Loading...
1# event_analyzing_sample.py: general event handler in python
2# SPDX-License-Identifier: GPL-2.0
3#
4# Current perf report is already very powerful with the annotation integrated,
5# and this script is not trying to be as powerful as perf report, but
6# providing end user/developer a flexible way to analyze the events other
7# than trace points.
8#
9# The 2 database related functions in this script just show how to gather
10# the basic information, and users can modify and write their own functions
11# according to their specific requirement.
12#
13# The first function "show_general_events" just does a basic grouping for all
14# generic events with the help of sqlite, and the 2nd one "show_pebs_ll" is
15# for a x86 HW PMU event: PEBS with load latency data.
16#
17
18from __future__ import print_function
19
20import os
21import sys
22import math
23import struct
24import sqlite3
25
26sys.path.append(os.environ['PERF_EXEC_PATH'] + \
27 '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
28
29from perf_trace_context import *
30from EventClass import *
31
32#
33# If the perf.data has a big number of samples, then the insert operation
34# will be very time consuming (about 10+ minutes for 10000 samples) if the
35# .db database is on disk. Move the .db file to RAM based FS to speedup
36# the handling, which will cut the time down to several seconds.
37#
38con = sqlite3.connect("/dev/shm/perf.db")
39con.isolation_level = None
40
41def trace_begin():
42 print("In trace_begin:\n")
43
44 #
45 # Will create several tables at the start, pebs_ll is for PEBS data with
46 # load latency info, while gen_events is for general event.
47 #
48 con.execute("""
49 create table if not exists gen_events (
50 name text,
51 symbol text,
52 comm text,
53 dso text
54 );""")
55 con.execute("""
56 create table if not exists pebs_ll (
57 name text,
58 symbol text,
59 comm text,
60 dso text,
61 flags integer,
62 ip integer,
63 status integer,
64 dse integer,
65 dla integer,
66 lat integer
67 );""")
68
69#
70# Create and insert event object to a database so that user could
71# do more analysis with simple database commands.
72#
73def process_event(param_dict):
74 event_attr = param_dict["attr"]
75 sample = param_dict["sample"]
76 raw_buf = param_dict["raw_buf"]
77 comm = param_dict["comm"]
78 name = param_dict["ev_name"]
79
80 # Symbol and dso info are not always resolved
81 if ("dso" in param_dict):
82 dso = param_dict["dso"]
83 else:
84 dso = "Unknown_dso"
85
86 if ("symbol" in param_dict):
87 symbol = param_dict["symbol"]
88 else:
89 symbol = "Unknown_symbol"
90
91 # Create the event object and insert it to the right table in database
92 event = create_event(name, comm, dso, symbol, raw_buf)
93 insert_db(event)
94
95def insert_db(event):
96 if event.ev_type == EVTYPE_GENERIC:
97 con.execute("insert into gen_events values(?, ?, ?, ?)",
98 (event.name, event.symbol, event.comm, event.dso))
99 elif event.ev_type == EVTYPE_PEBS_LL:
100 event.ip &= 0x7fffffffffffffff
101 event.dla &= 0x7fffffffffffffff
102 con.execute("insert into pebs_ll values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
103 (event.name, event.symbol, event.comm, event.dso, event.flags,
104 event.ip, event.status, event.dse, event.dla, event.lat))
105
106def trace_end():
107 print("In trace_end:\n")
108 # We show the basic info for the 2 type of event classes
109 show_general_events()
110 show_pebs_ll()
111 con.close()
112
113#
114# As the event number may be very big, so we can't use linear way
115# to show the histogram in real number, but use a log2 algorithm.
116#
117
118def num2sym(num):
119 # Each number will have at least one '#'
120 snum = '#' * (int)(math.log(num, 2) + 1)
121 return snum
122
123def show_general_events():
124
125 # Check the total record number in the table
126 count = con.execute("select count(*) from gen_events")
127 for t in count:
128 print("There is %d records in gen_events table" % t[0])
129 if t[0] == 0:
130 return
131
132 print("Statistics about the general events grouped by thread/symbol/dso: \n")
133
134 # Group by thread
135 commq = con.execute("select comm, count(comm) from gen_events group by comm order by -count(comm)")
136 print("\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42))
137 for row in commq:
138 print("%16s %8d %s" % (row[0], row[1], num2sym(row[1])))
139
140 # Group by symbol
141 print("\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58))
142 symbolq = con.execute("select symbol, count(symbol) from gen_events group by symbol order by -count(symbol)")
143 for row in symbolq:
144 print("%32s %8d %s" % (row[0], row[1], num2sym(row[1])))
145
146 # Group by dso
147 print("\n%40s %8s %16s\n%s" % ("dso", "number", "histogram", "="*74))
148 dsoq = con.execute("select dso, count(dso) from gen_events group by dso order by -count(dso)")
149 for row in dsoq:
150 print("%40s %8d %s" % (row[0], row[1], num2sym(row[1])))
151
152#
153# This function just shows the basic info, and we could do more with the
154# data in the tables, like checking the function parameters when some
155# big latency events happen.
156#
157def show_pebs_ll():
158
159 count = con.execute("select count(*) from pebs_ll")
160 for t in count:
161 print("There is %d records in pebs_ll table" % t[0])
162 if t[0] == 0:
163 return
164
165 print("Statistics about the PEBS Load Latency events grouped by thread/symbol/dse/latency: \n")
166
167 # Group by thread
168 commq = con.execute("select comm, count(comm) from pebs_ll group by comm order by -count(comm)")
169 print("\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42))
170 for row in commq:
171 print("%16s %8d %s" % (row[0], row[1], num2sym(row[1])))
172
173 # Group by symbol
174 print("\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58))
175 symbolq = con.execute("select symbol, count(symbol) from pebs_ll group by symbol order by -count(symbol)")
176 for row in symbolq:
177 print("%32s %8d %s" % (row[0], row[1], num2sym(row[1])))
178
179 # Group by dse
180 dseq = con.execute("select dse, count(dse) from pebs_ll group by dse order by -count(dse)")
181 print("\n%32s %8s %16s\n%s" % ("dse", "number", "histogram", "="*58))
182 for row in dseq:
183 print("%32s %8d %s" % (row[0], row[1], num2sym(row[1])))
184
185 # Group by latency
186 latq = con.execute("select lat, count(lat) from pebs_ll group by lat order by lat")
187 print("\n%32s %8s %16s\n%s" % ("latency", "number", "histogram", "="*58))
188 for row in latq:
189 print("%32s %8d %s" % (row[0], row[1], num2sym(row[1])))
190
191def trace_unhandled(event_name, context, event_fields_dict):
192 print (' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())]))
1# event_analyzing_sample.py: general event handler in python
2# SPDX-License-Identifier: GPL-2.0
3#
4# Current perf report is already very powerful with the annotation integrated,
5# and this script is not trying to be as powerful as perf report, but
6# providing end user/developer a flexible way to analyze the events other
7# than trace points.
8#
9# The 2 database related functions in this script just show how to gather
10# the basic information, and users can modify and write their own functions
11# according to their specific requirement.
12#
13# The first function "show_general_events" just does a basic grouping for all
14# generic events with the help of sqlite, and the 2nd one "show_pebs_ll" is
15# for a x86 HW PMU event: PEBS with load latency data.
16#
17
18import os
19import sys
20import math
21import struct
22import sqlite3
23
24sys.path.append(os.environ['PERF_EXEC_PATH'] + \
25 '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
26
27from perf_trace_context import *
28from EventClass import *
29
30#
31# If the perf.data has a big number of samples, then the insert operation
32# will be very time consuming (about 10+ minutes for 10000 samples) if the
33# .db database is on disk. Move the .db file to RAM based FS to speedup
34# the handling, which will cut the time down to several seconds.
35#
36con = sqlite3.connect("/dev/shm/perf.db")
37con.isolation_level = None
38
39def trace_begin():
40 print "In trace_begin:\n"
41
42 #
43 # Will create several tables at the start, pebs_ll is for PEBS data with
44 # load latency info, while gen_events is for general event.
45 #
46 con.execute("""
47 create table if not exists gen_events (
48 name text,
49 symbol text,
50 comm text,
51 dso text
52 );""")
53 con.execute("""
54 create table if not exists pebs_ll (
55 name text,
56 symbol text,
57 comm text,
58 dso text,
59 flags integer,
60 ip integer,
61 status integer,
62 dse integer,
63 dla integer,
64 lat integer
65 );""")
66
67#
68# Create and insert event object to a database so that user could
69# do more analysis with simple database commands.
70#
71def process_event(param_dict):
72 event_attr = param_dict["attr"]
73 sample = param_dict["sample"]
74 raw_buf = param_dict["raw_buf"]
75 comm = param_dict["comm"]
76 name = param_dict["ev_name"]
77
78 # Symbol and dso info are not always resolved
79 if (param_dict.has_key("dso")):
80 dso = param_dict["dso"]
81 else:
82 dso = "Unknown_dso"
83
84 if (param_dict.has_key("symbol")):
85 symbol = param_dict["symbol"]
86 else:
87 symbol = "Unknown_symbol"
88
89 # Create the event object and insert it to the right table in database
90 event = create_event(name, comm, dso, symbol, raw_buf)
91 insert_db(event)
92
93def insert_db(event):
94 if event.ev_type == EVTYPE_GENERIC:
95 con.execute("insert into gen_events values(?, ?, ?, ?)",
96 (event.name, event.symbol, event.comm, event.dso))
97 elif event.ev_type == EVTYPE_PEBS_LL:
98 event.ip &= 0x7fffffffffffffff
99 event.dla &= 0x7fffffffffffffff
100 con.execute("insert into pebs_ll values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
101 (event.name, event.symbol, event.comm, event.dso, event.flags,
102 event.ip, event.status, event.dse, event.dla, event.lat))
103
104def trace_end():
105 print "In trace_end:\n"
106 # We show the basic info for the 2 type of event classes
107 show_general_events()
108 show_pebs_ll()
109 con.close()
110
111#
112# As the event number may be very big, so we can't use linear way
113# to show the histogram in real number, but use a log2 algorithm.
114#
115
116def num2sym(num):
117 # Each number will have at least one '#'
118 snum = '#' * (int)(math.log(num, 2) + 1)
119 return snum
120
121def show_general_events():
122
123 # Check the total record number in the table
124 count = con.execute("select count(*) from gen_events")
125 for t in count:
126 print "There is %d records in gen_events table" % t[0]
127 if t[0] == 0:
128 return
129
130 print "Statistics about the general events grouped by thread/symbol/dso: \n"
131
132 # Group by thread
133 commq = con.execute("select comm, count(comm) from gen_events group by comm order by -count(comm)")
134 print "\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42)
135 for row in commq:
136 print "%16s %8d %s" % (row[0], row[1], num2sym(row[1]))
137
138 # Group by symbol
139 print "\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58)
140 symbolq = con.execute("select symbol, count(symbol) from gen_events group by symbol order by -count(symbol)")
141 for row in symbolq:
142 print "%32s %8d %s" % (row[0], row[1], num2sym(row[1]))
143
144 # Group by dso
145 print "\n%40s %8s %16s\n%s" % ("dso", "number", "histogram", "="*74)
146 dsoq = con.execute("select dso, count(dso) from gen_events group by dso order by -count(dso)")
147 for row in dsoq:
148 print "%40s %8d %s" % (row[0], row[1], num2sym(row[1]))
149
150#
151# This function just shows the basic info, and we could do more with the
152# data in the tables, like checking the function parameters when some
153# big latency events happen.
154#
155def show_pebs_ll():
156
157 count = con.execute("select count(*) from pebs_ll")
158 for t in count:
159 print "There is %d records in pebs_ll table" % t[0]
160 if t[0] == 0:
161 return
162
163 print "Statistics about the PEBS Load Latency events grouped by thread/symbol/dse/latency: \n"
164
165 # Group by thread
166 commq = con.execute("select comm, count(comm) from pebs_ll group by comm order by -count(comm)")
167 print "\n%16s %8s %16s\n%s" % ("comm", "number", "histogram", "="*42)
168 for row in commq:
169 print "%16s %8d %s" % (row[0], row[1], num2sym(row[1]))
170
171 # Group by symbol
172 print "\n%32s %8s %16s\n%s" % ("symbol", "number", "histogram", "="*58)
173 symbolq = con.execute("select symbol, count(symbol) from pebs_ll group by symbol order by -count(symbol)")
174 for row in symbolq:
175 print "%32s %8d %s" % (row[0], row[1], num2sym(row[1]))
176
177 # Group by dse
178 dseq = con.execute("select dse, count(dse) from pebs_ll group by dse order by -count(dse)")
179 print "\n%32s %8s %16s\n%s" % ("dse", "number", "histogram", "="*58)
180 for row in dseq:
181 print "%32s %8d %s" % (row[0], row[1], num2sym(row[1]))
182
183 # Group by latency
184 latq = con.execute("select lat, count(lat) from pebs_ll group by lat order by lat")
185 print "\n%32s %8s %16s\n%s" % ("latency", "number", "histogram", "="*58)
186 for row in latq:
187 print "%32s %8d %s" % (row[0], row[1], num2sym(row[1]))
188
189def trace_unhandled(event_name, context, event_fields_dict):
190 print ' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())])