(POWER_CSTATE, (ax>>4)+1);
 	if (!need_resched()) {
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
@@ -374,15 +367,14 @@ void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 		if (!need_resched())
 			__mwait(ax, cx);
 	}
-	trace_power_end(&it);
+	trace_power_end(0);
 }
 
 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 static void mwait_idle(void)
 {
-	struct power_trace it;
 	if (!need_resched()) {
-		trace_power_start(&it, POWER_CSTATE, 1);
+		trace_power_start(POWER_CSTATE, 1);
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
 
@@ -392,7 +384,7 @@ static void mwait_idle(void)
 			__sti_mwait(0, 0);
 		else
 			local_irq_enable();
-		trace_power_end(&it);
+		trace_power_end(0);
 	} else
 		local_irq_enable();
 }
@@ -404,13 +396,11 @@ static void mwait_idle(void)
  */
 static void poll_idle(void)
 {
-	struct power_trace it;
-
-	trace_power_start(&it, POWER_CSTATE, 0);
+	trace_power_start(POWER_CSTATE, 0);
 	local_irq_enable();
 	while (!need_resched())
 		cpu_relax();
-	trace_power_end(&it);
+	trace_power_end(0);
 }
 
 /*
diff --git a/include/trace/events/power.h b/include/trace/events/power.h
new file mode 100644
index 0000000..ea6d579
--- /dev/null
+++ b/include/trace/events/power.h
@@ -0,0 +1,81 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM power
+
+#if !defined(_TRACE_POWER_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_POWER_H
+
+#include <linux/ktime.h>
+#include <linux/tracepoint.h>
+
+#ifndef _TRACE_POWER_ENUM_
+#define _TRACE_POWER_ENUM_
+enum {
+	POWER_NONE = 0,
+	POWER_CSTATE = 1,
+	POWER_PSTATE = 2,
+};
+#endif
+
+
+
+TRACE_EVENT(power_start,
+
+	TP_PROTO(unsigned int type, unsigned int state),
+
+	TP_ARGS(type, state),
+
+	TP_STRUCT__entry(
+		__field(	u64,		type		)
+		__field(	u64,		state		)
+	),
+
+	TP_fast_assign(
+		__entry->type = type;
+		__entry->state = state;
+	),
+
+	TP_printk("type=%lu state=%lu", (unsigned long)__entry->type, (unsigned long)__entry->state)
+);
+
+TRACE_EVENT(power_end,
+
+	TP_PROTO(int dummy),
+
+	TP_ARGS(dummy),
+
+	TP_STRUCT__entry(
+		__field(	u64,		dummy		)
+	),
+
+	TP_fast_assign(
+		__entry->dummy = 0xffff;
+	),
+
+	TP_printk("dummy=%lu", (unsigned long)__entry->dummy)
+
+);
+
+
+TRACE_EVENT(power_frequency,
+
+	TP_PROTO(unsigned int type, unsigned int state),
+
+	TP_ARGS(type, state),
+
+	TP_STRUCT__entry(
+		__field(	u64,		type		)
+		__field(	u64,		state		)
+	),
+
+	TP_fast_assign(
+		__entry->type = type;
+		__entry->state = state;
+	),
+
+	TP_printk("type=%lu state=%lu", (unsigned long)__entry->type, (unsigned long) __entry->state)
+);
+
+#endif /* _TRACE_POWER_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/include/trace/power.h b/include/trace/power.h
deleted file mode 100644
index ef20466..0000000
--- a/include/trace/power.h
+++ /dev/null
@@ -1,32 +0,0 @@
-#ifndef _TRACE_POWER_H
-#define _TRACE_POWER_H
-
-#include <linux/ktime.h>
-#include <linux/tracepoint.h>
-
-enum {
-	POWER_NONE = 0,
-	POWER_CSTATE = 1,
-	POWER_PSTATE = 2,
-};
-
-struct power_trace {
-	ktime_t			stamp;
-	ktime_t			end;
-	int			type;
-	int			state;
-};
-
-DECLARE_TRACE(power_start,
-	TP_PROTO(struct power_trace *it, unsigned int type, unsigned int state),
-	      TP_ARGS(it, type, state));
-
-DECLARE_TRACE(power_mark,
-	TP_PROTO(struct power_trace *it, unsigned int type, unsigned int state),
-	      TP_ARGS(it, type, state));
-
-DECLARE_TRACE(power_end,
-	TP_PROTO(struct power_trace *it),
-	      TP_ARGS(it));
-
-#endif /* _TRACE_POWER_H */
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index ce3b1cd..b5ca90b 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -42,7 +42,6 @@ obj-$(CONFIG_BOOT_TRACER) += trace_boot.o
 obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o
 obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o
 obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o
-obj-$(CONFIG_POWER_TRACER) += trace_power.o
 obj-$(CONFIG_KMEMTRACE) += kmemtrace.o
 obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o
 obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
@@ -54,6 +53,7 @@ obj-$(CONFIG_EVENT_TRACING) += trace_export.o
 obj-$(CONFIG_FTRACE_SYSCALLS) += trace_syscalls.o
 obj-$(CONFIG_EVENT_PROFILE) += trace_event_profile.o
 obj-$(CONFIG_EVENT_TRACING) += trace_events_filter.o
+obj-$(CONFIG_EVENT_TRACING) += power-traces.o
 obj-$(CONFIG_KSYM_TRACER) += trace_ksym.o
 
 libftrace-y := ftrace.o
diff --git a/kernel/trace/power-traces.c b/kernel/trace/power-traces.c
new file mode 100644
index 0000000..cb4fc51
--- /dev/null
+++ b/kernel/trace/power-traces.c
@@ -0,0 +1,18 @@
+/*
+ * Power trace points
+ *
+ * Copyright (C) 2009 Arjan van de Ven <arjan@linux.intel.com>
+ */
+
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/workqueue.h>
+#include <linux/sched.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/power.h>
+
+EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
+
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 9027112..f7d0f46 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -10,7 +10,6 @@
 #include <linux/ftrace.h>
 #include <trace/boot.h>
 #include <linux/kmemtrace.h>
-#include <trace/power.h>
 
 #include <linux/trace_seq.h>
 #include <linux/ftrace_event.h>
@@ -40,7 +39,6 @@ enum trace_type {
 	TRACE_HW_BRANCHES,
 	TRACE_KMEM_ALLOC,
 	TRACE_KMEM_FREE,
-	TRACE_POWER,
 	TRACE_BLK,
 	TRACE_KSYM,
 
@@ -162,11 +160,6 @@ struct hw_branch_entry {
 	u64			to;
 };
 
-struct trace_power {
-	struct trace_entry	ent;
-	struct power_trace	state_data;
-};
-
 enum kmemtrace_type_id {
 	KMEMTRACE_TYPE_KMALLOC = 0,	/* kmalloc() or kfree(). */
 	KMEMTRACE_TYPE_CACHE,		/* kmem_cache_*(). */
@@ -318,7 +311,6 @@ extern void __ftrace_bad_type(void);
 		IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,	\
 			  TRACE_GRAPH_RET);		\
 		IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
-		IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
 		IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry,	\
 			  TRACE_KMEM_ALLOC);	\
 		IF_ASSIGN(var, ent, struct kmemtrace_free_entry,	\
diff --git a/kernel/trace/trace_event_types.h b/kernel/trace/trace_event_types.h
index 6db005e..d7ced2c 100644
--- a/kernel/trace/trace_event_types.h
+++ b/kernel/trace/trace_event_types.h
@@ -142,16 +142,6 @@ TRACE_EVENT_FORMAT(hw_branch, TRACE_HW_BRANCHES, hw_branch_entry, ignore,
 	TP_RAW_FMT("from: %llx to: %llx")
 );
 
-TRACE_EVENT_FORMAT(power, TRACE_POWER, trace_power, ignore,
-	TRACE_STRUCT(
-		TRACE_FIELD_SIGN(ktime_t, state_data.stamp, stamp, 1)
-		TRACE_FIELD_SIGN(ktime_t, state_data.end, end, 1)
-		TRACE_FIELD(int, state_data.type, type)
-		TRACE_FIELD(int, state_data.state, state)
-	),
-	TP_RAW_FMT("%llx->%llx type:%u state:%u")
-);
-
 TRACE_EVENT_FORMAT(kmem_alloc, TRACE_KMEM_ALLOC, kmemtrace_alloc_entry, ignore,
 	TRACE_STRUCT(
 		TRACE_FIELD(enum kmemtrace_type_id, type_id, type_id)
diff --git a/kernel/trace/trace_power.c b/kernel/trace/trace_power.c
deleted file mode 100644
index fe1a00f..0000000
--- a/kernel/trace/trace_power.c
+++ /dev/null
@@ -1,218 +0,0 @@
-/*
- * ring buffer based C-state tracer
- *
- * Arjan van de Ven <arjan@linux.intel.com>
- * Copyright (C) 2008 Intel Corporation
- *
- * Much is borrowed from trace_boot.c which is
- * Copyright (C) 2008 Frederic Weisbecker <fweisbec@gmail.com>
- *
- */
-
-#include <linux/init.h>
-#include <linux/debugfs.h>
-#include <trace/power.h>
-#include <linux/kallsyms.h>
-#include <linux/module.h>
-
-#include "trace.h"
-#include "trace_output.h"
-
-static struct trace_array *power_trace;
-static int __read_mostly trace_power_enabled;
-
-static void probe_power_start(struct power_trace *it, unsigned int type,
-				unsigned int level)
-{
-	if (!trace_power_enabled)
-		return;
-
-	memset(it, 0, sizeof(struct power_trace));
-	it->state = level;
-	it->type = type;
-	it->stamp = ktime_get();
-}
-
-
-static void probe_power_end(struct power_trace *it)
-{
-	struct ftrace_event_call *call = &event_power;
-	struct ring_buffer_event *event;
-	struct ring_buffer *buffer;
-	struct trace_power *entry;
-	struct trace_array_cpu *data;
-	struct trace_array *tr = power_trace;
-
-	if (!trace_power_enabled)
-		return;
-
-	buffer = tr->buffer;
-
-	preempt_disable();
-	it->end = ktime_get();
-	data = tr->data[smp_processor_id()];
-
-	event = trace_buffer_lock_reserve(buffer, TRACE_POWER,
-					  sizeof(*entry), 0, 0);
-	if (!event)
-		goto out;
-	entry	= ring_buffer_event_data(event);
-	entry->state_data = *it;
-	if (!filter_check_discard(call, entry, buffer, event))
-		trace_buffer_unlock_commit(buffer, event, 0, 0);
- out:
-	preempt_enable();
-}
-
-static void probe_power_mark(struct power_trace *it, unsigned int type,
-				unsigned int level)
-{
-	struct ftrace_event_call *call = &event_power;
-	struct ring_buffer_event *event;
-	struct ring_buffer *buffer;
-	struct trace_power *entry;
-	struct trace_array_cpu *data;
-	struct trace_array *tr = power_trace;
-
-	if (!trace_power_enabled)
-		return;
-
-	buffer = tr->buffer;
-
-	memset(it, 0, sizeof(struct power_trace));
-	it->state = level;
-	it->type = type;
-	it->stamp = ktime_get();
-	preempt_disable();
-	it->end = it->stamp;
-	data = tr->data[smp_processor_id()];
-
-	event = trace_buffer_lock_reserve(buffer, TRACE_POWER,
-					  sizeof(*entry), 0, 0);
-	if (!event)
-		goto out;
-	entry	= ring_buffer_event_data(event);
-	entry->state_data = *it;
-	if (!filter_check_discard(call, entry, buffer, event))
-		trace_buffer_unlock_commit(buffer, event, 0, 0);
- out:
-	preempt_enable();
-}
-
-static int tracing_power_register(void)
-{
-	int ret;
-
-	ret = register_trace_power_start(probe_power_start);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_start\n");
-		return ret;
-	}
-	ret = register_trace_power_end(probe_power_end);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_end\n");
-		goto fail_start;
-	}
-	ret = register_trace_power_mark(probe_power_mark);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_mark\n");
-		goto fail_end;
-	}
-	return ret;
-fail_end:
-	unregister_trace_power_end(probe_power_end);
-fail_start:
-	unregister_trace_power_start(probe_power_start);
-	return ret;
-}
-
-static void start_power_trace(struct trace_array *tr)
-{
-	trace_power_enabled = 1;
-}
-
-static void stop_power_trace(struct trace_array *tr)
-{
-	trace_power_enabled = 0;
-}
-
-static void power_trace_reset(struct trace_array *tr)
-{
-	trace_power_enabled = 0;
-	unregister_trace_power_start(probe_power_start);
-	unregister_trace_power_end(probe_power_end);
-	unregister_trace_power_mark(probe_power_mark);
-}
-
-
-static int power_trace_init(struct trace_array *tr)
-{
-	power_trace = tr;
-
-	trace_power_enabled = 1;
-	tracing_power_register();
-
-	tracing_reset_online_cpus(tr);
-	return 0;
-}
-
-static enum print_line_t power_print_line(struct trace_iterator *iter)
-{
-	int ret = 0;
-	struct trace_entry *entry = iter->ent;
-	struct trace_power *field ;
-	struct power_trace *it;
-	struct trace_seq *s = &iter->seq;
-	struct timespec stamp;
-	struct timespec duration;
-
-	trace_assign_type(field, entry);
-	it = &field->state_data;
-	stamp = ktime_to_timespec(it->stamp);
-	duration = ktime_to_timespec(ktime_sub(it->end, it->stamp));
-
-	if (entry->type == TRACE_POWER) {
-		if (it->type == POWER_CSTATE)
-			ret = trace_seq_printf(s, "[%5ld.%09ld] CSTATE: Going to C%i on cpu %i for %ld.%09ld\n",
-					  stamp.tv_sec,
-					  stamp.tv_nsec,
-					  it->state, iter->cpu,
-					  duration.tv_sec,
-					  duration.tv_nsec);
-		if (it->type == POWER_PSTATE)
-			ret = trace_seq_printf(s, "[%5ld.%09ld] PSTATE: Going to P%i on cpu %i\n",
-					  stamp.tv_sec,
-					  stamp.tv_nsec,
-					  it->state, iter->cpu);
-		if (!ret)
-			return TRACE_TYPE_PARTIAL_LINE;
-		return TRACE_TYPE_HANDLED;
-	}
-	return TRACE_TYPE_UNHANDLED;
-}
-
-static void power_print_header(struct seq_file *s)
-{
-	seq_puts(s, "#   TIMESTAMP      STATE  EVENT\n");
-	seq_puts(s, "#       |            |      |\n");
-}
-
-static struct tracer power_tracer __read_mostly =
-{
-	.name		= "power",
-	.init		= power_trace_init,
-	.start		= start_power_trace,
-	.stop		= stop_power_trace,
-	.reset		= power_trace_reset,
-	.print_line	= power_print_line,
-	.print_header	= power_print_header,
-};
-
-static int init_power_trace(void)
-{
-	return register_tracer(&power_tracer);
-}
-device_initcall(init_power_trace);
diff --git a/scripts/tracing/power.pl b/scripts/tracing/power.pl
deleted file mode 100644
index 4f729b3..0000000
--- a/scripts/tracing/power.pl
+++ /dev/null
@@ -1,108 +0,0 @@
-#!/usr/bin/perl
-
-# Copyright 2008, Intel Corporation
-#
-# This file is part of the Linux kernel
-#
-# This program file 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.
-#
-# This program is distributed in the hope that it will be useful, but WITHOUT
-# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
-# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
-# for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program in a file named COPYING; if not, write to the
-# Free Software Foundation, Inc.,
-# 51 Franklin Street, Fifth Floor,
-# Boston, MA 02110-1301 USA
-#
-# Authors:
-# 	Arjan van de Ven <arjan@linux.intel.com>
-
-
-#
-# This script turns a cstate ftrace output into a SVG graphic that shows
-# historic C-state information
-#
-#
-# 	cat /sys/kernel/debug/tracing/trace | perl power.pl > out.svg
-#
-
-my @styles;
-my $base = 0;
-
-my @pstate_last;
-my @pstate_level;
-
-$styles[0] = "fill:rgb(0,0,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[1] = "fill:rgb(0,255,0);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[2] = "fill:rgb(255,0,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[3] = "fill:rgb(255,255,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[4] = "fill:rgb(255,0,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[5] = "fill:rgb(0,255,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[6] = "fill:rgb(0,128,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[7] = "fill:rgb(0,255,128);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[8] = "fill:rgb(0,25,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-
-
-print "<?xml version=\"1.0\" standalone=\"no\"?> \n";
-print "<svg width=\"10000\" height=\"100%\" version=\"1.1\" xmlns=\"http://www.w3.org/2000/svg\">\n";
-
-my $scale = 30000.0;
-while (<>) {
-	my $line = $_;
-	if ($line =~ /([0-9\.]+)\] CSTATE: Going to C([0-9]) on cpu ([0-9]+) for ([0-9\.]+)/) {
-		if ($base == 0) {
-			$base = $1;
-		}
-		my $time = $1 - $base;
-		$time = $time * $scale;
-		my $C = $2;
-		my $cpu = $3;
-		my $y = 400 * $cpu;
-		my $duration = $4 * $scale;
-		my $msec = int($4 * 100000)/100.0;
-		my $height = $C * 20;
-		$style = $styles[$C];
-
-		$y = $y + 140 - $height;
-
-		$x2 = $time + 4;
-		$y2 = $y + 4;
-
-
-		print "<rect x=\"$time\" width=\"$duration\" y=\"$y\" height=\"$height\" style=\"$style\"/>\n";
-		print "<text transform=\"translate($x2,$y2) rotate(90)\">C$C $msec</text>\n";
-	}
-	if ($line =~ /([0-9\.]+)\] PSTATE: Going to P([0-9]) on cpu ([0-9]+)/) {
-		my $time = $1 - $base;
-		my $state = $2;
-		my $cpu = $3;
-
-		if (defined($pstate_last[$cpu])) {
-			my $from = $pstate_last[$cpu];
-			my $oldstate = $pstate_state[$cpu];
-			my $duration = ($time-$from) * $scale;
-
-			$from = $from * $scale;
-			my $to = $from + $duration;
-			my $height = 140 - ($oldstate * (140/8));
-
-			my $y = 400 * $cpu + 200 + $height;
-			my $y2 = $y+4;
-			my $style = $styles[8];
-
-			print "<rect x=\"$from\" y=\"$y\" width=\"$duration\" height=\"5\" style=\"$style\"/>\n";
-			print "<text transform=\"translate($from,$y2)\">P$oldstate (cpu $cpu)</text>\n";
-		};
-
-		$pstate_last[$cpu] = $time;
-		$pstate_state[$cpu] = $state;
-	}
-}
-
-
-print "</svg>\n";
-- 
1.6.0.6



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 6/8] perf: Add a SVG helper library file

[Arjan van de Ven]

>From 0aeceb5dd0714f68bc8fc120fc010791c670558d Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 12:52:11 +0200
Subject: [PATCH] perf: Add a SVG helper library file

The timechart tool writes out SVG format output; this patch
adds a set of helper functions to abstract dealing with SVG
from the core timechart code.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/Makefile         |    1 +
 tools/perf/util/svghelper.c |  382 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/svghelper.h |   25 +++
 3 files changed, 408 insertions(+), 0 deletions(-)
 create mode 100644 tools/perf/util/svghelper.c
 create mode 100644 tools/perf/util/svghelper.h

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 9f8d207..a087bf8 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -373,6 +373,7 @@ LIB_OBJS += util/thread.o
 LIB_OBJS += util/trace-event-parse.o
 LIB_OBJS += util/trace-event-read.o
 LIB_OBJS += util/trace-event-info.o
+LIB_OBJS += util/svghelper.o
 
 BUILTIN_OBJS += builtin-annotate.o
 BUILTIN_OBJS += builtin-help.o
diff --git a/tools/perf/util/svghelper.c b/tools/perf/util/svghelper.c
new file mode 100644
index 0000000..20cac41
--- /dev/null
+++ b/tools/perf/util/svghelper.c
@@ -0,0 +1,382 @@
+/*
+ * svghelper.c - helper functions for outputting svg
+ *
+ * (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 <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+#include "svghelper.h"
+
+static u64 first_time, last_time;
+static u64 turbo_frequency, max_freq;
+
+
+#define SLOT_MULT 30.0
+#define SLOT_HEIGHT 25.0
+#define WIDTH 1000.0
+
+static u64 total_height;
+static FILE *svgfile;
+
+static double cpu2slot(int cpu)
+{
+	return 2 * cpu + 1;
+}
+
+static double cpu2y(int cpu)
+{
+	return cpu2slot(cpu) * SLOT_MULT;
+}
+
+static double time2pixels(u64 time)
+{
+	double X;
+
+	X = WIDTH * (time - first_time) / (last_time - first_time);
+	return X;
+}
+
+void open_svg(const char *filename, int cpus, int rows)
+{
+
+	svgfile = fopen(filename, "w");
+	if (!svgfile) {
+		fprintf(stderr, "Cannot open %s for output\n", filename);
+		return;
+	}
+	total_height = (1 + rows + cpu2slot(cpus)) * SLOT_MULT;
+	fprintf(svgfile, "<?xml version=\"1.0\" standalone=\"no\"?> \n");
+	fprintf(svgfile, "<svg width=\"%4.1f\" height=\"%llu\" version=\"1.1\" xmlns=\"http://www.w3.org/2000/svg\">\n", WIDTH, total_height);
+
+	fprintf(svgfile, "<defs>\n  <style type=\"text/css\">\n    <![CDATA[\n");
+
+	fprintf(svgfile, "      rect          { stroke-width: 1; }\n");
+	fprintf(svgfile, "      rect.process  { fill:rgb(180,180,180); fill-opacity:0.9; stroke-width:1;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.process2 { fill:rgb(180,180,180); fill-opacity:0.9; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.sample   { fill:rgb(  0,  0,255); fill-opacity:0.8; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.blocked  { fill:rgb(255,  0,  0); fill-opacity:0.5; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.waiting  { fill:rgb(255,255,  0); fill-opacity:0.3; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.cpu      { fill:rgb(192,192,192); fill-opacity:0.2; stroke-width:0.5; stroke:rgb(128,128,128); } \n");
+	fprintf(svgfile, "      rect.pstate   { fill:rgb(128,128,128); fill-opacity:0.8; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c1       { fill:rgb(255,214,214); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c2       { fill:rgb(255,172,172); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c3       { fill:rgb(255,130,130); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c4       { fill:rgb(255, 88, 88); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c5       { fill:rgb(255, 44, 44); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c6       { fill:rgb(255,  0,  0); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      line.pstate   { stroke:rgb(255,255,  0); stroke-opacity:0.8; stroke-width:2; } \n");
+
+	fprintf(svgfile, "    ]]>\n   </style>\n</defs>\n");
+}
+
+void svg_box(int Yslot, u64 start, u64 end, const char *type)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), Yslot * SLOT_MULT, SLOT_HEIGHT, type);
+}
+
+void svg_sample(int Yslot, int cpu, u64 start, u64 end, const char *type)
+{
+	double text_size;
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), Yslot * SLOT_MULT, SLOT_HEIGHT, type);
+
+	text_size = (time2pixels(end)-time2pixels(start));
+	if (cpu > 9)
+		text_size = text_size/2;
+	if (text_size > 1.25)
+		text_size = 1.25;
+	if (text_size > 0.001)
+		fprintf(svgfile, "<text transform=\"translate(%1.4f,%1.4f)\" font-size=\"%1.4fpt\">%i</text>\n",
+			time2pixels(start), Yslot *  SLOT_MULT + SLOT_HEIGHT - 1, text_size,  cpu + 1);
+
+}
+
+static char *cpu_model(void)
+{
+	static char cpu_m[255];
+	char buf[256];
+	FILE *file;
+
+	cpu_m[0] = 0;
+	/* CPU type */
+	file = fopen("/proc/cpuinfo", "r");
+	if (file) {
+		while (fgets(buf, 255, file)) {
+			if (strstr(buf, "model name")) {
+				strncpy(cpu_m, &buf[13], 255);
+				break;
+			}
+		}
+		fclose(file);
+	}
+	return cpu_m;
+}
+
+void svg_cpu_box(int cpu, u64 __max_freq, u64 __turbo_freq)
+{
+	char cpu_string[80];
+	if (!svgfile)
+		return;
+
+	max_freq = __max_freq;
+	turbo_frequency = __turbo_freq;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"cpu\"/>\n",
+		time2pixels(first_time),
+		time2pixels(last_time)-time2pixels(first_time),
+		cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT);
+
+	sprintf(cpu_string, "CPU %i", (int)cpu+1);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\">%s</text>\n",
+		10+time2pixels(first_time), cpu2y(cpu) + SLOT_HEIGHT/2, cpu_string);
+
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\" font-size=\"1.25pt\">%s</text>\n",
+		10+time2pixels(first_time), cpu2y(cpu) + SLOT_MULT + SLOT_HEIGHT - 4, cpu_model());
+}
+
+void svg_process(int cpu, u64 start, u64 end, const char *type, const char *name)
+{
+	double width;
+
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT, type);
+	width = time2pixels(end)-time2pixels(start);
+	if (width > 6)
+		width = 6;
+
+	if (width > 0.1)
+		fprintf(svgfile, "<text  transform=\"translate(%4.1f,%4.1f) rotate(90)\" font-size=\"%3.4fpt\">%s</text>\n",
+			time2pixels(start), cpu2y(cpu), width, name);
+}
+
+void svg_cstate(int cpu, u64 start, u64 end, int type)
+{
+	double width;
+	char style[128];
+
+	if (!svgfile)
+		return;
+
+
+	if (type > 6)
+		type = 6;
+	sprintf(style, "c%i", type);
+
+	fprintf(svgfile, "<rect class=\"%s\" x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\"/>\n",
+		style,
+		time2pixels(start), time2pixels(end)-time2pixels(start),
+		cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT);
+
+	width = time2pixels(end)-time2pixels(start);
+	if (width > 6)
+		width = 6;
+
+	if (width > 0.05)
+		fprintf(svgfile, "<text  transform=\"translate(%4.1f,%4.1f) rotate(90)\" font-size=\"%3.4fpt\">C%i</text>\n",
+			time2pixels(start), cpu2y(cpu), width, type);
+}
+
+static char *HzToHuman(unsigned long hz)
+{
+	static char buffer[1024];
+	unsigned long long Hz;
+
+	memset(buffer, 0, 1024);
+
+	Hz = hz;
+
+	/* default: just put the Number in */
+	sprintf(buffer, "%9lli", Hz);
+
+	if (Hz > 1000)
+		sprintf(buffer, " %6lli Mhz", (Hz+500)/1000);
+
+	if (Hz > 1500000)
+		sprintf(buffer, " %6.2f Ghz", (Hz+5000.0)/1000000);
+
+	if (Hz == turbo_frequency)
+		sprintf(buffer, "Turbo");
+
+	return buffer;
+}
+
+void svg_pstate(int cpu, u64 start, u64 end, u64 freq)
+{
+	double height = 0;
+
+	if (!svgfile)
+		return;
+
+	if (max_freq)
+		height = freq * 1.0 / max_freq * (SLOT_HEIGHT + SLOT_MULT);
+	height = 1 + cpu2y(cpu) + SLOT_MULT + SLOT_HEIGHT - height;
+	fprintf(svgfile, "<line x1=\"%4.8f\" x2=\"%4.8f\" y1=\"%4.1f\" y2=\"%4.1f\" class=\"pstate\"/>\n",
+		time2pixels(start), time2pixels(end), height, height);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\" font-size=\"0.25pt\">%s</text>\n",
+		time2pixels(start), height+0.9, HzToHuman(freq));
+
+}
+
+
+void svg_partial_wakeline(u64 start, int row1, int row2)
+{
+	double height;
+
+	if (!svgfile)
+		return;
+
+
+	if (row1 < row2) {
+		if (row1)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT + SLOT_MULT/32);
+
+		if (row2)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row2 * SLOT_MULT - SLOT_MULT/32,  time2pixels(start), row2 * SLOT_MULT);
+	} else {
+		if (row2)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT + SLOT_MULT/32);
+
+		if (row1)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row1 * SLOT_MULT - SLOT_MULT/32,  time2pixels(start), row1 * SLOT_MULT);
+	}
+	height = row1 * SLOT_MULT;
+	if (row2 > row1)
+		height += SLOT_HEIGHT;
+	if (row1)
+		fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(32,255,32)\"/>\n",
+			time2pixels(start), height);
+}
+
+void svg_wakeline(u64 start, int row1, int row2)
+{
+	double height;
+
+	if (!svgfile)
+		return;
+
+
+	if (row1 < row2)
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+			time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row2 * SLOT_MULT);
+	else
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+			time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row1 * SLOT_MULT);
+
+	height = row1 * SLOT_MULT;
+	if (row2 > row1)
+		height += SLOT_HEIGHT;
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(32,255,32)\"/>\n",
+			time2pixels(start), height);
+}
+
+void svg_interrupt(u64 start, int row)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(255,128,128)\"/>\n",
+			time2pixels(start), row * SLOT_MULT);
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(255,128,128)\"/>\n",
+			time2pixels(start), row * SLOT_MULT + SLOT_HEIGHT);
+}
+
+void svg_text(int Yslot, u64 start, const char *text)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\">%s</text>\n",
+		time2pixels(start), Yslot * SLOT_MULT+SLOT_HEIGHT/2, text);
+}
+
+static void svg_legenda_box(int X, const char *text, const char *style)
+{
+	double boxsize;
+	boxsize = SLOT_HEIGHT / 2;
+
+	fprintf(svgfile, "<rect x=\"%i\" width=\"%4.8f\" y=\"0\" height=\"%4.1f\" class=\"%s\"/>\n",
+		X, boxsize, boxsize, style);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f, %4.1f)\" font-size=\"%4.4fpt\">%s</text>\n",
+		X + boxsize + 5, boxsize, 0.8 * boxsize, text);
+}
+
+void svg_legenda(void)
+{
+	if (!svgfile)
+		return;
+
+	svg_legenda_box(0,	"Running", "sample");
+	svg_legenda_box(100,	"Idle","rect.c1");
+	svg_legenda_box(200,	"Deeper Idle", "rect.c3");
+	svg_legenda_box(350,	"Deepest Idle", "rect.c6");
+	svg_legenda_box(550,	"Sleeping", "process2");
+	svg_legenda_box(650,	"Waiting for cpu", "waiting");
+	svg_legenda_box(800,	"Blocked on IO", "blocked");
+}
+
+void svg_time_grid(u64 start, u64 end)
+{
+	u64 i;
+
+	first_time = start;
+	last_time = end;
+
+	first_time = first_time / 100000000 * 100000000;
+
+	if (!svgfile)
+		return;
+
+	i = first_time;
+	while (i < last_time) {
+		int color = 220;
+		double thickness = 0.075;
+		if ((i % 100000000) == 0) {
+			thickness = 0.5;
+			color = 192;
+		}
+		if ((i % 1000000000) == 0) {
+			thickness = 2.0;
+			color = 128;
+		}
+
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%llu\" style=\"stroke:rgb(%i,%i,%i);stroke-width:%1.3f\"/>\n",
+			time2pixels(i), SLOT_MULT/2, time2pixels(i), total_height, color, color, color, thickness);
+
+		i += 10000000;
+	}
+}
+
+void svg_close(void)
+{
+	if (svgfile) {
+		fprintf(svgfile, "</svg>\n");
+		fclose(svgfile);
+		svgfile = NULL;
+	}
+}
diff --git a/tools/perf/util/svghelper.h b/tools/perf/util/svghelper.h
new file mode 100644
index 0000000..ad79b5d
--- /dev/null
+++ b/tools/perf/util/svghelper.h
@@ -0,0 +1,25 @@
+#ifndef _INCLUDE_GUARD_SVG_HELPER_
+#define _INCLUDE_GUARD_SVG_HELPER_
+
+#include "types.h"
+
+extern void open_svg(const char *filename, int cpus, int rows);
+extern void svg_box(int Yslot, u64 start, u64 end, const char *type);
+extern void svg_sample(int Yslot, int cpu, u64 start, u64 end, const char *type);
+extern void svg_cpu_box(int cpu, u64 max_frequency, u64 turbo_frequency);
+
+
+extern void svg_process(int cpu, u64 start, u64 end, const char *type, const char *name);
+extern void svg_cstate(int cpu, u64 start, u64 end, int type);
+extern void svg_pstate(int cpu, u64 start, u64 end, u64 freq);
+
+
+extern void svg_time_grid(u64 start, u64 end);
+extern void svg_legenda(void);
+extern void svg_wakeline(u64 start, int row1, int row2);
+extern void svg_partial_wakeline(u64 start, int row1, int row2);
+extern void svg_interrupt(u64 start, int row);
+extern void svg_text(int Yslot, u64 start, const char *text);
+extern void svg_close(void);
+
+#endif
-- 
1.6.0.6




-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 7/8] perf: Add a perf record --timechart option

[Arjan van de Ven]

>From 2f2ee48c6cbf556724929c55b6c0842f737e9658 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 12:52:13 +0200
Subject: [PATCH] perf: Add a perf record --timechart option

This patch adds a "perf record --timechart" option, to make it easy
to capture traces for use by the timechart tool.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/builtin-record.c |   28 ++++++++++++++++++++++++++++
 1 files changed, 28 insertions(+), 0 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 99a12fe..f06a447 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -47,6 +47,7 @@ static int			append_file			= 0;
 static int			call_graph			= 0;
 static int			inherit_stat			= 0;
 static int			no_samples			= 0;
+static int			timechart			= 0;
 static int			sample_address			= 0;
 
 static long			samples;
@@ -385,6 +386,12 @@ static void create_counter(int counter, int cpu, pid_t pid)
 
 	attr->sample_type	|= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
 
+	if (timechart) {
+		attr->sample_type |= PERF_SAMPLE_CPU;
+		attr->sample_type |= PERF_SAMPLE_TIME;
+		attr->sample_type |= PERF_SAMPLE_RAW;
+	}
+
 	if (freq) {
 		attr->sample_type	|= PERF_SAMPLE_PERIOD;
 		attr->freq		= 1;
@@ -637,6 +644,24 @@ static int __cmd_record(int argc, const char **argv)
 	return 0;
 }
 
+static int do_timechart(const struct option *opt __used, const char *str __used, int unset __used)
+{
+	parse_events(NULL, "power:power_start", 0);
+	parse_events(NULL, "power:power_end", 0);
+	parse_events(NULL, "power:power_frequency", 0);
+	parse_events(NULL, "sched:sched_wakeup", 0);
+	parse_events(NULL, "sched:sched_switch", 0);
+	default_interval = 1;
+	system_wide = 1;
+	timechart = 1;
+	if (!nr_counters) {
+		printf("No counters found.. is debugfs mounted?\n");
+		exit(0);
+	}
+	return 0;
+}
+
+
 static const char * const record_usage[] = {
 	"perf record [<options>] [<command>]",
 	"perf record [<options>] -- <command> [<options>]",
@@ -681,6 +706,9 @@ static const struct option options[] = {
 		    "Sample addresses"),
 	OPT_BOOLEAN('n', "no-samples", &no_samples,
 		    "don't sample"),
+	OPT_CALLBACK_NOOPT('t', "timechart", NULL, NULL,
+		     "create a timechart recording",
+		     do_timechart),
 	OPT_END()
 };
 
-- 
1.6.0.6




-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

[PATCH 8/8] perf: Add the timechart tool

[Arjan van de Ven]

>From 4935620ab9ca8217ed00ae2fdf856b115a5081c0 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 12 Sep 2009 07:53:05 +0200
Subject: [PATCH] perf: Add the timechart tool

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 "infinitly 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.

Detailed preview of how this works can be found at http://blog.fenrus.org/?p=5

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/Makefile            |    7 +
 tools/perf/builtin-timechart.c | 1120 ++++++++++++++++++++++++++++++++++++++++
 tools/perf/builtin.h           |    1 +
 tools/perf/perf.c              |    1 +
 4 files changed, 1129 insertions(+), 0 deletions(-)
 create mode 100644 tools/perf/builtin-timechart.c

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index a087bf8..847437b 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -381,6 +381,7 @@ 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
 
@@ -711,6 +712,12 @@ 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
index 0000000..9b79d4a
--- /dev/null
+++ b/tools/perf/builtin-timechart.c
@@ -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();
+}
diff --git a/tools/perf/builtin.h b/tools/perf/builtin.h
index 3a63e41..f2dafae 100644
--- a/tools/perf/builtin.h
+++ b/tools/perf/builtin.h
@@ -19,6 +19,7 @@ extern int cmd_help(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_version(int argc, const char **argv, const char *prefix);
 extern int cmd_list(int argc, const char **argv, const char *prefix);
diff --git a/tools/perf/perf.c b/tools/perf/perf.c
index fe4589d..7585b5d 100644
--- a/tools/perf/perf.c
+++ b/tools/perf/perf.c
@@ -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 },
-- 
1.6.0.6

 


-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: [PATCH 7/8] perf: Add a perf record --timechart option

[Peter Zijlstra]

On Sat, 2009-09-12 at 13:06 +0200, Arjan van de Ven wrote:
> This patch adds a "perf record --timechart" option, to make it easy
> to capture traces for use by the timechart tool.

Would it perhaps make sense to make that:

 perf timechart record

?

Re: [PATCH 7/8] perf: Add a perf record --timechart option

[Arjan van de Ven]

On Mon, 14 Sep 2009 10:06:00 +0200
Peter Zijlstra <peterz@infradead.org> wrote:

> On Sat, 2009-09-12 at 13:06 +0200, Arjan van de Ven wrote:
> > This patch adds a "perf record --timechart" option, to make it easy
> > to capture traces for use by the timechart tool.
> 
> Would it perhaps make sense to make that:
> 
>  perf timechart record

well yes, but it would just be an alias for the former, since it's the
record code that has to do the recording... and that's a rather obscene
layering violation in how perf is set up I think.


-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: [PATCH 5/8] trace: Convert the power tracer into an event tracer

[Li Zefan]

Arjan van de Ven wrote:
>>From f591a0f126263733d606838a5d113b9807b12c84 Mon Sep 17 00:00:00 2001
> From: Arjan van de Ven <arjan@linux.intel.com>
> Date: Sat, 12 Sep 2009 07:53:03 +0200
> Subject: [PATCH] trace: Convert the power tracer into an event tracer
> 
> This patch converts the existing power tracer into an event tracer,
> so that power events (C states and frequency changes) can be tracked
> via "perf".
> 
> This also removes the perl script that was used to demo the tracer;
> its functionality is being replaced entirely with timechart.
> 

And Documentation/power.txt needs update, or just remove it?

> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>

...

> diff --git a/kernel/trace/power-traces.c b/kernel/trace/power-traces.c
> new file mode 100644
> index 0000000..cb4fc51
> --- /dev/null
> +++ b/kernel/trace/power-traces.c

This file is not necessarily needed.

You can just add definition of CREATE_TRACE_POINTS in process.c.

> @@ -0,0 +1,18 @@
> +/*
> + * Power trace points
> + *
> + * Copyright (C) 2009 Arjan van de Ven <arjan@linux.intel.com>
> + */
> +
> +#include <linux/string.h>
> +#include <linux/types.h>
> +#include <linux/workqueue.h>
> +#include <linux/sched.h>
> +#include <linux/module.h>
> +#include <linux/slab.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/power.h>
> +
> +EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
> +

I think it's power_frequency that needs to be exported?

Re: [PATCH 5/8] trace: Convert the power tracer into an event tracer

[Arjan van de Ven]

On Tue, 15 Sep 2009 11:43:16 +0800
Li Zefan <lizf@cn.fujitsu.com> wrote:

> Arjan van de Ven wrote:
> >>From f591a0f126263733d606838a5d113b9807b12c84 Mon Sep 17 00:00:00
> >>2001
> > From: Arjan van de Ven <arjan@linux.intel.com>
> > Date: Sat, 12 Sep 2009 07:53:03 +0200
> > Subject: [PATCH] trace: Convert the power tracer into an event
> > tracer
> > 
> > This patch converts the existing power tracer into an event tracer,
> > so that power events (C states and frequency changes) can be tracked
> > via "perf".
> > 
> > This also removes the perl script that was used to demo the tracer;
> > its functionality is being replaced entirely with timechart.
> > 
> 
> And Documentation/power.txt needs update, or just remove it?

remove it; will do
> 
> > Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
> 
> ...
> 
> > diff --git a/kernel/trace/power-traces.c
> > b/kernel/trace/power-traces.c new file mode 100644
> > index 0000000..cb4fc51
> > --- /dev/null
> > +++ b/kernel/trace/power-traces.c
> 
> This file is not necessarily needed.
> 
> You can just add definition of CREATE_TRACE_POINTS in process.c.

hmm it seems existing style to do this in kernel/trace/ though

> > +
> > +EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
> > +
> 
> I think it's power_frequency that needs to be exported?

very valid point, will fix.

thank you for reviewing the patch; I will send an updated patch out
asap.


-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: [PATCH 2/8] perf: Store trace event name/id pairs in perf.data

[Ingo Molnar]

* Arjan van de Ven <arjan@infradead.org> wrote:

> From c9d9dfc617df7a6a6029feb486f2433b7d825ef4 Mon Sep 17 00:00:00 2001
> From: Arjan van de Ven <arjan@linux.intel.com>
> Date: Sat, 12 Sep 2009 07:52:51 +0200
> Subject: [PATCH] perf: Store trace event name/id pairs in perf.data
> 
> The trace event name<->id mapping is dynamic for each kernel compile.
> In order for perf.data to be useable outside the actual system, we thus
> need to store a table of this mapping for later use.
> 
> This patch adds this table to perf.data, and provides helper functions
> for lookup up fields from this table.
> 
> To avoid mistakes, lookup-from-table is kept completely seprate from
> lookup-from-local-debugfs.
> 
> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
> ---
>  tools/perf/util/header.c       |   59 ++++++++++++++++++++++++++++++++++++++++
>  tools/perf/util/header.h       |    6 ++++
>  tools/perf/util/parse-events.c |   25 +++++++++++++++++
>  3 files changed, 90 insertions(+), 0 deletions(-)
> 
> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> index ec4d4c2..ef91145 100644
> --- a/tools/perf/util/header.c
> +++ b/tools/perf/util/header.c
> @@ -86,6 +86,44 @@ void perf_header__add_attr(struct perf_header *self,
>  	self->attr[pos] = attr;
>  }
>  
> +struct perf_trace_event_type {
> +	u64	event_id;
> +	char	name[64];
> +};
> +
> +static int event_count;
> +static struct perf_trace_event_type *events;
> +
> +void perf_header__push_event(u64 id, const char *name)
> +{
> +	if (strlen(name) > 64)
> +		printf("Event %s will be truncated\n", name);
> +
> +	if (!events) {
> +		events = malloc(sizeof(struct perf_trace_event_type));
> +		if (!events)
> +			die("nomem");
> +	} else {
> +		events = realloc(events, (event_count + 1) * sizeof(struct perf_trace_event_type));
> +		if (!events)
> +			die("nomem");
> +	}
> +	memset(&events[event_count], 0, sizeof(struct perf_trace_event_type));
> +	events[event_count].event_id = id;
> +	strncpy(events[event_count].name, name, 63);
> +	event_count++;

This has the string length limit '64' embedded in 3 open-coded forms...

	Ingo

Re: [PATCH 7/8] perf: Add a perf record --timechart option

[Ingo Molnar]

* Arjan van de Ven <arjan@infradead.org> wrote:

> On Mon, 14 Sep 2009 10:06:00 +0200
> Peter Zijlstra <peterz@infradead.org> wrote:
> 
> > On Sat, 2009-09-12 at 13:06 +0200, Arjan van de Ven wrote:
> > > This patch adds a "perf record --timechart" option, to make it easy
> > > to capture traces for use by the timechart tool.
> > 
> > Would it perhaps make sense to make that:
> > 
> >  perf timechart record
> 
> well yes, but it would just be an alias for the former, since it's the 
> record code that has to do the recording... and that's a rather 
> obscene layering violation in how perf is set up I think.

No, please do it like 'perf sched record' does - it constructs an 
argument list of pre-cooked options. See builtin-sched.c's record_args[] 
array.

	Ingo

Re: [PATCH 8/8] perf: Add the timechart tool

[Ingo Molnar]

* Arjan van de Ven <arjan@infradead.org> wrote:

> ---
>  tools/perf/Makefile            |    7 +
>  tools/perf/builtin-timechart.c | 1120 ++++++++++++++++++++++++++++++++++++++++
>  tools/perf/builtin.h           |    1 +
>  tools/perf/perf.c              |    1 +
>  4 files changed, 1129 insertions(+), 0 deletions(-)

i've applied your latest patches (sans the perf record --timechart 
thing) - thanks Arjan.

There's a few missing things with the main timechart command:

 - it's not in tools/perf/common-cmds.h that way 'perf help' does not 
   list 'timechart' as an available command.

 - there's no tools/perf/Documentation/perf-timechart.txt file - hence 
   'perf help timechart' does not work and there's no manpage/etc.

	Ingo

Re: [PATCH 8/8] perf: Add the timechart tool

[Ingo Molnar]

there's also a few other infrastructure things that should be improved:

 - struct wake_event / trace_entry / wakeup_entry / sched_switch etc. 
   are all present in the existing tracing bits or in builtin-sched. 
   These should be factored out into common facilities and shared 
   between perf timechart and perf sched. We in essence want a good 
   facility to access tracepoint data.

 - the kernel side scheduler tracepoints relied on here should be 
   updated from TRACE_EVENT to TRACE_EVENT_ABI. (TRACE_EVENT_ABI should 
   just be an alias of TRACE_EVENT)

	Ingo

[tip:perfcounters/core] perf: Add a timestamp to fork events

[tip-bot for Arjan van de Ven]

Commit-ID:  393b2ad8c757ba3ccd2a99ca5bbcd6db4d3fa53d
Gitweb:     http://git.kernel.org/tip/393b2ad8c757ba3ccd2a99ca5bbcd6db4d3fa53d
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 07:52:47 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:10 +0200

perf: Add a timestamp to fork events

perf timechart needs to know when a process forked, in order to be
able to visualize properly when tasks start.

This patch adds a time field to the event structure, and fills it
in appropriately.

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: <20090912130341.51ad2de2@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 include/linux/perf_counter.h |    2 ++
 kernel/perf_counter.c        |   11 +++++++++--
 tools/perf/util/event.h      |    1 +
 3 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index c7375f9..bd34100 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -336,6 +336,7 @@ enum perf_event_type {
 	 *	struct perf_event_header	header;
 	 *	u32				pid, ppid;
 	 *	u32				tid, ptid;
+	 *	u64				time;
 	 * };
 	 */
 	PERF_EVENT_EXIT			= 4,
@@ -356,6 +357,7 @@ enum perf_event_type {
 	 *	struct perf_event_header	header;
 	 *	u32				pid, ppid;
 	 *	u32				tid, ptid;
+	 *	{ u64				time;     } && PERF_SAMPLE_TIME
 	 * };
 	 */
 	PERF_EVENT_FORK			= 7,
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index d013f4e..d5899b6 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -3083,6 +3083,7 @@ struct perf_task_event {
 		u32				ppid;
 		u32				tid;
 		u32				ptid;
+		u64				time;
 	} event;
 };
 
@@ -3090,9 +3091,12 @@ static void perf_counter_task_output(struct perf_counter *counter,
 				     struct perf_task_event *task_event)
 {
 	struct perf_output_handle handle;
-	int size = task_event->event.header.size;
+	int size;
 	struct task_struct *task = task_event->task;
-	int ret = perf_output_begin(&handle, counter, size, 0, 0);
+	int ret;
+
+	size  = task_event->event.header.size;
+	ret = perf_output_begin(&handle, counter, size, 0, 0);
 
 	if (ret)
 		return;
@@ -3103,7 +3107,10 @@ static void perf_counter_task_output(struct perf_counter *counter,
 	task_event->event.tid = perf_counter_tid(counter, task);
 	task_event->event.ptid = perf_counter_tid(counter, current);
 
+	task_event->event.time = perf_clock();
+
 	perf_output_put(&handle, task_event->event);
+
 	perf_output_end(&handle);
 }
 
diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index 2495529..28a579f 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -39,6 +39,7 @@ struct fork_event {
 	struct perf_event_header header;
 	u32 pid, ppid;
 	u32 tid, ptid;
+	u64 time;
 };
 
 struct lost_event {

[tip:perfcounters/core] perf: Store trace event name/id pairs in perf.data

[tip-bot for Arjan van de Ven]

Commit-ID:  8755a8f27ae590dde225f9005ee18c6d9c4c6d78
Gitweb:     http://git.kernel.org/tip/8755a8f27ae590dde225f9005ee18c6d9c4c6d78
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 07:52:51 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:11 +0200

perf: Store trace event name/id pairs in perf.data

The trace event name<->id mapping is dynamic for each kernel
compile. In order for perf.data to be useable outside the actual
system, we thus need to store a table of this mapping for later
use.

This patch adds this table to perf.data, and provides helper
functions for lookup up fields from this table.

To avoid mistakes, lookup-from-table is kept completely seprate
from lookup-from-local-debugfs.

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: <20090912130405.6960d099@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 tools/perf/util/header.c       |   59 ++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/header.h       |    6 ++++
 tools/perf/util/parse-events.c |   25 +++++++++++++++++
 3 files changed, 90 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index ec4d4c2..ef91145 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -86,6 +86,44 @@ void perf_header__add_attr(struct perf_header *self,
 	self->attr[pos] = attr;
 }
 
+struct perf_trace_event_type {
+	u64	event_id;
+	char	name[64];
+};
+
+static int event_count;
+static struct perf_trace_event_type *events;
+
+void perf_header__push_event(u64 id, const char *name)
+{
+	if (strlen(name) > 64)
+		printf("Event %s will be truncated\n", name);
+
+	if (!events) {
+		events = malloc(sizeof(struct perf_trace_event_type));
+		if (!events)
+			die("nomem");
+	} else {
+		events = realloc(events, (event_count + 1) * sizeof(struct perf_trace_event_type));
+		if (!events)
+			die("nomem");
+	}
+	memset(&events[event_count], 0, sizeof(struct perf_trace_event_type));
+	events[event_count].event_id = id;
+	strncpy(events[event_count].name, name, 63);
+	event_count++;
+}
+
+char *perf_header__find_event(u64 id)
+{
+	int i;
+	for (i = 0 ; i < event_count; i++) {
+		if (events[i].event_id == id)
+			return events[i].name;
+	}
+	return NULL;
+}
+
 static const char *__perf_magic = "PERFFILE";
 
 #define PERF_MAGIC	(*(u64 *)__perf_magic)
@@ -106,6 +144,7 @@ struct perf_file_header {
 	u64				attr_size;
 	struct perf_file_section	attrs;
 	struct perf_file_section	data;
+	struct perf_file_section	event_types;
 };
 
 static void do_write(int fd, void *buf, size_t size)
@@ -154,6 +193,11 @@ void perf_header__write(struct perf_header *self, int fd)
 		do_write(fd, &f_attr, sizeof(f_attr));
 	}
 
+	self->event_offset = lseek(fd, 0, SEEK_CUR);
+	self->event_size = event_count * sizeof(struct perf_trace_event_type);
+	if (events)
+		do_write(fd, events, self->event_size);
+
 
 	self->data_offset = lseek(fd, 0, SEEK_CUR);
 
@@ -169,6 +213,10 @@ void perf_header__write(struct perf_header *self, int fd)
 			.offset = self->data_offset,
 			.size	= self->data_size,
 		},
+		.event_types = {
+			.offset = self->event_offset,
+			.size	= self->event_size,
+		},
 	};
 
 	lseek(fd, 0, SEEK_SET);
@@ -234,6 +282,17 @@ struct perf_header *perf_header__read(int fd)
 		lseek(fd, tmp, SEEK_SET);
 	}
 
+	if (f_header.event_types.size) {
+		lseek(fd, f_header.event_types.offset, SEEK_SET);
+		events = malloc(f_header.event_types.size);
+		if (!events)
+			die("nomem");
+		do_read(fd, events, f_header.event_types.size);
+		event_count =  f_header.event_types.size / sizeof(struct perf_trace_event_type);
+	}
+	self->event_offset = f_header.event_types.offset;
+	self->event_size   = f_header.event_types.size;
+
 	self->data_offset = f_header.data.offset;
 	self->data_size   = f_header.data.size;
 
diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
index 5d0a72e..7b0e84a 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -19,6 +19,8 @@ struct perf_header {
 	s64 attr_offset;
 	u64 data_offset;
 	u64 data_size;
+	u64 event_offset;
+	u64 event_size;
 };
 
 struct perf_header *perf_header__read(int fd);
@@ -27,6 +29,10 @@ void perf_header__write(struct perf_header *self, int fd);
 void perf_header__add_attr(struct perf_header *self,
 			   struct perf_header_attr *attr);
 
+void perf_header__push_event(u64 id, const char *name);
+char *perf_header__find_event(u64 id);
+
+
 struct perf_header_attr *
 perf_header_attr__new(struct perf_counter_attr *attr);
 void perf_header_attr__add_id(struct perf_header_attr *self, u64 id);
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index a9bdcab..89172fd 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -6,6 +6,7 @@
 #include "exec_cmd.h"
 #include "string.h"
 #include "cache.h"
+#include "header.h"
 
 int					nr_counters;
 
@@ -687,11 +688,35 @@ modifier:
 	return ret;
 }
 
+static void store_event_type(const char *orgname)
+{
+	char filename[PATH_MAX], *c;
+	FILE *file;
+	int id;
+
+	sprintf(filename, "/sys/kernel/debug/tracing/events/%s/id", orgname);
+	c = strchr(filename, ':');
+	if (c)
+		*c = '/';
+
+	file = fopen(filename, "r");
+	if (!file)
+		return;
+	if (fscanf(file, "%i", &id) < 1)
+		die("cannot store event ID");
+	fclose(file);
+	perf_header__push_event(id, orgname);
+}
+
+
 int parse_events(const struct option *opt __used, const char *str, int unset __used)
 {
 	struct perf_counter_attr attr;
 	enum event_result ret;
 
+	if (strchr(str, ':'))
+		store_event_type(str);
+
 	for (;;) {
 		if (nr_counters == MAX_COUNTERS)
 			return -1;

[tip:perfcounters/core] perf: Allow perf utilities to have "callback" options without arguments

[tip-bot for Arjan van de Ven]

Commit-ID:  a79d7db9fdb0037925ffa31f6ff9b189ff75d235
Gitweb:     http://git.kernel.org/tip/a79d7db9fdb0037925ffa31f6ff9b189ff75d235
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 07:52:54 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:11 +0200

perf: Allow perf utilities to have "callback" options without arguments

timechart needs to add a "callback" type command line argument that
does not take arguments.

This patch adds the parse-options.h infrastructure to make this
possible.

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: <20090912130440.548666c1@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 tools/perf/util/parse-options.h |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/parse-options.h b/tools/perf/util/parse-options.h
index 8aa3464..2ee248f 100644
--- a/tools/perf/util/parse-options.h
+++ b/tools/perf/util/parse-options.h
@@ -104,6 +104,8 @@ struct option {
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), .argh = "time", .help = (h), .callback = parse_opt_approxidate_cb }
 #define OPT_CALLBACK(s, l, v, a, h, f) \
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f) }
+#define OPT_CALLBACK_NOOPT(s, l, v, a, h, f) \
+	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f), .flags = PARSE_OPT_NOARG }
 #define OPT_CALLBACK_DEFAULT(s, l, v, a, h, f, d) \
 	{ .type = OPTION_CALLBACK, .short_name = (s), .long_name = (l), .value = (v), (a), .help = (h), .callback = (f), .defval = (intptr_t)d, .flags = PARSE_OPT_LASTARG_DEFAULT }
 

[tip:perfcounters/core] perf: Add a sample_event type to the event_union

[tip-bot for Arjan van de Ven]

Commit-ID:  fd39e055c4fefa4553efc9030f9903afdc6ee323
Gitweb:     http://git.kernel.org/tip/fd39e055c4fefa4553efc9030f9903afdc6ee323
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 07:53:00 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:12 +0200

perf: Add a sample_event type to the event_union

Add a sample_event type to the event_union so that raw samples can
be processed easily.

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: <20090912130511.411434b5@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 tools/perf/util/event.h |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index 28a579f..018d414 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -60,6 +60,12 @@ struct read_event {
 	u64 id;
 };
 
+struct sample_event{
+	struct perf_event_header        header;
+	u64 array[];
+};
+
+
 typedef union event_union {
 	struct perf_event_header	header;
 	struct ip_event			ip;
@@ -68,6 +74,7 @@ typedef union event_union {
 	struct fork_event		fork;
 	struct lost_event		lost;
 	struct read_event		read;
+	struct sample_event		sample;
 } event_t;
 
 struct map {

[tip:perfcounters/core] tracing, perf: Convert the power tracer into an event tracer

[tip-bot for Arjan van de Ven]

Commit-ID:  6161352142d5fed4cd753b32e5ccde66e705b14e
Gitweb:     http://git.kernel.org/tip/6161352142d5fed4cd753b32e5ccde66e705b14e
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Thu, 17 Sep 2009 16:11:28 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:12 +0200

tracing, perf: Convert the power tracer into an event tracer

This patch converts the existing power tracer into an event tracer,
so that power events (C states and frequency changes) can be
tracked via "perf".

This also removes the perl script that was used to demo the tracer;
its functionality is being replaced entirely with timechart.

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: <20090912130542.6d314860@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 Documentation/trace/power.txt              |   17 --
 arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c |    7 +-
 arch/x86/kernel/process.c                  |   28 +---
 include/trace/events/power.h               |   81 ++++++++++
 kernel/trace/Makefile                      |    2 +-
 kernel/trace/power-traces.c                |   20 +++
 kernel/trace/trace.h                       |    3 -
 kernel/trace/trace_entries.h               |   17 --
 kernel/trace/trace_power.c                 |  218 ----------------------------
 scripts/tracing/power.pl                   |  108 --------------
 10 files changed, 113 insertions(+), 388 deletions(-)

diff --git a/Documentation/trace/power.txt b/Documentation/trace/power.txt
deleted file mode 100644
index cd805e1..0000000
--- a/Documentation/trace/power.txt
+++ /dev/null
@@ -1,17 +0,0 @@
-The power tracer collects detailed information about C-state and P-state
-transitions, instead of just looking at the high-level "average"
-information.
-
-There is a helper script found in scrips/tracing/power.pl in the kernel
-sources which can be used to parse this information and create a
-Scalable Vector Graphics (SVG) picture from the trace data.
-
-To use this tracer:
-
-	echo 0 > /sys/kernel/debug/tracing/tracing_enabled
-	echo power > /sys/kernel/debug/tracing/current_tracer
-	echo 1 > /sys/kernel/debug/tracing/tracing_enabled
-	sleep 1
-	echo 0 > /sys/kernel/debug/tracing/tracing_enabled
-	cat /sys/kernel/debug/tracing/trace | \
-		perl scripts/tracing/power.pl > out.sv
diff --git a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
index 4109679..479cc8c 100644
--- a/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -33,7 +33,7 @@
 #include <linux/cpufreq.h>
 #include <linux/compiler.h>
 #include <linux/dmi.h>
-#include <trace/power.h>
+#include <trace/events/power.h>
 
 #include <linux/acpi.h>
 #include <linux/io.h>
@@ -72,8 +72,6 @@ static DEFINE_PER_CPU(struct acpi_cpufreq_data *, drv_data);
 
 static DEFINE_PER_CPU(struct aperfmperf, old_perf);
 
-DEFINE_TRACE(power_mark);
-
 /* acpi_perf_data is a pointer to percpu data. */
 static struct acpi_processor_performance *acpi_perf_data;
 
@@ -332,7 +330,6 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 	unsigned int next_perf_state = 0; /* Index into perf table */
 	unsigned int i;
 	int result = 0;
-	struct power_trace it;
 
 	dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 
@@ -364,7 +361,7 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 		}
 	}
 
-	trace_power_mark(&it, POWER_PSTATE, next_perf_state);
+	trace_power_frequency(POWER_PSTATE, data->freq_table[next_state].frequency);
 
 	switch (data->cpu_feature) {
 	case SYSTEM_INTEL_MSR_CAPABLE:
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 071166a..7b60e39 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -9,7 +9,7 @@
 #include <linux/pm.h>
 #include <linux/clockchips.h>
 #include <linux/random.h>
-#include <trace/power.h>
+#include <trace/events/power.h>
 #include <asm/system.h>
 #include <asm/apic.h>
 #include <asm/syscalls.h>
@@ -25,9 +25,6 @@ EXPORT_SYMBOL(idle_nomwait);
 
 struct kmem_cache *task_xstate_cachep;
 
-DEFINE_TRACE(power_start);
-DEFINE_TRACE(power_end);
-
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
 	*dst = *src;
@@ -299,9 +296,7 @@ static inline int hlt_use_halt(void)
 void default_idle(void)
 {
 	if (hlt_use_halt()) {
-		struct power_trace it;
-
-		trace_power_start(&it, POWER_CSTATE, 1);
+		trace_power_start(POWER_CSTATE, 1);
 		current_thread_info()->status &= ~TS_POLLING;
 		/*
 		 * TS_POLLING-cleared state must be visible before we
@@ -314,7 +309,7 @@ void default_idle(void)
 		else
 			local_irq_enable();
 		current_thread_info()->status |= TS_POLLING;
-		trace_power_end(&it);
+		trace_power_end(0);
 	} else {
 		local_irq_enable();
 		/* loop is done by the caller */
@@ -372,9 +367,7 @@ EXPORT_SYMBOL_GPL(cpu_idle_wait);
  */
 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 {
-	struct power_trace it;
-
-	trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
+	trace_power_start(POWER_CSTATE, (ax>>4)+1);
 	if (!need_resched()) {
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
@@ -384,15 +377,14 @@ void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 		if (!need_resched())
 			__mwait(ax, cx);
 	}
-	trace_power_end(&it);
+	trace_power_end(0);
 }
 
 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 static void mwait_idle(void)
 {
-	struct power_trace it;
 	if (!need_resched()) {
-		trace_power_start(&it, POWER_CSTATE, 1);
+		trace_power_start(POWER_CSTATE, 1);
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
 
@@ -402,7 +394,7 @@ static void mwait_idle(void)
 			__sti_mwait(0, 0);
 		else
 			local_irq_enable();
-		trace_power_end(&it);
+		trace_power_end(0);
 	} else
 		local_irq_enable();
 }
@@ -414,13 +406,11 @@ static void mwait_idle(void)
  */
 static void poll_idle(void)
 {
-	struct power_trace it;
-
-	trace_power_start(&it, POWER_CSTATE, 0);
+	trace_power_start(POWER_CSTATE, 0);
 	local_irq_enable();
 	while (!need_resched())
 		cpu_relax();
-	trace_power_end(&it);
+	trace_power_end(0);
 }
 
 /*
diff --git a/include/trace/events/power.h b/include/trace/events/power.h
new file mode 100644
index 0000000..ea6d579
--- /dev/null
+++ b/include/trace/events/power.h
@@ -0,0 +1,81 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM power
+
+#if !defined(_TRACE_POWER_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_POWER_H
+
+#include <linux/ktime.h>
+#include <linux/tracepoint.h>
+
+#ifndef _TRACE_POWER_ENUM_
+#define _TRACE_POWER_ENUM_
+enum {
+	POWER_NONE = 0,
+	POWER_CSTATE = 1,
+	POWER_PSTATE = 2,
+};
+#endif
+
+
+
+TRACE_EVENT(power_start,
+
+	TP_PROTO(unsigned int type, unsigned int state),
+
+	TP_ARGS(type, state),
+
+	TP_STRUCT__entry(
+		__field(	u64,		type		)
+		__field(	u64,		state		)
+	),
+
+	TP_fast_assign(
+		__entry->type = type;
+		__entry->state = state;
+	),
+
+	TP_printk("type=%lu state=%lu", (unsigned long)__entry->type, (unsigned long)__entry->state)
+);
+
+TRACE_EVENT(power_end,
+
+	TP_PROTO(int dummy),
+
+	TP_ARGS(dummy),
+
+	TP_STRUCT__entry(
+		__field(	u64,		dummy		)
+	),
+
+	TP_fast_assign(
+		__entry->dummy = 0xffff;
+	),
+
+	TP_printk("dummy=%lu", (unsigned long)__entry->dummy)
+
+);
+
+
+TRACE_EVENT(power_frequency,
+
+	TP_PROTO(unsigned int type, unsigned int state),
+
+	TP_ARGS(type, state),
+
+	TP_STRUCT__entry(
+		__field(	u64,		type		)
+		__field(	u64,		state		)
+	),
+
+	TP_fast_assign(
+		__entry->type = type;
+		__entry->state = state;
+	),
+
+	TP_printk("type=%lu state=%lu", (unsigned long)__entry->type, (unsigned long) __entry->state)
+);
+
+#endif /* _TRACE_POWER_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 844164d..26f03ac 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -42,7 +42,6 @@ obj-$(CONFIG_BOOT_TRACER) += trace_boot.o
 obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o
 obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o
 obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o
-obj-$(CONFIG_POWER_TRACER) += trace_power.o
 obj-$(CONFIG_KMEMTRACE) += kmemtrace.o
 obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o
 obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
@@ -54,5 +53,6 @@ obj-$(CONFIG_EVENT_TRACING) += trace_export.o
 obj-$(CONFIG_FTRACE_SYSCALLS) += trace_syscalls.o
 obj-$(CONFIG_EVENT_PROFILE) += trace_event_profile.o
 obj-$(CONFIG_EVENT_TRACING) += trace_events_filter.o
+obj-$(CONFIG_EVENT_TRACING) += power-traces.o
 
 libftrace-y := ftrace.o
diff --git a/kernel/trace/power-traces.c b/kernel/trace/power-traces.c
new file mode 100644
index 0000000..e06c6e3
--- /dev/null
+++ b/kernel/trace/power-traces.c
@@ -0,0 +1,20 @@
+/*
+ * Power trace points
+ *
+ * Copyright (C) 2009 Arjan van de Ven <arjan@linux.intel.com>
+ */
+
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/workqueue.h>
+#include <linux/sched.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/power.h>
+
+EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
+EXPORT_TRACEPOINT_SYMBOL_GPL(power_end);
+EXPORT_TRACEPOINT_SYMBOL_GPL(power_frequency);
+
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 86bcff9..405cb85 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -11,7 +11,6 @@
 #include <linux/ftrace.h>
 #include <trace/boot.h>
 #include <linux/kmemtrace.h>
-#include <trace/power.h>
 
 #include <linux/trace_seq.h>
 #include <linux/ftrace_event.h>
@@ -37,7 +36,6 @@ enum trace_type {
 	TRACE_HW_BRANCHES,
 	TRACE_KMEM_ALLOC,
 	TRACE_KMEM_FREE,
-	TRACE_POWER,
 	TRACE_BLK,
 
 	__TRACE_LAST_TYPE,
@@ -207,7 +205,6 @@ extern void __ftrace_bad_type(void);
 		IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,	\
 			  TRACE_GRAPH_RET);		\
 		IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
-		IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
 		IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry,	\
 			  TRACE_KMEM_ALLOC);	\
 		IF_ASSIGN(var, ent, struct kmemtrace_free_entry,	\
diff --git a/kernel/trace/trace_entries.h b/kernel/trace/trace_entries.h
index a431748..ead3d72 100644
--- a/kernel/trace/trace_entries.h
+++ b/kernel/trace/trace_entries.h
@@ -330,23 +330,6 @@ FTRACE_ENTRY(hw_branch, hw_branch_entry,
 	F_printk("from: %llx to: %llx", __entry->from, __entry->to)
 );
 
-FTRACE_ENTRY(power, trace_power,
-
-	TRACE_POWER,
-
-	F_STRUCT(
-		__field_struct(	struct power_trace,	state_data	)
-		__field_desc(	s64,	state_data,	stamp		)
-		__field_desc(	s64,	state_data,	end		)
-		__field_desc(	int,	state_data,	type		)
-		__field_desc(	int,	state_data,	state		)
-	),
-
-	F_printk("%llx->%llx type:%u state:%u",
-		 __entry->stamp, __entry->end,
-		 __entry->type, __entry->state)
-);
-
 FTRACE_ENTRY(kmem_alloc, kmemtrace_alloc_entry,
 
 	TRACE_KMEM_ALLOC,
diff --git a/kernel/trace/trace_power.c b/kernel/trace/trace_power.c
deleted file mode 100644
index fe1a00f..0000000
--- a/kernel/trace/trace_power.c
+++ /dev/null
@@ -1,218 +0,0 @@
-/*
- * ring buffer based C-state tracer
- *
- * Arjan van de Ven <arjan@linux.intel.com>
- * Copyright (C) 2008 Intel Corporation
- *
- * Much is borrowed from trace_boot.c which is
- * Copyright (C) 2008 Frederic Weisbecker <fweisbec@gmail.com>
- *
- */
-
-#include <linux/init.h>
-#include <linux/debugfs.h>
-#include <trace/power.h>
-#include <linux/kallsyms.h>
-#include <linux/module.h>
-
-#include "trace.h"
-#include "trace_output.h"
-
-static struct trace_array *power_trace;
-static int __read_mostly trace_power_enabled;
-
-static void probe_power_start(struct power_trace *it, unsigned int type,
-				unsigned int level)
-{
-	if (!trace_power_enabled)
-		return;
-
-	memset(it, 0, sizeof(struct power_trace));
-	it->state = level;
-	it->type = type;
-	it->stamp = ktime_get();
-}
-
-
-static void probe_power_end(struct power_trace *it)
-{
-	struct ftrace_event_call *call = &event_power;
-	struct ring_buffer_event *event;
-	struct ring_buffer *buffer;
-	struct trace_power *entry;
-	struct trace_array_cpu *data;
-	struct trace_array *tr = power_trace;
-
-	if (!trace_power_enabled)
-		return;
-
-	buffer = tr->buffer;
-
-	preempt_disable();
-	it->end = ktime_get();
-	data = tr->data[smp_processor_id()];
-
-	event = trace_buffer_lock_reserve(buffer, TRACE_POWER,
-					  sizeof(*entry), 0, 0);
-	if (!event)
-		goto out;
-	entry	= ring_buffer_event_data(event);
-	entry->state_data = *it;
-	if (!filter_check_discard(call, entry, buffer, event))
-		trace_buffer_unlock_commit(buffer, event, 0, 0);
- out:
-	preempt_enable();
-}
-
-static void probe_power_mark(struct power_trace *it, unsigned int type,
-				unsigned int level)
-{
-	struct ftrace_event_call *call = &event_power;
-	struct ring_buffer_event *event;
-	struct ring_buffer *buffer;
-	struct trace_power *entry;
-	struct trace_array_cpu *data;
-	struct trace_array *tr = power_trace;
-
-	if (!trace_power_enabled)
-		return;
-
-	buffer = tr->buffer;
-
-	memset(it, 0, sizeof(struct power_trace));
-	it->state = level;
-	it->type = type;
-	it->stamp = ktime_get();
-	preempt_disable();
-	it->end = it->stamp;
-	data = tr->data[smp_processor_id()];
-
-	event = trace_buffer_lock_reserve(buffer, TRACE_POWER,
-					  sizeof(*entry), 0, 0);
-	if (!event)
-		goto out;
-	entry	= ring_buffer_event_data(event);
-	entry->state_data = *it;
-	if (!filter_check_discard(call, entry, buffer, event))
-		trace_buffer_unlock_commit(buffer, event, 0, 0);
- out:
-	preempt_enable();
-}
-
-static int tracing_power_register(void)
-{
-	int ret;
-
-	ret = register_trace_power_start(probe_power_start);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_start\n");
-		return ret;
-	}
-	ret = register_trace_power_end(probe_power_end);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_end\n");
-		goto fail_start;
-	}
-	ret = register_trace_power_mark(probe_power_mark);
-	if (ret) {
-		pr_info("power trace: Couldn't activate tracepoint"
-			" probe to trace_power_mark\n");
-		goto fail_end;
-	}
-	return ret;
-fail_end:
-	unregister_trace_power_end(probe_power_end);
-fail_start:
-	unregister_trace_power_start(probe_power_start);
-	return ret;
-}
-
-static void start_power_trace(struct trace_array *tr)
-{
-	trace_power_enabled = 1;
-}
-
-static void stop_power_trace(struct trace_array *tr)
-{
-	trace_power_enabled = 0;
-}
-
-static void power_trace_reset(struct trace_array *tr)
-{
-	trace_power_enabled = 0;
-	unregister_trace_power_start(probe_power_start);
-	unregister_trace_power_end(probe_power_end);
-	unregister_trace_power_mark(probe_power_mark);
-}
-
-
-static int power_trace_init(struct trace_array *tr)
-{
-	power_trace = tr;
-
-	trace_power_enabled = 1;
-	tracing_power_register();
-
-	tracing_reset_online_cpus(tr);
-	return 0;
-}
-
-static enum print_line_t power_print_line(struct trace_iterator *iter)
-{
-	int ret = 0;
-	struct trace_entry *entry = iter->ent;
-	struct trace_power *field ;
-	struct power_trace *it;
-	struct trace_seq *s = &iter->seq;
-	struct timespec stamp;
-	struct timespec duration;
-
-	trace_assign_type(field, entry);
-	it = &field->state_data;
-	stamp = ktime_to_timespec(it->stamp);
-	duration = ktime_to_timespec(ktime_sub(it->end, it->stamp));
-
-	if (entry->type == TRACE_POWER) {
-		if (it->type == POWER_CSTATE)
-			ret = trace_seq_printf(s, "[%5ld.%09ld] CSTATE: Going to C%i on cpu %i for %ld.%09ld\n",
-					  stamp.tv_sec,
-					  stamp.tv_nsec,
-					  it->state, iter->cpu,
-					  duration.tv_sec,
-					  duration.tv_nsec);
-		if (it->type == POWER_PSTATE)
-			ret = trace_seq_printf(s, "[%5ld.%09ld] PSTATE: Going to P%i on cpu %i\n",
-					  stamp.tv_sec,
-					  stamp.tv_nsec,
-					  it->state, iter->cpu);
-		if (!ret)
-			return TRACE_TYPE_PARTIAL_LINE;
-		return TRACE_TYPE_HANDLED;
-	}
-	return TRACE_TYPE_UNHANDLED;
-}
-
-static void power_print_header(struct seq_file *s)
-{
-	seq_puts(s, "#   TIMESTAMP      STATE  EVENT\n");
-	seq_puts(s, "#       |            |      |\n");
-}
-
-static struct tracer power_tracer __read_mostly =
-{
-	.name		= "power",
-	.init		= power_trace_init,
-	.start		= start_power_trace,
-	.stop		= stop_power_trace,
-	.reset		= power_trace_reset,
-	.print_line	= power_print_line,
-	.print_header	= power_print_header,
-};
-
-static int init_power_trace(void)
-{
-	return register_tracer(&power_tracer);
-}
-device_initcall(init_power_trace);
diff --git a/scripts/tracing/power.pl b/scripts/tracing/power.pl
deleted file mode 100644
index 4f729b3..0000000
--- a/scripts/tracing/power.pl
+++ /dev/null
@@ -1,108 +0,0 @@
-#!/usr/bin/perl
-
-# Copyright 2008, Intel Corporation
-#
-# This file is part of the Linux kernel
-#
-# This program file 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.
-#
-# This program is distributed in the hope that it will be useful, but WITHOUT
-# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
-# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
-# for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program in a file named COPYING; if not, write to the
-# Free Software Foundation, Inc.,
-# 51 Franklin Street, Fifth Floor,
-# Boston, MA 02110-1301 USA
-#
-# Authors:
-# 	Arjan van de Ven <arjan@linux.intel.com>
-
-
-#
-# This script turns a cstate ftrace output into a SVG graphic that shows
-# historic C-state information
-#
-#
-# 	cat /sys/kernel/debug/tracing/trace | perl power.pl > out.svg
-#
-
-my @styles;
-my $base = 0;
-
-my @pstate_last;
-my @pstate_level;
-
-$styles[0] = "fill:rgb(0,0,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[1] = "fill:rgb(0,255,0);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[2] = "fill:rgb(255,0,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[3] = "fill:rgb(255,255,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[4] = "fill:rgb(255,0,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[5] = "fill:rgb(0,255,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[6] = "fill:rgb(0,128,255);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[7] = "fill:rgb(0,255,128);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-$styles[8] = "fill:rgb(0,25,20);fill-opacity:0.5;stroke-width:1;stroke:rgb(0,0,0)";
-
-
-print "<?xml version=\"1.0\" standalone=\"no\"?> \n";
-print "<svg width=\"10000\" height=\"100%\" version=\"1.1\" xmlns=\"http://www.w3.org/2000/svg\">\n";
-
-my $scale = 30000.0;
-while (<>) {
-	my $line = $_;
-	if ($line =~ /([0-9\.]+)\] CSTATE: Going to C([0-9]) on cpu ([0-9]+) for ([0-9\.]+)/) {
-		if ($base == 0) {
-			$base = $1;
-		}
-		my $time = $1 - $base;
-		$time = $time * $scale;
-		my $C = $2;
-		my $cpu = $3;
-		my $y = 400 * $cpu;
-		my $duration = $4 * $scale;
-		my $msec = int($4 * 100000)/100.0;
-		my $height = $C * 20;
-		$style = $styles[$C];
-
-		$y = $y + 140 - $height;
-
-		$x2 = $time + 4;
-		$y2 = $y + 4;
-
-
-		print "<rect x=\"$time\" width=\"$duration\" y=\"$y\" height=\"$height\" style=\"$style\"/>\n";
-		print "<text transform=\"translate($x2,$y2) rotate(90)\">C$C $msec</text>\n";
-	}
-	if ($line =~ /([0-9\.]+)\] PSTATE: Going to P([0-9]) on cpu ([0-9]+)/) {
-		my $time = $1 - $base;
-		my $state = $2;
-		my $cpu = $3;
-
-		if (defined($pstate_last[$cpu])) {
-			my $from = $pstate_last[$cpu];
-			my $oldstate = $pstate_state[$cpu];
-			my $duration = ($time-$from) * $scale;
-
-			$from = $from * $scale;
-			my $to = $from + $duration;
-			my $height = 140 - ($oldstate * (140/8));
-
-			my $y = 400 * $cpu + 200 + $height;
-			my $y2 = $y+4;
-			my $style = $styles[8];
-
-			print "<rect x=\"$from\" y=\"$y\" width=\"$duration\" height=\"5\" style=\"$style\"/>\n";
-			print "<text transform=\"translate($from,$y2)\">P$oldstate (cpu $cpu)</text>\n";
-		};
-
-		$pstate_last[$cpu] = $time;
-		$pstate_state[$cpu] = $state;
-	}
-}
-
-
-print "</svg>\n";

[tip:perfcounters/core] perf: Add a SVG helper library file

[tip-bot for Arjan van de Ven]

Commit-ID:  f48d55ce7871824eae3065f4d81956d7113eff19
Gitweb:     http://git.kernel.org/tip/f48d55ce7871824eae3065f4d81956d7113eff19
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 12:52:11 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:13 +0200

perf: Add a SVG helper library file

The timechart tool writes out SVG format output; this patch adds a
set of helper functions to abstract dealing with SVG from the core
timechart code.

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: <20090912130613.677f0516@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 tools/perf/Makefile         |    1 +
 tools/perf/util/svghelper.c |  382 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/svghelper.h |   25 +++
 3 files changed, 408 insertions(+), 0 deletions(-)

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 2cb8cc3..0388e36 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -373,6 +373,7 @@ LIB_OBJS += util/thread.o
 LIB_OBJS += util/trace-event-parse.o
 LIB_OBJS += util/trace-event-read.o
 LIB_OBJS += util/trace-event-info.o
+LIB_OBJS += util/svghelper.o
 
 BUILTIN_OBJS += builtin-annotate.o
 BUILTIN_OBJS += builtin-help.o
diff --git a/tools/perf/util/svghelper.c b/tools/perf/util/svghelper.c
new file mode 100644
index 0000000..c7a29af
--- /dev/null
+++ b/tools/perf/util/svghelper.c
@@ -0,0 +1,382 @@
+/*
+ * svghelper.c - helper functions for outputting svg
+ *
+ * (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 <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+#include "svghelper.h"
+
+static u64 first_time, last_time;
+static u64 turbo_frequency, max_freq;
+
+
+#define SLOT_MULT 30.0
+#define SLOT_HEIGHT 25.0
+#define WIDTH 1000.0
+
+static u64 total_height;
+static FILE *svgfile;
+
+static double cpu2slot(int cpu)
+{
+	return 2 * cpu + 1;
+}
+
+static double cpu2y(int cpu)
+{
+	return cpu2slot(cpu) * SLOT_MULT;
+}
+
+static double time2pixels(u64 time)
+{
+	double X;
+
+	X = WIDTH * (time - first_time) / (last_time - first_time);
+	return X;
+}
+
+void open_svg(const char *filename, int cpus, int rows)
+{
+
+	svgfile = fopen(filename, "w");
+	if (!svgfile) {
+		fprintf(stderr, "Cannot open %s for output\n", filename);
+		return;
+	}
+	total_height = (1 + rows + cpu2slot(cpus)) * SLOT_MULT;
+	fprintf(svgfile, "<?xml version=\"1.0\" standalone=\"no\"?> \n");
+	fprintf(svgfile, "<svg width=\"%4.1f\" height=\"%llu\" version=\"1.1\" xmlns=\"http://www.w3.org/2000/svg\">\n", WIDTH, total_height);
+
+	fprintf(svgfile, "<defs>\n  <style type=\"text/css\">\n    <![CDATA[\n");
+
+	fprintf(svgfile, "      rect          { stroke-width: 1; }\n");
+	fprintf(svgfile, "      rect.process  { fill:rgb(180,180,180); fill-opacity:0.9; stroke-width:1;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.process2 { fill:rgb(180,180,180); fill-opacity:0.9; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.sample   { fill:rgb(  0,  0,255); fill-opacity:0.8; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.blocked  { fill:rgb(255,  0,  0); fill-opacity:0.5; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.waiting  { fill:rgb(255,255,  0); fill-opacity:0.3; stroke-width:0;   stroke:rgb(  0,  0,  0); } \n");
+	fprintf(svgfile, "      rect.cpu      { fill:rgb(192,192,192); fill-opacity:0.2; stroke-width:0.5; stroke:rgb(128,128,128); } \n");
+	fprintf(svgfile, "      rect.pstate   { fill:rgb(128,128,128); fill-opacity:0.8; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c1       { fill:rgb(255,214,214); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c2       { fill:rgb(255,172,172); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c3       { fill:rgb(255,130,130); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c4       { fill:rgb(255, 88, 88); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c5       { fill:rgb(255, 44, 44); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      rect.c6       { fill:rgb(255,  0,  0); fill-opacity:0.5; stroke-width:0; } \n");
+	fprintf(svgfile, "      line.pstate   { stroke:rgb(255,255,  0); stroke-opacity:0.8; stroke-width:2; } \n");
+
+	fprintf(svgfile, "    ]]>\n   </style>\n</defs>\n");
+}
+
+void svg_box(int Yslot, u64 start, u64 end, const char *type)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), Yslot * SLOT_MULT, SLOT_HEIGHT, type);
+}
+
+void svg_sample(int Yslot, int cpu, u64 start, u64 end, const char *type)
+{
+	double text_size;
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), Yslot * SLOT_MULT, SLOT_HEIGHT, type);
+
+	text_size = (time2pixels(end)-time2pixels(start));
+	if (cpu > 9)
+		text_size = text_size/2;
+	if (text_size > 1.25)
+		text_size = 1.25;
+	if (text_size > 0.0001)
+		fprintf(svgfile, "<text transform=\"translate(%1.6f,%1.6f)\" font-size=\"%1.6fpt\">%i</text>\n",
+			time2pixels(start), Yslot *  SLOT_MULT + SLOT_HEIGHT - 1, text_size,  cpu + 1);
+
+}
+
+static char *cpu_model(void)
+{
+	static char cpu_m[255];
+	char buf[256];
+	FILE *file;
+
+	cpu_m[0] = 0;
+	/* CPU type */
+	file = fopen("/proc/cpuinfo", "r");
+	if (file) {
+		while (fgets(buf, 255, file)) {
+			if (strstr(buf, "model name")) {
+				strncpy(cpu_m, &buf[13], 255);
+				break;
+			}
+		}
+		fclose(file);
+	}
+	return cpu_m;
+}
+
+void svg_cpu_box(int cpu, u64 __max_freq, u64 __turbo_freq)
+{
+	char cpu_string[80];
+	if (!svgfile)
+		return;
+
+	max_freq = __max_freq;
+	turbo_frequency = __turbo_freq;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"cpu\"/>\n",
+		time2pixels(first_time),
+		time2pixels(last_time)-time2pixels(first_time),
+		cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT);
+
+	sprintf(cpu_string, "CPU %i", (int)cpu+1);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\">%s</text>\n",
+		10+time2pixels(first_time), cpu2y(cpu) + SLOT_HEIGHT/2, cpu_string);
+
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\" font-size=\"1.25pt\">%s</text>\n",
+		10+time2pixels(first_time), cpu2y(cpu) + SLOT_MULT + SLOT_HEIGHT - 4, cpu_model());
+}
+
+void svg_process(int cpu, u64 start, u64 end, const char *type, const char *name)
+{
+	double width;
+
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<rect x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\" class=\"%s\"/>\n",
+		time2pixels(start), time2pixels(end)-time2pixels(start), cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT, type);
+	width = time2pixels(end)-time2pixels(start);
+	if (width > 6)
+		width = 6;
+
+	if (width > 0.001)
+		fprintf(svgfile, "<text  transform=\"translate(%4.5f,%4.5f) rotate(90)\" font-size=\"%3.4fpt\">%s</text>\n",
+			time2pixels(start), cpu2y(cpu), width, name);
+}
+
+void svg_cstate(int cpu, u64 start, u64 end, int type)
+{
+	double width;
+	char style[128];
+
+	if (!svgfile)
+		return;
+
+
+	if (type > 6)
+		type = 6;
+	sprintf(style, "c%i", type);
+
+	fprintf(svgfile, "<rect class=\"%s\" x=\"%4.8f\" width=\"%4.8f\" y=\"%4.1f\" height=\"%4.1f\"/>\n",
+		style,
+		time2pixels(start), time2pixels(end)-time2pixels(start),
+		cpu2y(cpu), SLOT_MULT+SLOT_HEIGHT);
+
+	width = time2pixels(end)-time2pixels(start);
+	if (width > 6)
+		width = 6;
+
+	if (width > 0.05)
+		fprintf(svgfile, "<text  transform=\"translate(%4.5f,%4.5f) rotate(90)\" font-size=\"%3.4fpt\">C%i</text>\n",
+			time2pixels(start), cpu2y(cpu), width, type);
+}
+
+static char *HzToHuman(unsigned long hz)
+{
+	static char buffer[1024];
+	unsigned long long Hz;
+
+	memset(buffer, 0, 1024);
+
+	Hz = hz;
+
+	/* default: just put the Number in */
+	sprintf(buffer, "%9lli", Hz);
+
+	if (Hz > 1000)
+		sprintf(buffer, " %6lli Mhz", (Hz+500)/1000);
+
+	if (Hz > 1500000)
+		sprintf(buffer, " %6.2f Ghz", (Hz+5000.0)/1000000);
+
+	if (Hz == turbo_frequency)
+		sprintf(buffer, "Turbo");
+
+	return buffer;
+}
+
+void svg_pstate(int cpu, u64 start, u64 end, u64 freq)
+{
+	double height = 0;
+
+	if (!svgfile)
+		return;
+
+	if (max_freq)
+		height = freq * 1.0 / max_freq * (SLOT_HEIGHT + SLOT_MULT);
+	height = 1 + cpu2y(cpu) + SLOT_MULT + SLOT_HEIGHT - height;
+	fprintf(svgfile, "<line x1=\"%4.8f\" x2=\"%4.8f\" y1=\"%4.1f\" y2=\"%4.1f\" class=\"pstate\"/>\n",
+		time2pixels(start), time2pixels(end), height, height);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\" font-size=\"0.25pt\">%s</text>\n",
+		time2pixels(start), height+0.9, HzToHuman(freq));
+
+}
+
+
+void svg_partial_wakeline(u64 start, int row1, int row2)
+{
+	double height;
+
+	if (!svgfile)
+		return;
+
+
+	if (row1 < row2) {
+		if (row1)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT + SLOT_MULT/32);
+
+		if (row2)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row2 * SLOT_MULT - SLOT_MULT/32,  time2pixels(start), row2 * SLOT_MULT);
+	} else {
+		if (row2)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT + SLOT_MULT/32);
+
+		if (row1)
+			fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+				time2pixels(start), row1 * SLOT_MULT - SLOT_MULT/32,  time2pixels(start), row1 * SLOT_MULT);
+	}
+	height = row1 * SLOT_MULT;
+	if (row2 > row1)
+		height += SLOT_HEIGHT;
+	if (row1)
+		fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(32,255,32)\"/>\n",
+			time2pixels(start), height);
+}
+
+void svg_wakeline(u64 start, int row1, int row2)
+{
+	double height;
+
+	if (!svgfile)
+		return;
+
+
+	if (row1 < row2)
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+			time2pixels(start), row1 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row2 * SLOT_MULT);
+	else
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%4.2f\" style=\"stroke:rgb(32,255,32);stroke-width:0.009\"/>\n",
+			time2pixels(start), row2 * SLOT_MULT + SLOT_HEIGHT,  time2pixels(start), row1 * SLOT_MULT);
+
+	height = row1 * SLOT_MULT;
+	if (row2 > row1)
+		height += SLOT_HEIGHT;
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(32,255,32)\"/>\n",
+			time2pixels(start), height);
+}
+
+void svg_interrupt(u64 start, int row)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(255,128,128)\"/>\n",
+			time2pixels(start), row * SLOT_MULT);
+	fprintf(svgfile, "<circle  cx=\"%4.8f\" cy=\"%4.2f\" r = \"0.01\"  style=\"fill:rgb(255,128,128)\"/>\n",
+			time2pixels(start), row * SLOT_MULT + SLOT_HEIGHT);
+}
+
+void svg_text(int Yslot, u64 start, const char *text)
+{
+	if (!svgfile)
+		return;
+
+	fprintf(svgfile, "<text transform=\"translate(%4.1f,%4.1f)\">%s</text>\n",
+		time2pixels(start), Yslot * SLOT_MULT+SLOT_HEIGHT/2, text);
+}
+
+static void svg_legenda_box(int X, const char *text, const char *style)
+{
+	double boxsize;
+	boxsize = SLOT_HEIGHT / 2;
+
+	fprintf(svgfile, "<rect x=\"%i\" width=\"%4.8f\" y=\"0\" height=\"%4.1f\" class=\"%s\"/>\n",
+		X, boxsize, boxsize, style);
+	fprintf(svgfile, "<text transform=\"translate(%4.1f, %4.1f)\" font-size=\"%4.4fpt\">%s</text>\n",
+		X + boxsize + 5, boxsize, 0.8 * boxsize, text);
+}
+
+void svg_legenda(void)
+{
+	if (!svgfile)
+		return;
+
+	svg_legenda_box(0,	"Running", "sample");
+	svg_legenda_box(100,	"Idle","rect.c1");
+	svg_legenda_box(200,	"Deeper Idle", "rect.c3");
+	svg_legenda_box(350,	"Deepest Idle", "rect.c6");
+	svg_legenda_box(550,	"Sleeping", "process2");
+	svg_legenda_box(650,	"Waiting for cpu", "waiting");
+	svg_legenda_box(800,	"Blocked on IO", "blocked");
+}
+
+void svg_time_grid(u64 start, u64 end)
+{
+	u64 i;
+
+	first_time = start;
+	last_time = end;
+
+	first_time = first_time / 100000000 * 100000000;
+
+	if (!svgfile)
+		return;
+
+	i = first_time;
+	while (i < last_time) {
+		int color = 220;
+		double thickness = 0.075;
+		if ((i % 100000000) == 0) {
+			thickness = 0.5;
+			color = 192;
+		}
+		if ((i % 1000000000) == 0) {
+			thickness = 2.0;
+			color = 128;
+		}
+
+		fprintf(svgfile, "<line x1=\"%4.8f\" y1=\"%4.2f\" x2=\"%4.8f\" y2=\"%llu\" style=\"stroke:rgb(%i,%i,%i);stroke-width:%1.3f\"/>\n",
+			time2pixels(i), SLOT_MULT/2, time2pixels(i), total_height, color, color, color, thickness);
+
+		i += 10000000;
+	}
+}
+
+void svg_close(void)
+{
+	if (svgfile) {
+		fprintf(svgfile, "</svg>\n");
+		fclose(svgfile);
+		svgfile = NULL;
+	}
+}
diff --git a/tools/perf/util/svghelper.h b/tools/perf/util/svghelper.h
new file mode 100644
index 0000000..ad79b5d
--- /dev/null
+++ b/tools/perf/util/svghelper.h
@@ -0,0 +1,25 @@
+#ifndef _INCLUDE_GUARD_SVG_HELPER_
+#define _INCLUDE_GUARD_SVG_HELPER_
+
+#include "types.h"
+
+extern void open_svg(const char *filename, int cpus, int rows);
+extern void svg_box(int Yslot, u64 start, u64 end, const char *type);
+extern void svg_sample(int Yslot, int cpu, u64 start, u64 end, const char *type);
+extern void svg_cpu_box(int cpu, u64 max_frequency, u64 turbo_frequency);
+
+
+extern void svg_process(int cpu, u64 start, u64 end, const char *type, const char *name);
+extern void svg_cstate(int cpu, u64 start, u64 end, int type);
+extern void svg_pstate(int cpu, u64 start, u64 end, u64 freq);
+
+
+extern void svg_time_grid(u64 start, u64 end);
+extern void svg_legenda(void);
+extern void svg_wakeline(u64 start, int row1, int row2);
+extern void svg_partial_wakeline(u64 start, int row1, int row2);
+extern void svg_interrupt(u64 start, int row);
+extern void svg_text(int Yslot, u64 start, const char *text);
+extern void svg_close(void);
+
+#endif

[tip:perfcounters/core] perf: Add the timechart tool

[tip-bot for Arjan van de Ven]

Commit-ID:  10274989fd595db455874fc2c83272fb33f6b27b
Gitweb:     http://git.kernel.org/tip/10274989fd595db455874fc2c83272fb33f6b27b
Author:     Arjan van de Ven <arjan@linux.intel.com>
AuthorDate: Sat, 12 Sep 2009 07:53:05 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Sat, 19 Sep 2009 11:42:13 +0200

perf: Add the timechart tool

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            |    7 +
 tools/perf/builtin-timechart.c | 1120 ++++++++++++++++++++++++++++++++++++++++
 tools/perf/builtin.h           |    1 +
 tools/perf/perf.c              |    1 +
 4 files changed, 1129 insertions(+), 0 deletions(-)

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 0388e36..0aba8b6 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -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
index 0000000..00fac1b
--- /dev/null
+++ b/tools/perf/builtin-timechart.c
@@ -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();
+}
diff --git a/tools/perf/builtin.h b/tools/perf/builtin.h
index b09cadb..e11d8d2 100644
--- a/tools/perf/builtin.h
+++ b/tools/perf/builtin.h
@@ -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);
diff --git a/tools/perf/perf.c b/tools/perf/perf.c
index c972d1c..19fc7fe 100644
--- a/tools/perf/perf.c
+++ b/tools/perf/perf.c
@@ -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 },

[PATCH] perf: Add "perf timechart record"

[Arjan van de Ven]

>From f11a27881ef0fcdf459e6f52dd57885c7af93426 Mon Sep 17 00:00:00 2001
From: Arjan van de Ven <arjan@linux.intel.com>
Date: Sat, 19 Sep 2009 12:59:47 +0200
Subject: [PATCH] perf: Add "perf timechart record"

Add a command line option to record a trace, similar to "perf sched record".

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
---
 tools/perf/builtin-timechart.c |   47 +++++++++++++++++++++++++++++++++------
 1 files changed, 39 insertions(+), 8 deletions(-)

diff --git a/tools/perf/builtin-timechart.c b/tools/perf/builtin-timechart.c
index 00fac1b..58d737e 100644
--- a/tools/perf/builtin-timechart.c
+++ b/tools/perf/builtin-timechart.c
@@ -1086,11 +1086,42 @@ done:
 	return rc;
 }
 
-static const char * const report_usage[] = {
-	"perf report [<options>] <command>",
+static const char * const timechart_usage[] = {
+	"perf timechart [<options>] {record}",
 	NULL
 };
 
+static const char *record_args[] = {
+	"record",
+	"-a",
+	"-R",
+	"-M",
+	"-f",
+	"-c", "1",
+	"-e", "power:power_start",
+	"-e", "power:power_end",
+	"-e", "power:power_frequency",
+	"-e", "sched:sched_wakeup",
+	"-e", "sched:sched_switch",
+};
+
+static int __cmd_record(int argc, const char **argv)
+{
+	unsigned int rec_argc, i, j;
+	const char **rec_argv;
+
+	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
+	rec_argv = calloc(rec_argc + 1, sizeof(char *));
+
+	for (i = 0; i < ARRAY_SIZE(record_args); i++)
+		rec_argv[i] = strdup(record_args[i]);
+
+	for (j = 1; j < (unsigned int)argc; j++, i++)
+		rec_argv[i] = argv[j];
+
+	return cmd_record(i, rec_argv, NULL);
+}
+
 static const struct option options[] = {
 	OPT_STRING('i', "input", &input_name, "file",
 		    "input file name"),
@@ -1106,13 +1137,13 @@ int cmd_timechart(int argc, const char **argv, const char *prefix __used)
 
 	page_size = getpagesize();
 
-	argc = parse_options(argc, argv, options, report_usage, 0);
+	argc = parse_options(argc, argv, options, timechart_usage,
+			PARSE_OPT_STOP_AT_NON_OPTION);
 
-	/*
-	 * Any (unrecognized) arguments left?
-	 */
-	if (argc)
-		usage_with_options(report_usage, options);
+	if (argc && !strncmp(argv[0], "rec", 3))
+		return __cmd_record(argc, argv);
+	else if (argc)
+		usage_with_options(timechart_usage, options);
 
 	setup_pager();
 
-- 
1.6.0.6



-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org