blob: 2ac83931cd7ba7b9ab316c1aad6689e7981374e9 [file] [log] [blame]
# Copyright 2013 The Chromium Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
# Use: ../ `ls cyglog.* -Sr` > merged_cyglog
""""Merge multiple logs files from different processes into a single log.
Given two log files of execution traces, merge the traces into a single trace.
Merging will use timestamps (i.e. the first two columns of logged calls) to
create a single log that is an ordered trace of calls by both processes.
import optparse
import string
import sys
def ParseLogLines(lines):
"""Parse log file lines.
lines: lines from log file produced by profiled run
Below is an example of a small log file:
5086e000-52e92000 r-xp 00000000 b3:02 51276
secs usecs pid:threadid func
1314897086 795828 3587:1074648168 0x509e105c
1314897086 795874 3587:1074648168 0x509e0eb4
1314897086 796326 3587:1074648168 0x509e0e3c
1314897086 796552 3587:1074648168 0x509e07bc
tuple conisiting of 1) an ordered list of the logged calls, as an array of
fields, 2) the virtual start address of the library, used to compute the
offset of the symbol in the library and 3) the virtual end address
call_lines = []
vm_start = 0
vm_end = 0
dash_index = lines[0].find ('-')
space_index = lines[0].find (' ')
vm_start = int (lines[0][:dash_index], 16)
vm_end = int (lines[0][dash_index+1:space_index], 16)
for line in lines[2:]:
line = line.strip()
fields = line.split()
call_lines.append (fields)
return (call_lines, vm_start, vm_end)
def HasDuplicates(calls):
"""Makes sure that calls are only logged once.
calls: list of calls logged
boolean indicating if calls has duplicate calls
seen = set([])
for call in calls:
if call[3] in seen:
return True
return False
def CheckTimestamps(calls):
"""Prints warning to stderr if the call timestamps are not in order.
calls: list of calls logged
index = 0
last_timestamp_secs = -1
last_timestamp_us = -1
while (index < len (calls)):
timestamp_secs = int (calls[index][0])
timestamp_us = int (calls[index][1])
timestamp = (timestamp_secs * 1000000) + timestamp_us
last_timestamp = (last_timestamp_secs * 1000000) + last_timestamp_us
if (timestamp < last_timestamp):
raise Exception("last_timestamp: " + str(last_timestamp_secs)
+ " " + str(last_timestamp_us) + " timestamp: "
+ str(timestamp_secs) + " " + str(timestamp_us) + "\n")
last_timestamp_secs = timestamp_secs
last_timestamp_us = timestamp_us
index = index + 1
def Convert(call_lines, start_address, end_address):
"""Converts the call addresses to static offsets and removes invalid calls.
Removes profiled calls not in shared library using start and end virtual
addresses, converts strings to integer values, coverts virtual addresses to
address in shared library.
list of calls as tuples (sec, usec, pid:tid, callee)
converted_calls = []
call_addresses = set()
for fields in call_lines:
secs = int (fields[0])
usecs = int (fields[1])
callee = int (fields[3], 16)
# Eliminate repetitions of the same function.
if callee in call_addresses:
# Eliminate small addresses. It should be safe to do so because these point
# before the .text section (it is in .plt or earlier).
# TODO(pasko): understand why __cyg_profile_func_enter may output a small
# offset sometimes.
if callee < start_address + 4096:
sys.stderr.write('WARNING: ignoring small address: %s' %
hex(callee - start_address))
if start_address <= callee < end_address:
converted_calls.append((secs, usecs, fields[2], (callee - start_address)))
return converted_calls
def Timestamp(trace_entry):
return int (trace_entry[0]) * 1000000 + int(trace_entry[1])
def AddTrace (tracemap, trace):
"""Adds a trace to the tracemap.
Adds entries in the trace to the tracemap. All new calls will be added to
the tracemap. If the calls already exist in the tracemap then they will be
replaced if they happened sooner in the new trace.
tracemap: the tracemap
trace: the trace
for trace_entry in trace:
call = trace_entry[3]
if (not call in tracemap) or (
Timestamp(tracemap[call]) > Timestamp(trace_entry)):
tracemap[call] = trace_entry
def GroupByProcessAndThreadId(input_trace):
"""Returns an array of traces grouped by pid and tid.
This is used to make the order of functions not depend on thread scheduling
which can be greatly impacted when profiling is done with cygprofile. As a
result each thread has its own contiguous segment of code (ordered by
timestamp) and processes also have their code isolated (i.e. not interleaved).
def MakeTimestamp(sec, usec):
return sec * 1000000 + usec
def PidAndTidFromString(pid_and_tid):
strings = pid_and_tid.split(':')
return (int(strings[0]), int(strings[1]))
tid_to_pid_map = {}
pid_first_seen = {}
tid_first_seen = {}
for (sec, usec, pid_and_tid, _) in input_trace:
(pid, tid) = PidAndTidFromString(pid_and_tid)
# Make sure that thread IDs are unique since this is a property we rely on.
if tid_to_pid_map.setdefault(tid, pid) != pid:
raise Exception(
'Seen PIDs %d and %d for TID=%d. Thread-IDs must be unique' % (
tid_to_pid_map[tid], pid, tid))
if not pid in pid_first_seen:
pid_first_seen[pid] = MakeTimestamp(sec, usec)
if not tid in tid_first_seen:
tid_first_seen[tid] = MakeTimestamp(sec, usec)
def CompareEvents(event1, event2):
(sec1, usec1, pid_and_tid, _) = event1
(pid1, tid1) = PidAndTidFromString(pid_and_tid)
(sec2, usec2, pid_and_tid, _) = event2
(pid2, tid2) = PidAndTidFromString(pid_and_tid)
pid_cmp = cmp(pid_first_seen[pid1], pid_first_seen[pid2])
if pid_cmp != 0:
return pid_cmp
tid_cmp = cmp(tid_first_seen[tid1], tid_first_seen[tid2])
if tid_cmp != 0:
return tid_cmp
return cmp(MakeTimestamp(sec1, usec1), MakeTimestamp(sec2, usec2))
return sorted(input_trace, cmp=CompareEvents)
def Main():
"""Merge two traces for code in specified library and write to stdout.
Merges the two traces and coverts the virtual addresses to the offsets in the
library. First line of merged trace has dummy virtual address of 0-ffffffff
so that symbolizing the addresses uses the addresses in the log, since the
addresses have already been converted to static offsets.
parser = optparse.OptionParser('usage: %prog trace1 ... traceN')
(_, args) = parser.parse_args()
if len(args) <= 1:
parser.error('expected at least the following args: trace1 trace2')
step = 0
# Maps function addresses to their corresponding trace entry.
tracemap = dict()
for trace_file in args:
step += 1
sys.stderr.write(" " + str(step) + "/" + str(len(args)) +
": " + trace_file + ":\n")
trace_lines = map(string.rstrip, open(trace_file).readlines())
(trace_calls, trace_start, trace_end) = ParseLogLines(trace_lines)
sys.stderr.write("Len: " + str(len(trace_calls)) +
". Start: " + hex(trace_start) +
", end: " + hex(trace_end) + '\n')
trace_calls = Convert(trace_calls, trace_start, trace_end)
sys.stderr.write("Converted len: " + str(len(trace_calls)) + "\n")
AddTrace(tracemap, trace_calls)
sys.stderr.write("Merged len: " + str(len(tracemap)) + "\n")
# Extract the resulting trace from the tracemap
merged_trace = []
for call in tracemap:
grouped_trace = GroupByProcessAndThreadId(merged_trace)
print "0-ffffffff r-xp 00000000 xx:00 00000 ./"
print "secs\tusecs\tpid:threadid\tfunc"
for call in grouped_trace:
print (str(call[0]) + "\t" + str(call[1]) + "\t" + call[2] + "\t" +
if __name__ == '__main__':