| 1 |
#!/usr/bin/env python |
| 2 |
# |
| 3 |
# Copyright (C) 2017 The Android Open Source Project |
| 4 |
# |
| 5 |
# Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 |
# you may not use this file except in compliance with the License. |
| 7 |
# You may obtain a copy of the License at |
| 8 |
# |
| 9 |
# http://www.apache.org/licenses/LICENSE-2.0 |
| 10 |
# |
| 11 |
# Unless required by applicable law or agreed to in writing, software |
| 12 |
# distributed under the License is distributed on an "AS IS" BASIS, |
| 13 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 |
# See the License for the specific language governing permissions and |
| 15 |
# limitations under the License. |
| 16 |
# |
| 17 |
|
| 18 |
"""debug_unwind_reporter.py: report dwarf unwinding results generated by debug-unwind cmd. |
| 19 |
Below is an example using debug_unwind_reporter.py: |
| 20 |
1. Record with "-g --no-unwind" option on device. |
| 21 |
simpleperf record -g --no-unwind --app com.google.sample.tunnel --duration 10 |
| 22 |
2. Use debug-unwind cmd to unwind samples in perf.data on device. |
| 23 |
simpleperf debug-unwind |
| 24 |
3. Pull perf.data.debug on host, and report it with this script. |
| 25 |
python debug_unwind_reporter.py |
| 26 |
|
| 27 |
It reports below items: |
| 28 |
1. time used for offline dwarf unwinding for each sample. |
| 29 |
2. mem consumption before and after unwinding samples. |
| 30 |
3. For samples that don't have complete callchains, report their regs, stack data, |
| 31 |
unwinding failure info, callchains. Samples having the same failure reason |
| 32 |
in the same function are only reported once. |
| 33 |
|
| 34 |
It can be used to: |
| 35 |
1. show unwinding failures. |
| 36 |
2. show performance and memory consumption change caused by simpleperf/unwinder changes. |
| 37 |
""" |
| 38 |
|
| 39 |
from __future__ import print_function |
| 40 |
import argparse |
| 41 |
import bisect |
| 42 |
import collections |
| 43 |
import copy |
| 44 |
import re |
| 45 |
import subprocess |
| 46 |
|
| 47 |
from utils import bytes_to_str, log_exit, log_fatal, get_host_binary_path |
| 48 |
|
| 49 |
|
| 50 |
class MapEntry(object): |
| 51 |
|
| 52 |
def __init__(self, start, end, filename): |
| 53 |
self.start = start |
| 54 |
self.end = end |
| 55 |
self.filename = filename |
| 56 |
|
| 57 |
def __lt__(self, other): |
| 58 |
return self.start < other.start |
| 59 |
|
| 60 |
class ProcessMaps(object): |
| 61 |
|
| 62 |
def __init__(self): |
| 63 |
self.process_maps = {} # map from pid to a sorted list of MapEntry. |
| 64 |
|
| 65 |
def add(self, pid, map_entry): |
| 66 |
old_list = self.process_maps.get(pid, []) |
| 67 |
new_list = [] |
| 68 |
map_entry_used = False |
| 69 |
for entry in old_list: |
| 70 |
if entry.end <= map_entry.start: |
| 71 |
new_list.append(entry) |
| 72 |
elif entry.start < map_entry.start: |
| 73 |
entry.end = map_entry.start |
| 74 |
new_list.append(entry) |
| 75 |
else: |
| 76 |
if not map_entry_used: |
| 77 |
new_list.append(map_entry) |
| 78 |
map_entry_used = True |
| 79 |
if entry.start >= map_entry.end: |
| 80 |
new_list.append(entry) |
| 81 |
elif entry.end > map_entry.end: |
| 82 |
entry.start = map_entry.end |
| 83 |
new_list.append(entry) |
| 84 |
if not map_entry_used: |
| 85 |
new_list.append(map_entry) |
| 86 |
self.process_maps[pid] = new_list |
| 87 |
|
| 88 |
def fork_pid(self, pid, ppid): |
| 89 |
if pid == ppid: |
| 90 |
return |
| 91 |
entry_list = self.process_maps.get(ppid, []) |
| 92 |
self.process_maps[pid] = copy.deepcopy(entry_list) |
| 93 |
|
| 94 |
def find(self, pid, addr): |
| 95 |
key = MapEntry(addr, addr, '') |
| 96 |
entry_list = self.process_maps.get(pid, []) |
| 97 |
pos = bisect.bisect_right(entry_list, key) |
| 98 |
if pos > 0 and entry_list[pos - 1].end > addr: |
| 99 |
return entry_list[pos - 1] |
| 100 |
return None |
| 101 |
|
| 102 |
def show(self): |
| 103 |
for pid in sorted(self.process_maps): |
| 104 |
print(' pid %d' % pid) |
| 105 |
for entry in self.process_maps[pid]: |
| 106 |
print(' map [%x-%x] %s' % |
| 107 |
(entry.start, entry.end, entry.filename)) |
| 108 |
|
| 109 |
|
| 110 |
class UnwindingTimes(object): |
| 111 |
|
| 112 |
def __init__(self): |
| 113 |
self.total_time = 0 |
| 114 |
self.count = 0 |
| 115 |
self.max_time = 0 |
| 116 |
|
| 117 |
def add_time(self, used_time): |
| 118 |
self.total_time += used_time |
| 119 |
self.count += 1 |
| 120 |
self.max_time = max(self.max_time, used_time) |
| 121 |
|
| 122 |
|
| 123 |
class UnwindingMemConsumption(object): |
| 124 |
|
| 125 |
def __init__(self): |
| 126 |
self.before_unwinding = None |
| 127 |
self.after_unwinding = None |
| 128 |
|
| 129 |
|
| 130 |
class CallChainNode(object): |
| 131 |
|
| 132 |
""" Representing a node in a call chain.""" |
| 133 |
|
| 134 |
def __init__(self, ip, sp, filename, vaddr_in_file, function_name, map_start_addr, |
| 135 |
map_end_addr): |
| 136 |
self.ip = ip |
| 137 |
self.sp = sp |
| 138 |
self.filename = filename |
| 139 |
self.vaddr_in_file = vaddr_in_file |
| 140 |
self.function_name = function_name |
| 141 |
self.map_start_addr = map_start_addr |
| 142 |
self.map_end_addr = map_end_addr |
| 143 |
|
| 144 |
|
| 145 |
class SampleResult(object): |
| 146 |
|
| 147 |
""" Unwinding result per sample. """ |
| 148 |
|
| 149 |
def __init__(self, pid, tid, unwinding_result, callchain, sample_record): |
| 150 |
self.pid = pid |
| 151 |
self.tid = tid |
| 152 |
self.unwinding_result = unwinding_result |
| 153 |
self.callchain = callchain |
| 154 |
self.sample_record = sample_record |
| 155 |
|
| 156 |
def show(self): |
| 157 |
print(' pid: %d' % self.pid) |
| 158 |
print(' tid: %d' % self.tid) |
| 159 |
for key, value in self.unwinding_result.items(): |
| 160 |
print(' %s: %s' % (key, value)) |
| 161 |
for i, node in enumerate(self.callchain): |
| 162 |
print(' node %d: ip 0x%x, sp 0x%x, %s (%s[+%x]), map [%x-%x]' % ( |
| 163 |
i, node.ip, node.sp, node.function_name, node.filename, node.vaddr_in_file, |
| 164 |
node.map_start_addr, node.map_end_addr)) |
| 165 |
if self.sample_record: |
| 166 |
print(' original sample record:') |
| 167 |
for line in self.sample_record: |
| 168 |
print(' %s' % line) |
| 169 |
|
| 170 |
|
| 171 |
class FunctionResult(object): |
| 172 |
|
| 173 |
""" Unwinding result per function. """ |
| 174 |
|
| 175 |
def __init__(self): |
| 176 |
# Map from Unwinding result reason to [SampleResult]. |
| 177 |
self.sample_results = {} |
| 178 |
|
| 179 |
def add_sample_result(self, sample_result): |
| 180 |
stop_reason = sample_result.unwinding_result['stop_reason'] |
| 181 |
result_list = self.sample_results.get(stop_reason) |
| 182 |
if not result_list: |
| 183 |
result_list = self.sample_results[stop_reason] = [] |
| 184 |
for result in result_list: |
| 185 |
if result.callchain[-1].vaddr_in_file == sample_result.callchain[-1].vaddr_in_file: |
| 186 |
# This sample_result duplicates with an existing one. |
| 187 |
return |
| 188 |
# We don't want to store too many sample results for a function. |
| 189 |
if len(result_list) < 10: |
| 190 |
result_list.append(sample_result) |
| 191 |
|
| 192 |
def show(self): |
| 193 |
for stop_reason in sorted(self.sample_results): |
| 194 |
for sample_result in self.sample_results[stop_reason]: |
| 195 |
sample_result.show() |
| 196 |
|
| 197 |
|
| 198 |
class FileResult(object): |
| 199 |
|
| 200 |
""" Unwinding result per shared library. """ |
| 201 |
|
| 202 |
def __init__(self): |
| 203 |
self.function_results = {} # Map from function_name to FunctionResult. |
| 204 |
|
| 205 |
def add_sample_result(self, sample_result): |
| 206 |
function_name = sample_result.callchain[-1].function_name |
| 207 |
function_result = self.function_results.get(function_name) |
| 208 |
if not function_result: |
| 209 |
function_result = self.function_results[ |
| 210 |
function_name] = FunctionResult() |
| 211 |
function_result.add_sample_result(sample_result) |
| 212 |
|
| 213 |
def show(self): |
| 214 |
for function_name in sorted(self.function_results): |
| 215 |
print(' function %s' % function_name) |
| 216 |
self.function_results[function_name].show() |
| 217 |
print('\n') |
| 218 |
|
| 219 |
|
| 220 |
class UnwindingResultErrorReport(object): |
| 221 |
|
| 222 |
""" Report time used for unwinding and unwinding result errors. """ |
| 223 |
|
| 224 |
def __init__(self, omit_callchains_fixed_by_joiner): |
| 225 |
self.omit_callchains_fixed_by_joiner = omit_callchains_fixed_by_joiner |
| 226 |
self.process_maps = ProcessMaps() |
| 227 |
self.unwinding_times = UnwindingTimes() |
| 228 |
self.mem_stat = UnwindingMemConsumption() |
| 229 |
self.file_results = {} # map from filename to FileResult. |
| 230 |
|
| 231 |
def add_sample_result(self, sample_result, joined_record): |
| 232 |
self.unwinding_times.add_time(int(sample_result.unwinding_result['used_time'])) |
| 233 |
if self.should_omit(sample_result, joined_record): |
| 234 |
return |
| 235 |
filename = sample_result.callchain[-1].filename |
| 236 |
file_result = self.file_results.get(filename) |
| 237 |
if not file_result: |
| 238 |
file_result = self.file_results[filename] = FileResult() |
| 239 |
file_result.add_sample_result(sample_result) |
| 240 |
|
| 241 |
def add_mem_stat(self, name, mem_str): |
| 242 |
# mem_str is like VmPeak:202464 kB;VmSize:183456 kB;VmHWM:98256 kB;VmRSS:33680 kB. |
| 243 |
items = [] |
| 244 |
for p in mem_str.split(';'): |
| 245 |
key, value = p.split(':') |
| 246 |
items.append((key, value)) |
| 247 |
if name == 'debug_unwind_mem_before': |
| 248 |
self.mem_stat.before_unwinding = items |
| 249 |
else: |
| 250 |
self.mem_stat.after_unwinding = items |
| 251 |
|
| 252 |
def should_omit(self, sample_result, joined_record): |
| 253 |
# 1. Can't unwind code generated in memory. |
| 254 |
for name in ['/dev/ashmem/dalvik-jit-code-cache', '[anon:dalvik-jit-code-cache]', '//anon']: |
| 255 |
if name in sample_result.callchain[-1].filename: |
| 256 |
return True |
| 257 |
# 2. Don't report complete callchains, which can reach __libc_init or __start_thread in |
| 258 |
# libc.so. |
| 259 |
def is_callchain_complete(callchain): |
| 260 |
for node in callchain: |
| 261 |
if (node.filename.endswith('libc.so') and |
| 262 |
node.function_name in ['__libc_init', '__start_thread']): |
| 263 |
return True |
| 264 |
return False |
| 265 |
if is_callchain_complete(sample_result.callchain): |
| 266 |
return True |
| 267 |
# 3. Omit callchains made complete by callchain joiner. |
| 268 |
if self.omit_callchains_fixed_by_joiner and is_callchain_complete(joined_record.callchain): |
| 269 |
return True |
| 270 |
return False |
| 271 |
|
| 272 |
def show(self): |
| 273 |
print('Unwinding time info:') |
| 274 |
print(' total time: %f ms' % (self.unwinding_times.total_time / 1e6)) |
| 275 |
print(' unwinding count: %d' % self.unwinding_times.count) |
| 276 |
if self.unwinding_times.count > 0: |
| 277 |
print(' average time: %f us' % ( |
| 278 |
self.unwinding_times.total_time / 1e3 / self.unwinding_times.count)) |
| 279 |
print(' max time: %f us' % (self.unwinding_times.max_time / 1e3)) |
| 280 |
print('Unwinding mem info:') |
| 281 |
for items in zip(self.mem_stat.before_unwinding, self.mem_stat.after_unwinding): |
| 282 |
assert items[0][0] == items[1][0] |
| 283 |
print(' %s: %s -> %s' % (items[0][0], items[0][1], items[1][1])) |
| 284 |
print('Process maps:') |
| 285 |
self.process_maps.show() |
| 286 |
for filename in sorted(self.file_results): |
| 287 |
print('filename %s' % filename) |
| 288 |
self.file_results[filename].show() |
| 289 |
print('\n') |
| 290 |
|
| 291 |
|
| 292 |
class CallChainRecord(object): |
| 293 |
""" Store data of a callchain record. """ |
| 294 |
|
| 295 |
def __init__(self): |
| 296 |
self.pid = None |
| 297 |
self.tid = None |
| 298 |
self.callchain = [] |
| 299 |
|
| 300 |
def parse_sample_record(lines, i): |
| 301 |
""" Read the lines belong to a SampleRecord.""" |
| 302 |
if i == len(lines) or not lines[i].startswith('record sample:'): |
| 303 |
log_fatal('unexpected dump output near line %d' % i) |
| 304 |
start_line = i |
| 305 |
i += 1 |
| 306 |
while i < len(lines) and (not lines[i] or lines[i].startswith(' ')): |
| 307 |
i += 1 |
| 308 |
return i, lines[start_line:i] |
| 309 |
|
| 310 |
def parse_callchain_record(lines, i, chain_type, process_maps): |
| 311 |
if i == len(lines) or not lines[i].startswith('record callchain:'): |
| 312 |
log_fatal('unexpected dump output near line %d' % i) |
| 313 |
i += 1 |
| 314 |
record = CallChainRecord() |
| 315 |
ips = [] |
| 316 |
sps = [] |
| 317 |
function_names = [] |
| 318 |
filenames = [] |
| 319 |
vaddr_in_files = [] |
| 320 |
map_start_addrs = [] |
| 321 |
map_end_addrs = [] |
| 322 |
in_callchain = False |
| 323 |
while i < len(lines) and not lines[i].startswith('record'): |
| 324 |
line = lines[i].strip() |
| 325 |
items = line.split() |
| 326 |
if not items: |
| 327 |
i += 1 |
| 328 |
continue |
| 329 |
if items[0] == 'pid' and len(items) == 2: |
| 330 |
record.pid = int(items[1]) |
| 331 |
elif items[0] == 'tid' and len(items) == 2: |
| 332 |
record.tid = int(items[1]) |
| 333 |
elif items[0] == 'chain_type' and len(items) == 2: |
| 334 |
if items[1] != chain_type: |
| 335 |
log_fatal('unexpected dump output near line %d' % i) |
| 336 |
elif items[0] == 'ip': |
| 337 |
m = re.search(r'ip\s+0x(\w+),\s+sp\s+0x(\w+)$', line) |
| 338 |
if m: |
| 339 |
ips.append(int(m.group(1), 16)) |
| 340 |
sps.append(int(m.group(2), 16)) |
| 341 |
elif items[0] == 'callchain:': |
| 342 |
in_callchain = True |
| 343 |
elif in_callchain: |
| 344 |
# "dalvik-jit-code-cache (deleted)[+346c] ([anon:dalvik-jit-code-cache] |
| 345 |
# (deleted)[+346c])" |
| 346 |
if re.search(r'\)\[\+\w+\]\)$', line): |
| 347 |
break_pos = line.rfind('(', 0, line.rfind('(')) |
| 348 |
else: |
| 349 |
break_pos = line.rfind('(') |
| 350 |
if break_pos > 0: |
| 351 |
m = re.match(r'(.*)\[\+(\w+)\]\)', line[break_pos + 1:]) |
| 352 |
if m: |
| 353 |
function_names.append(line[:break_pos].strip()) |
| 354 |
filenames.append(m.group(1)) |
| 355 |
vaddr_in_files.append(int(m.group(2), 16)) |
| 356 |
i += 1 |
| 357 |
|
| 358 |
for ip in ips: |
| 359 |
map_entry = process_maps.find(record.pid, ip) |
| 360 |
if map_entry: |
| 361 |
map_start_addrs.append(map_entry.start) |
| 362 |
map_end_addrs.append(map_entry.end) |
| 363 |
else: |
| 364 |
map_start_addrs.append(0) |
| 365 |
map_end_addrs.append(0) |
| 366 |
n = len(ips) |
| 367 |
if (None in [record.pid, record.tid] or n == 0 or len(sps) != n or |
| 368 |
len(function_names) != n or len(filenames) != n or len(vaddr_in_files) != n or |
| 369 |
len(map_start_addrs) != n or len(map_end_addrs) != n): |
| 370 |
log_fatal('unexpected dump output near line %d' % i) |
| 371 |
for j in range(n): |
| 372 |
record.callchain.append(CallChainNode(ips[j], sps[j], filenames[j], vaddr_in_files[j], |
| 373 |
function_names[j], map_start_addrs[j], |
| 374 |
map_end_addrs[j])) |
| 375 |
return i, record |
| 376 |
|
| 377 |
|
| 378 |
def build_unwinding_result_report(args): |
| 379 |
simpleperf_path = get_host_binary_path('simpleperf') |
| 380 |
proc = subprocess.Popen([simpleperf_path, 'dump', args.record_file[0]], |
| 381 |
stdout=subprocess.PIPE) |
| 382 |
(stdoutdata, _) = proc.communicate() |
| 383 |
stdoutdata = bytes_to_str(stdoutdata) |
| 384 |
if 'debug_unwind = true' not in stdoutdata: |
| 385 |
log_exit("Can't parse unwinding result. Because " + |
| 386 |
"%s was not generated by the debug-unwind cmd." % args.record_file[0]) |
| 387 |
unwinding_report = UnwindingResultErrorReport(args.omit_callchains_fixed_by_joiner) |
| 388 |
process_maps = unwinding_report.process_maps |
| 389 |
lines = stdoutdata.split('\n') |
| 390 |
i = 0 |
| 391 |
while i < len(lines): |
| 392 |
if lines[i].startswith('record mmap:') or lines[i].startswith('record mmap2:'): |
| 393 |
i += 1 |
| 394 |
pid = None |
| 395 |
start = None |
| 396 |
end = None |
| 397 |
filename = None |
| 398 |
while i < len(lines) and not lines[i].startswith('record'): |
| 399 |
if lines[i].startswith(' pid'): |
| 400 |
m = re.search(r'pid\s+(\d+).+addr\s+0x(\w+).+len\s+0x(\w+)', lines[i]) |
| 401 |
if m: |
| 402 |
pid = int(m.group(1)) |
| 403 |
start = int(m.group(2), 16) |
| 404 |
end = start + int(m.group(3), 16) |
| 405 |
elif 'filename' in lines[i]: |
| 406 |
pos = lines[i].find('filename') + len('filename') |
| 407 |
filename = lines[i][pos:].strip() |
| 408 |
i += 1 |
| 409 |
if None in [pid, start, end, filename]: |
| 410 |
log_fatal('unexpected dump output near line %d' % i) |
| 411 |
process_maps.add(pid, MapEntry(start, end, filename)) |
| 412 |
elif lines[i].startswith('record unwinding_result:'): |
| 413 |
i += 1 |
| 414 |
unwinding_result = collections.OrderedDict() |
| 415 |
while i < len(lines) and not lines[i].startswith('record'): |
| 416 |
strs = (lines[i].strip()).split() |
| 417 |
if len(strs) == 2: |
| 418 |
unwinding_result[strs[0]] = strs[1] |
| 419 |
i += 1 |
| 420 |
for key in ['time', 'used_time', 'stop_reason']: |
| 421 |
if key not in unwinding_result: |
| 422 |
log_fatal('unexpected dump output near line %d' % i) |
| 423 |
|
| 424 |
i, sample_record = parse_sample_record(lines, i) |
| 425 |
i, original_record = parse_callchain_record(lines, i, 'ORIGINAL_OFFLINE', process_maps) |
| 426 |
i, joined_record = parse_callchain_record(lines, i, 'JOINED_OFFLINE', process_maps) |
| 427 |
if args.omit_sample: |
| 428 |
sample_record = [] |
| 429 |
sample_result = SampleResult(original_record.pid, original_record.tid, |
| 430 |
unwinding_result, original_record.callchain, |
| 431 |
sample_record) |
| 432 |
unwinding_report.add_sample_result(sample_result, joined_record) |
| 433 |
elif lines[i].startswith('record fork:'): |
| 434 |
i += 1 |
| 435 |
pid = None |
| 436 |
ppid = None |
| 437 |
while i < len(lines) and not lines[i].startswith('record'): |
| 438 |
if lines[i].startswith(' pid'): |
| 439 |
m = re.search(r'pid\s+(\w+),\s+ppid\s+(\w+)', lines[i]) |
| 440 |
if m: |
| 441 |
pid = int(m.group(1)) |
| 442 |
ppid = int(m.group(2)) |
| 443 |
i += 1 |
| 444 |
if None in [pid, ppid]: |
| 445 |
log_fatal('unexpected dump output near line %d' % i) |
| 446 |
process_maps.fork_pid(pid, ppid) |
| 447 |
elif lines[i].startswith(' debug_unwind_mem'): |
| 448 |
items = lines[i].strip().split(' = ') |
| 449 |
if len(items) == 2: |
| 450 |
unwinding_report.add_mem_stat(items[0], items[1]) |
| 451 |
i += 1 |
| 452 |
else: |
| 453 |
i += 1 |
| 454 |
return unwinding_report |
| 455 |
|
| 456 |
|
| 457 |
def main(): |
| 458 |
parser = argparse.ArgumentParser( |
| 459 |
description='Report dwarf unwinding results generated by the debug-unwind cmd.') |
| 460 |
parser.add_argument('-i', '--record_file', nargs=1, default=['perf.data.debug'], help=""" |
| 461 |
Set profiling data to report. Default is perf.data.debug.""") |
| 462 |
parser.add_argument('--omit-callchains-fixed-by-joiner', action='store_true', help=""" |
| 463 |
Don't show incomplete callchains fixed by callchain joiner.""") |
| 464 |
parser.add_argument('--omit-sample', action='store_true', help="""Don't show original sample |
| 465 |
records.""") |
| 466 |
args = parser.parse_args() |
| 467 |
report = build_unwinding_result_report(args) |
| 468 |
report.show() |
| 469 |
|
| 470 |
if __name__ == '__main__': |
| 471 |
main() |