tools/perf/scripts/python/event_analyzing_sample.py


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189

# event_analyzing_sample.py: general event handler in python
#
# Current perf report is already very powerful with the annotation integrated,
# and this script is not trying to be as powerful as perf report, but
# providing end user/developer a flexible way to analyze the events other
# than trace points.
#
# The 2 database related functions in this script just show how to gather
# the basic information, and users can modify and write their own functions
# according to their specific requirement.
#
# The first function "show_general_events" just does a basic grouping for all
# generic events with the help of sqlite, and the 2nd one "show_pebs_ll" is
# for a x86 HW PMU event: PEBS with load latency data.
#

import os
import sys
import math
import struct
import sqlite3

sys.path.append(os.environ['PERF_EXEC_PATH'] + \
        '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')

from perf_trace_context import *
from EventClass import *

#
# If the perf.data has a big number of samples, then the insert operation
# will be very time consuming (about 10+ minutes for 10000 samples) if the
# .db database is on disk. Move the .db file to RAM based FS to speedup
# the handling, which will cut the time down to several seconds.
#
con = sqlite3.connect("/dev/shm/perf.db")
con.isolation_level = None

def trace_begin():
	print "In trace_begin:\n"

        #
        # Will create several tables at the start, pebs_ll is for PEBS data with
        # load latency info, while gen_events is for general event.
        #
        con.execute("""
                create table if not exists gen_events (
                        name text,
                        symbol text,
                        comm text,
                        dso text
                );""")
        con.execute("""
                create table if not exists pebs_ll (
                        name text,
                        symbol text,
                        comm text,
                        dso text,
                        flags integer,
                        ip integer,
                        status integer,
                        dse integer,
                        dla integer,
                        lat integer
                );""")

#
# Create and insert event object to a database so that user could
# do more analysis with simple database commands.
#
def process_event(param_dict):
        event_attr = param_dict["attr"]
        sample     = param_dict["sample"]
        raw_buf    = param_dict["raw_buf"]
        comm       = param_dict["comm"]
        name       = param_dict["ev_name"]

        # Symbol and dso info are not always resolved
        if (param_dict.has_key("dso")):
                dso = param_dict["dso"]
        else:
                dso = "Unknown_dso"

        if (param_dict.has_key("symbol")):
                symbol = param_dict["symbol"]
        else:
                symbol = "Unknown_symbol"

        # Create the event object and insert it to the right table in database
        event = create_event(name, comm, dso, symbol, raw_buf)
        insert_db(event)

def insert_db(event):
        if event.ev_type == EVTYPE_GENERIC:
                con.execute("insert into gen_events values(?, ?, ?, ?)",
                                (event.name, event.symbol, event.comm, event.dso))
        elif event.ev_type == EVTYPE_PEBS_LL:
                event.ip &= 0x7fffffffffffffff
                event.dla &= 0x7fffffffffffffff
                con.execute("insert into pebs_ll values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
                        (event.name, event.symbol, event.comm, event.dso, event.flags,
                                event.ip, event.status, event.dse, event.dla, event.lat))

def trace_end():
	print "In trace_end:\n"
        # We show the basic info for the 2 type of event classes
        show_general_events()
        show_pebs_ll()
        con.close()

#
# As the event number may be very big, so we can't use linear way
# to show the histogram in real number, but use a log2 algorithm.
#

def num2sym(num):
        # Each number will have at least one '#'
        snum = '#' * (int)(math.log(num, 2) + 1)
        return snum

def show_general_events():

        # Check the total record number in the table
        count = con.execute("select count(*) from gen_events")
        for t in count:
                print "There is %d records in gen_events table" % t[0]
                if t[0] == 0:
                        return

        print "Statistics about the general events grouped by thread/symbol/dso: \n"

         # Group by thread
        commq = con.execute("select comm, count(comm) from gen_events group by comm order by -count(comm)")
        print "\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42)
        for row in commq:
             print "%16s %8d     %s" % (row[0], row[1], num2sym(row[1]))

        # Group by symbol
        print "\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58)
        symbolq = con.execute("select symbol, count(symbol) from gen_events group by symbol order by -count(symbol)")
        for row in symbolq:
             print "%32s %8d     %s" % (row[0], row[1], num2sym(row[1]))

        # Group by dso
        print "\n%40s %8s %16s\n%s" % ("dso", "number", "histogram", "="*74)
        dsoq = con.execute("select dso, count(dso) from gen_events group by dso order by -count(dso)")
        for row in dsoq:
             print "%40s %8d     %s" % (row[0], row[1], num2sym(row[1]))

#
# This function just shows the basic info, and we could do more with the
# data in the tables, like checking the function parameters when some
# big latency events happen.
#
def show_pebs_ll():

        count = con.execute("select count(*) from pebs_ll")
        for t in count:
                print "There is %d records in pebs_ll table" % t[0]
                if t[0] == 0:
                        return

        print "Statistics about the PEBS Load Latency events grouped by thread/symbol/dse/latency: \n"

        # Group by thread
        commq = con.execute("select comm, count(comm) from pebs_ll group by comm order by -count(comm)")
        print "\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42)
        for row in commq:
             print "%16s %8d     %s" % (row[0], row[1], num2sym(row[1]))

        # Group by symbol
        print "\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58)
        symbolq = con.execute("select symbol, count(symbol) from pebs_ll group by symbol order by -count(symbol)")
        for row in symbolq:
             print "%32s %8d     %s" % (row[0], row[1], num2sym(row[1]))

        # Group by dse
        dseq = con.execute("select dse, count(dse) from pebs_ll group by dse order by -count(dse)")
        print "\n%32s %8s %16s\n%s" % ("dse", "number", "histogram", "="*58)
        for row in dseq:
             print "%32s %8d     %s" % (row[0], row[1], num2sym(row[1]))

        # Group by latency
        latq = con.execute("select lat, count(lat) from pebs_ll group by lat order by lat")
        print "\n%32s %8s %16s\n%s" % ("latency", "number", "histogram", "="*58)
        for row in latq:
             print "%32s %8d     %s" % (row[0], row[1], num2sym(row[1]))

def trace_unhandled(event_name, context, event_fields_dict):
		print ' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())])
#ifndef _LINUX_CGROUP_H
#define _LINUX_CGROUP_H
/*
 *  cgroup interface
 *
 *  Copyright (C) 2003 BULL SA
 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
 *
 */

#include <linux/sched.h>
#include <linux/cpumask.h>
#include <linux/nodemask.h>
#include <linux/rculist.h>
#include <linux/cgroupstats.h>
#include <linux/rwsem.h>
#include <linux/fs.h>
#include <linux/seq_file.h>
#include <linux/kernfs.h>

#include <linux/cgroup-defs.h>

#ifdef CONFIG_CGROUPS

/* a css_task_iter should be treated as an opaque object */
struct css_task_iter {
	struct cgroup_subsys		*ss;

	struct list_head		*cset_pos;
	struct list_head		*cset_head;

	struct list_head		*task_pos;
	struct list_head		*tasks_head;
	struct list_head		*mg_tasks_head;
};

extern struct cgroup_root cgrp_dfl_root;
extern struct css_set init_css_set;

#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
#include <linux/cgroup_subsys.h>
#undef SUBSYS

bool css_has_online_children(struct cgroup_subsys_state *css);
struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
					     struct cgroup_subsys *ss);
struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
						       struct cgroup_subsys *ss);

bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);

int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
int cgroup_rm_cftypes(struct cftype *cfts);

char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
		     struct pid *pid, struct task_struct *tsk);

void cgroup_fork(struct task_struct *p);
void cgroup_post_fork(struct task_struct *p);
void cgroup_exit(struct task_struct *p);

int cgroup_init_early(void);
int cgroup_init(void);

/*
 * Iteration helpers and macros.
 */

struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
					   struct cgroup_subsys_state *parent);
struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
						    struct cgroup_subsys_state *css);
struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
						     struct cgroup_subsys_state *css);

struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);

void css_task_iter_start(struct cgroup_subsys_state *css,
			 struct css_task_iter *it);
struct task_struct *css_task_iter_next(struct css_task_iter *it);
void css_task_iter_end(struct css_task_iter *it);

/**
 * css_for_each_child - iterate through children of a css
 * @pos: the css * to use as the loop cursor
 * @parent: css whose children to walk
 *
 * Walk @parent's children.  Must be called under rcu_read_lock().
 *
 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 * css which finished ->css_online() is guaranteed to be visible in the
 * future iterations and will stay visible until the last reference is put.
 * A css which hasn't finished ->css_online() or already finished
 * ->css_offline() may show up during traversal.  It's each subsystem's
 * responsibility to synchronize against on/offlining.
 *
 * It is allowed to temporarily drop RCU read lock during iteration.  The
 * caller is responsible for ensuring that @pos remains accessible until
 * the start of the next iteration by, for example, bumping the css refcnt.
 */
#define css_for_each_child(pos, parent)					\
	for ((pos) = css_next_child(NULL, (parent)); (pos);		\
	     (pos) = css_next_child((pos), (parent)))

/**
 * css_for_each_descendant_pre - pre-order walk of a css's descendants
 * @pos: the css * to use as the loop cursor
 * @root: css whose descendants to walk
 *
 * Walk @root's descendants.  @root is included in the iteration and the
 * first node to be visited.  Must be called under rcu_read_lock().
 *
 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 * css which finished ->css_online() is guaranteed to be visible in the
 * future iterations and will stay visible until the last reference is put.
 * A css which hasn't finished ->css_online() or already finished
 * ->css_offline() may show up during traversal.  It's each subsystem's
 * responsibility to synchronize against on/offlining.
 *
 * For example, the following guarantees that a descendant can't escape
 * state updates of its ancestors.
 *
 * my_online(@css)
 * {
 *	Lock @css's parent and @css;
 *	Inherit state from the parent;
 *	Unlock both.
 * }
 *
 * my_update_state(@css)
 * {
 *	css_for_each_descendant_pre(@pos, @css) {
 *		Lock @pos;
 *		if (@pos == @css)
 *			Update @css's state;
 *		else
 *			Verify @pos is alive and inherit state from its parent;
 *		Unlock @pos;
 *	}
 * }
 *
 * As long as the inheriting step, including checking the parent state, is
 * enclosed inside @pos locking, double-locking the parent isn't necessary
 * while inheriting.  The state update to the parent is guaranteed to be
 * visible by walking order and, as long as inheriting operations to the
 * same @pos are atomic to each other, multiple updates racing each other
 * still result in the correct state.  It's guaranateed that at least one
 * inheritance happens for any css after the latest update to its parent.
 *
 * If checking parent's state requires locking the parent, each inheriting
 * iteration should lock and unlock both @pos->parent and @pos.
 *
 * Alternatively, a subsystem may choose to use a single global lock to
 * synchronize ->css_online() and ->css_offline() against tree-walking
 * operations.
 *
 * It is allowed to temporarily drop RCU read lock during iteration.  The
 * caller is responsible for ensuring that @pos remains accessible until
 * the start of the next iteration by, for example, bumping the css refcnt.
 */
#define css_for_each_descendant_pre(pos, css)				\
	for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);	\
	     (pos) = css_next_descendant_pre((pos), (css)))

/**
 * css_for_each_descendant_post - post-order walk of a css's descendants
 * @pos: the css * to use as the loop cursor
 * @css: css whose descendants to walk
 *
 * Similar to css_for_each_descendant_pre() but performs post-order
 * traversal instead.  @root is included in the iteration and the last
 * node to be visited.
 *
 * If a subsystem synchronizes ->css_online() and the start of iteration, a
 * css which finished ->css_online() is guaranteed to be visible in the
 * future iterations and will stay visible until the last reference is put.
 * A css which hasn't finished ->css_online() or already finished
 * ->css_offline() may show up during traversal.  It's each subsystem's
 * responsibility to synchronize against on/offlining.
 *
 * Note that the walk visibility guarantee example described in pre-order
 * walk doesn't apply the same to post-order walks.
 */
#define css_for_each_descendant_post(pos, css)				\
	for ((pos) = css_next_descendant_post(NULL, (css)); (pos);	\
	     (pos) = css_next_descendant_post((pos), (css)))

/**
 * cgroup_taskset_for_each - iterate cgroup_taskset
 * @task: the loop cursor
 * @tset: taskset to iterate
 */
#define cgroup_taskset_for_each(task, tset)				\
	for ((task) = cgroup_taskset_first((tset)); (task);		\
	     (task) = cgroup_taskset_next((tset)))

/*
 * Inline functions.
 */

/**
 * css_get - obtain a reference on the specified css
 * @css: target css
 *
 * The caller must already have a reference.
 */
static inline void css_get(struct cgroup_subsys_state *css)
{
	if (!(css->flags & CSS_NO_REF))
		percpu_ref_get(&css->refcnt);
}

/**
 * css_get_many - obtain references on the specified css
 * @css: target css
 * @n: number of references to get
 *
 * The caller must already have a reference.
 */
static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
{
	if (!(css->flags & CSS_NO_REF))
		percpu_ref_get_many(&css->refcnt, n);
}

/**
 * css_tryget - try to obtain a reference on the specified css
 * @css: target css
 *
 * Obtain a reference on @css unless it already has reached zero and is
 * being released.  This function doesn't care whether @css is on or
 * offline.  The caller naturally needs to ensure that @css is accessible
 * but doesn't have to be holding a reference on it - IOW, RCU protected
 * access is good enough for this function.  Returns %true if a reference
 * count was successfully obtained; %false otherwise.
 */
static inline bool css_tryget(struct cgroup_subsys_state *css)
{
	if (!(css->flags & CSS_NO_REF))
		return percpu_ref_tryget(&css->refcnt);
	return true;
}

/**
 * css_tryget_online - try to obtain a reference on the specified css if online
 * @css: target css
 *
 * Obtain a reference on @css if it's online.  The caller naturally needs
 * to ensure that @css is accessible but doesn't have to be holding a
 * reference on it - IOW, RCU protected access is good enough for this
 * function.  Returns %true if a reference count was successfully obtained;
 * %false otherwise.
 */
static inline bool css_tryget_online(struct cgroup_subsys_state *css)
{
	if (!(css->flags & CSS_NO_REF))
		return percpu_ref_tryget_live(&css->refcnt);
	return true;
}

/**
 * css_put - put a css reference
 * @css: target css
 *
 * Put a reference obtained via css_get() and css_tryget_online().
 */
static inline void css_put(struct cgroup_subsys_state *css)
{
	if (!(css->flags & CSS_NO_REF))
		percpu_ref_put(&css->refcnt);
}

/**
 * css_put_many - put css references
 * @css: target css
 * @n: number of references to put
 *
 * Put references obtained via css_get() and css_tryget_online().
 */
static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
{
	if (!(css->flags & CSS_NO_REF))
		percpu_ref_put_many(&css->refcnt, n);
}

/**
 * task_css_set_check - obtain a task's css_set with extra access conditions
 * @task: the task to obtain css_set for
 * @__c: extra condition expression to be passed to rcu_dereference_check()
 *
 * A task's css_set is RCU protected, initialized and exited while holding
 * task_lock(), and can only be modified while holding both cgroup_mutex
 * and task_lock() while the task is alive.  This macro verifies that the
 * caller is inside proper critical section and returns @task's css_set.
 *
 * The caller can also specify additional allowed conditions via @__c, such
 * as locks used during the cgroup_subsys::attach() methods.
 */
#ifdef CONFIG_PROVE_RCU
extern struct mutex cgroup_mutex;
extern struct rw_semaphore css_set_rwsem;
#define task_css_set_check(task, __c)					\
	rcu_dereference_check((task)->cgroups,				\
		lockdep_is_held(&cgroup_mutex) ||			\
		lockdep_is_held(&css_set_rwsem) ||			\
		((task)->flags & PF_EXITING) || (__c))
#else
#define task_css_set_check(task, __c)					\
	rcu_dereference((task)->cgroups)
#endif

/**
 * task_css_check - obtain css for (task, subsys) w/ extra access conds
 * @task: the target task
 * @subsys_id: the target subsystem ID
 * @__c: extra condition expression to be passed to rcu_dereference_check()
 *
 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
 * synchronization rules are the same as task_css_set_check().
 */
#define task_css_check(task, subsys_id, __c)				\
	task_css_set_check((task), (__c))->subsys[(subsys_id)]

/**
 * task_css_set - obtain a task's css_set
 * @task: the task to obtain css_set for
 *
 * See task_css_set_check().
 */
static inline struct css_set *task_css_set(struct task_struct *task)
{
	return task_css_set_check(task, false);
}

/**
 * task_css - obtain css for (task, subsys)
 * @task: the target task
 * @subsys_id: the target subsystem ID
 *
 * See task_css_check().
 */
static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
						   int subsys_id)
{
	return task_css_check(task, subsys_id, false);
}

/**
 * task_get_css - find and get the css for (task, subsys)
 * @task: the target task
 * @subsys_id: the target subsystem ID
 *
 * Find the css for the (@task, @subsys_id) combination, increment a
 * reference on and return it.  This function is guaranteed to return a
 * valid css.
 */
static inline struct cgroup_subsys_state *
task_get_css(struct task_struct *task, int subsys_id)
{
	struct cgroup_subsys_state *css;

	rcu_read_lock();
	while (true) {
		css = task_css(task, subsys_id);
		if (likely(css_tryget_online(css)))
			break;
		cpu_relax();
	}
	rcu_read_unlock();
	return css;
}

/**
 * task_css_is_root - test whether a task belongs to the root css
 * @task: the target task
 * @subsys_id: the target subsystem ID
 *
 * Test whether @task belongs to the root css on the specified subsystem.
 * May be invoked in any context.
 */
static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
{
	return task_css_check(task, subsys_id, true) ==
		init_css_set.subsys[subsys_id];
}

static inline struct cgroup *task_cgroup(struct task_struct *task,
					 int subsys_id)
{
	return task_css(task, subsys_id)->cgroup;
}

/**
 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
 * @cgrp: the cgroup of interest
 *
 * The default hierarchy is the v2 interface of cgroup and this function
 * can be used to test whether a cgroup is on the default hierarchy for
 * cases where a subsystem should behave differnetly depending on the
 * interface version.
 *
 * The set of behaviors which change on the default hierarchy are still
 * being determined and the mount option is prefixed with __DEVEL__.
 *
 * List of changed behaviors:
 *
 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
 *   and "name" are disallowed.
 *
 * - When mounting an existing superblock, mount options should match.
 *
 * - Remount is disallowed.
 *
 * - rename(2) is disallowed.
 *
 * - "tasks" is removed.  Everything should be at process granularity.  Use
 *   "cgroup.procs" instead.
 *
 * - "cgroup.procs" is not sorted.  pids will be unique unless they got
 *   recycled inbetween reads.
 *
 * - "release_agent" and "notify_on_release" are removed.  Replacement
 *   notification mechanism will be implemented.
 *
 * - "cgroup.clone_children" is removed.
 *
 * - "cgroup.subtree_populated" is available.  Its value is 0 if the cgroup
 *   and its descendants contain no task; otherwise, 1.  The file also
 *   generates kernfs notification which can be monitored through poll and
 *   [di]notify when the value of the file changes.
 *
 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
 *   take masks of ancestors with non-empty cpus/mems, instead of being
 *   moved to an ancestor.
 *
 * - cpuset: a task can be moved into an empty cpuset, and again it takes
 *   masks of ancestors.
 *
 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
 *   is not created.
 *
 * - blkcg: blk-throttle becomes properly hierarchical.
 *
 * - debug: disallowed on the default hierarchy.
 */
static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
{
	return cgrp->root == &cgrp_dfl_root;
}

/* no synchronization, the result can only be used as a hint */
static inline bool cgroup_has_tasks(struct cgroup *cgrp)
{
	return !list_empty(&cgrp->cset_links);
}

/* returns ino associated with a cgroup */
static inline ino_t cgroup_ino(struct cgroup *cgrp)
{
	return cgrp->kn->ino;
}

/* cft/css accessors for cftype->write() operation */
static inline struct cftype *of_cft(struct kernfs_open_file *of)
{