event_analyzing_sample.py (7497B)
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())]))