1 #! /usr/bin/env perl 2 3 # Copyright (c) 1998-2007, Google Inc. 4 # All rights reserved. 5 # 6 # Redistribution and use in source and binary forms, with or without 7 # modification, are permitted provided that the following conditions are 8 # met: 9 # 10 # * Redistributions of source code must retain the above copyright 11 # notice, this list of conditions and the following disclaimer. 12 # * Redistributions in binary form must reproduce the above 13 # copyright notice, this list of conditions and the following disclaimer 14 # in the documentation and/or other materials provided with the 15 # distribution. 16 # * Neither the name of Google Inc. nor the names of its 17 # contributors may be used to endorse or promote products derived from 18 # this software without specific prior written permission. 19 # 20 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 23 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 24 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 25 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 26 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 27 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 28 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 29 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 30 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 31 32 # --- 33 # Program for printing the profile generated by common/profiler.cc, 34 # or by the heap profiler (common/debugallocation.cc) 35 # 36 # The profile contains a sequence of entries of the form: 37 # <count> <stack trace> 38 # This program parses the profile, and generates user-readable 39 # output. 40 # 41 # Examples: 42 # 43 # % tools/pprof "program" "profile" 44 # Enters "interactive" mode 45 # 46 # % tools/pprof --text "program" "profile" 47 # Generates one line per procedure 48 # 49 # % tools/pprof --gv "program" "profile" 50 # Generates annotated call-graph and displays via "gv" 51 # 52 # % tools/pprof --gv --focus=Mutex "program" "profile" 53 # Restrict to code paths that involve an entry that matches "Mutex" 54 # 55 # % tools/pprof --gv --focus=Mutex --ignore=string "program" "profile" 56 # Restrict to code paths that involve an entry that matches "Mutex" 57 # and does not match "string" 58 # 59 # % tools/pprof --list=IBF_CheckDocid "program" "profile" 60 # Generates disassembly listing of all routines with at least one 61 # sample that match the --list=<regexp> pattern. The listing is 62 # annotated with the flat and cumulative sample counts at each line. 63 # 64 # % tools/pprof --disasm=IBF_CheckDocid "program" "profile" 65 # Generates disassembly listing of all routines with at least one 66 # sample that match the --disasm=<regexp> pattern. The listing is 67 # annotated with the flat and cumulative sample counts at each PC value. 68 # 69 # TODO: Use color to indicate files? 70 71 use strict; 72 use warnings; 73 use Getopt::Long; 74 75 my $PPROF_VERSION = "2.0"; 76 77 # These are the object tools we use which can come from a 78 # user-specified location using --tools, from the PPROF_TOOLS 79 # environment variable, or from the environment. 80 my %obj_tool_map = ( 81 "objdump" => "objdump", 82 "nm" => "nm", 83 "addr2line" => "addr2line", 84 "c++filt" => "c++filt", 85 ## ConfigureObjTools may add architecture-specific entries: 86 #"nm_pdb" => "nm-pdb", # for reading windows (PDB-format) executables 87 #"addr2line_pdb" => "addr2line-pdb", # ditto 88 #"otool" => "otool", # equivalent of objdump on OS X 89 ); 90 # NOTE: these are lists, so you can put in commandline flags if you want. 91 my @DOT = ("dot"); # leave non-absolute, since it may be in /usr/local 92 my @GV = ("gv"); 93 my @EVINCE = ("evince"); # could also be xpdf or perhaps acroread 94 my @KCACHEGRIND = ("kcachegrind"); 95 my @PS2PDF = ("ps2pdf"); 96 # These are used for dynamic profiles 97 my @URL_FETCHER = ("curl", "-s"); 98 99 # These are the web pages that servers need to support for dynamic profiles 100 my $HEAP_PAGE = "/pprof/heap"; 101 my $PROFILE_PAGE = "/pprof/profile"; # must support cgi-param "?seconds=#" 102 my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param 103 # ?seconds=#&event=x&period=n 104 my $GROWTH_PAGE = "/pprof/growth"; 105 my $CONTENTION_PAGE = "/pprof/contention"; 106 my $WALL_PAGE = "/pprof/wall(?:\\?.*)?"; # accepts options like namefilter 107 my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?"; 108 my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param 109 # "?seconds=#", 110 # "?tags_regexp=#" and 111 # "?type=#". 112 my $SYMBOL_PAGE = "/pprof/symbol"; # must support symbol lookup via POST 113 my $PROGRAM_NAME_PAGE = "/pprof/cmdline"; 114 115 # These are the web pages that can be named on the command line. 116 # All the alternatives must begin with /. 117 my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" . 118 "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" . 119 "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)"; 120 121 # default binary name 122 my $UNKNOWN_BINARY = "(unknown)"; 123 124 # There is a pervasive dependency on the length (in hex characters, 125 # i.e., nibbles) of an address, distinguishing between 32-bit and 126 # 64-bit profiles. To err on the safe size, default to 64-bit here: 127 my $address_length = 16; 128 129 my $dev_null = "/dev/null"; 130 if (! -e $dev_null && $^O =~ /MSWin/) { # $^O is the OS perl was built for 131 $dev_null = "nul"; 132 } 133 134 # A list of paths to search for shared object files 135 my @prefix_list = (); 136 137 # Special routine name that should not have any symbols. 138 # Used as separator to parse "addr2line -i" output. 139 my $sep_symbol = '_fini'; 140 my $sep_address = undef; 141 142 ##### Argument parsing ##### 143 144 sub usage_string { 145 return <<EOF; 146 Usage: 147 pprof [options] <program> <profiles> 148 <profiles> is a space separated list of profile names. 149 pprof [options] <symbolized-profiles> 150 <symbolized-profiles> is a list of profile files where each file contains 151 the necessary symbol mappings as well as profile data (likely generated 152 with --raw). 153 pprof [options] <profile> 154 <profile> is a remote form. Symbols are obtained from host:port$SYMBOL_PAGE 155 156 Each name can be: 157 /path/to/profile - a path to a profile file 158 host:port[/<service>] - a location of a service to get profile from 159 160 The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile, 161 $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall, 162 $CENSUSPROFILE_PAGE, or /pprof/filteredprofile. 163 For instance: 164 pprof http://myserver.com:80$HEAP_PAGE 165 If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling). 166 pprof --symbols <program> 167 Maps addresses to symbol names. In this mode, stdin should be a 168 list of library mappings, in the same format as is found in the heap- 169 and cpu-profile files (this loosely matches that of /proc/self/maps 170 on linux), followed by a list of hex addresses to map, one per line. 171 172 For more help with querying remote servers, including how to add the 173 necessary server-side support code, see this filename (or one like it): 174 175 /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html 176 177 Options: 178 --cum Sort by cumulative data 179 --base=<base> Subtract <base> from <profile> before display 180 --interactive Run in interactive mode (interactive "help" gives help) [default] 181 --seconds=<n> Length of time for dynamic profiles [default=30 secs] 182 --add_lib=<file> Read additional symbols and line info from the given library 183 --lib_prefix=<dir> Comma separated list of library path prefixes 184 185 Reporting Granularity: 186 --addresses Report at address level 187 --lines Report at source line level 188 --functions Report at function level [default] 189 --files Report at source file level 190 191 Output type: 192 --text Generate text report 193 --callgrind Generate callgrind format to stdout 194 --gv Generate Postscript and display 195 --evince Generate PDF and display 196 --web Generate SVG and display 197 --list=<regexp> Generate source listing of matching routines 198 --disasm=<regexp> Generate disassembly of matching routines 199 --symbols Print demangled symbol names found at given addresses 200 --dot Generate DOT file to stdout 201 --ps Generate Postcript to stdout 202 --pdf Generate PDF to stdout 203 --svg Generate SVG to stdout 204 --gif Generate GIF to stdout 205 --raw Generate symbolized pprof data (useful with remote fetch) 206 207 Heap-Profile Options: 208 --inuse_space Display in-use (mega)bytes [default] 209 --inuse_objects Display in-use objects 210 --alloc_space Display allocated (mega)bytes 211 --alloc_objects Display allocated objects 212 --show_bytes Display space in bytes 213 --drop_negative Ignore negative differences 214 215 Contention-profile options: 216 --total_delay Display total delay at each region [default] 217 --contentions Display number of delays at each region 218 --mean_delay Display mean delay at each region 219 220 Call-graph Options: 221 --nodecount=<n> Show at most so many nodes [default=80] 222 --nodefraction=<f> Hide nodes below <f>*total [default=.005] 223 --edgefraction=<f> Hide edges below <f>*total [default=.001] 224 --maxdegree=<n> Max incoming/outgoing edges per node [default=8] 225 --focus=<regexp> Focus on nodes matching <regexp> 226 --ignore=<regexp> Ignore nodes matching <regexp> 227 --scale=<n> Set GV scaling [default=0] 228 --heapcheck Make nodes with non-0 object counts 229 (i.e. direct leak generators) more visible 230 231 Miscellaneous: 232 --tools=<prefix or binary:fullpath>[,...] \$PATH for object tool pathnames 233 --test Run unit tests 234 --help This message 235 --version Version information 236 237 Environment Variables: 238 PPROF_TMPDIR Profiles directory. Defaults to \$HOME/pprof 239 PPROF_TOOLS Prefix for object tools pathnames 240 241 Examples: 242 243 pprof /bin/ls ls.prof 244 Enters "interactive" mode 245 pprof --text /bin/ls ls.prof 246 Outputs one line per procedure 247 pprof --web /bin/ls ls.prof 248 Displays annotated call-graph in web browser 249 pprof --gv /bin/ls ls.prof 250 Displays annotated call-graph via 'gv' 251 pprof --gv --focus=Mutex /bin/ls ls.prof 252 Restricts to code paths including a .*Mutex.* entry 253 pprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof 254 Code paths including Mutex but not string 255 pprof --list=getdir /bin/ls ls.prof 256 (Per-line) annotated source listing for getdir() 257 pprof --disasm=getdir /bin/ls ls.prof 258 (Per-PC) annotated disassembly for getdir() 259 260 pprof http://localhost:1234/ 261 Enters "interactive" mode 262 pprof --text localhost:1234 263 Outputs one line per procedure for localhost:1234 264 pprof --raw localhost:1234 > ./local.raw 265 pprof --text ./local.raw 266 Fetches a remote profile for later analysis and then 267 analyzes it in text mode. 268 EOF 269 } 270 271 sub version_string { 272 return <<EOF 273 pprof (part of gperftools $PPROF_VERSION) 274 275 Copyright 1998-2007 Google Inc. 276 277 This is BSD licensed software; see the source for copying conditions 278 and license information. 279 There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A 280 PARTICULAR PURPOSE. 281 EOF 282 } 283 284 sub usage { 285 my $msg = shift; 286 print STDERR "$msg\n\n"; 287 print STDERR usage_string(); 288 print STDERR "\nFATAL ERROR: $msg\n"; # just as a reminder 289 exit(1); 290 } 291 292 sub Init() { 293 # Setup tmp-file name and handler to clean it up. 294 # We do this in the very beginning so that we can use 295 # error() and cleanup() function anytime here after. 296 $main::tmpfile_sym = "/tmp/pprof$$.sym"; 297 $main::tmpfile_ps = "/tmp/pprof$$"; 298 $main::next_tmpfile = 0; 299 $SIG{'INT'} = \&sighandler; 300 301 # Cache from filename/linenumber to source code 302 $main::source_cache = (); 303 304 $main::opt_help = 0; 305 $main::opt_version = 0; 306 307 $main::opt_cum = 0; 308 $main::opt_base = ''; 309 $main::opt_addresses = 0; 310 $main::opt_lines = 0; 311 $main::opt_functions = 0; 312 $main::opt_files = 0; 313 $main::opt_lib_prefix = ""; 314 315 $main::opt_text = 0; 316 $main::opt_callgrind = 0; 317 $main::opt_list = ""; 318 $main::opt_disasm = ""; 319 $main::opt_symbols = 0; 320 $main::opt_gv = 0; 321 $main::opt_evince = 0; 322 $main::opt_web = 0; 323 $main::opt_dot = 0; 324 $main::opt_ps = 0; 325 $main::opt_pdf = 0; 326 $main::opt_gif = 0; 327 $main::opt_svg = 0; 328 $main::opt_raw = 0; 329 330 $main::opt_nodecount = 80; 331 $main::opt_nodefraction = 0.005; 332 $main::opt_edgefraction = 0.001; 333 $main::opt_maxdegree = 8; 334 $main::opt_focus = ''; 335 $main::opt_ignore = ''; 336 $main::opt_scale = 0; 337 $main::opt_heapcheck = 0; 338 $main::opt_seconds = 30; 339 $main::opt_lib = ""; 340 341 $main::opt_inuse_space = 0; 342 $main::opt_inuse_objects = 0; 343 $main::opt_alloc_space = 0; 344 $main::opt_alloc_objects = 0; 345 $main::opt_show_bytes = 0; 346 $main::opt_drop_negative = 0; 347 $main::opt_interactive = 0; 348 349 $main::opt_total_delay = 0; 350 $main::opt_contentions = 0; 351 $main::opt_mean_delay = 0; 352 353 $main::opt_tools = ""; 354 $main::opt_debug = 0; 355 $main::opt_test = 0; 356 357 # These are undocumented flags used only by unittests. 358 $main::opt_test_stride = 0; 359 360 # Are we using $SYMBOL_PAGE? 361 $main::use_symbol_page = 0; 362 363 # Files returned by TempName. 364 %main::tempnames = (); 365 366 # Type of profile we are dealing with 367 # Supported types: 368 # cpu 369 # heap 370 # growth 371 # contention 372 $main::profile_type = ''; # Empty type means "unknown" 373 374 GetOptions("help!" => \$main::opt_help, 375 "version!" => \$main::opt_version, 376 "cum!" => \$main::opt_cum, 377 "base=s" => \$main::opt_base, 378 "seconds=i" => \$main::opt_seconds, 379 "add_lib=s" => \$main::opt_lib, 380 "lib_prefix=s" => \$main::opt_lib_prefix, 381 "functions!" => \$main::opt_functions, 382 "lines!" => \$main::opt_lines, 383 "addresses!" => \$main::opt_addresses, 384 "files!" => \$main::opt_files, 385 "text!" => \$main::opt_text, 386 "callgrind!" => \$main::opt_callgrind, 387 "list=s" => \$main::opt_list, 388 "disasm=s" => \$main::opt_disasm, 389 "symbols!" => \$main::opt_symbols, 390 "gv!" => \$main::opt_gv, 391 "evince!" => \$main::opt_evince, 392 "web!" => \$main::opt_web, 393 "dot!" => \$main::opt_dot, 394 "ps!" => \$main::opt_ps, 395 "pdf!" => \$main::opt_pdf, 396 "svg!" => \$main::opt_svg, 397 "gif!" => \$main::opt_gif, 398 "raw!" => \$main::opt_raw, 399 "interactive!" => \$main::opt_interactive, 400 "nodecount=i" => \$main::opt_nodecount, 401 "nodefraction=f" => \$main::opt_nodefraction, 402 "edgefraction=f" => \$main::opt_edgefraction, 403 "maxdegree=i" => \$main::opt_maxdegree, 404 "focus=s" => \$main::opt_focus, 405 "ignore=s" => \$main::opt_ignore, 406 "scale=i" => \$main::opt_scale, 407 "heapcheck" => \$main::opt_heapcheck, 408 "inuse_space!" => \$main::opt_inuse_space, 409 "inuse_objects!" => \$main::opt_inuse_objects, 410 "alloc_space!" => \$main::opt_alloc_space, 411 "alloc_objects!" => \$main::opt_alloc_objects, 412 "show_bytes!" => \$main::opt_show_bytes, 413 "drop_negative!" => \$main::opt_drop_negative, 414 "total_delay!" => \$main::opt_total_delay, 415 "contentions!" => \$main::opt_contentions, 416 "mean_delay!" => \$main::opt_mean_delay, 417 "tools=s" => \$main::opt_tools, 418 "test!" => \$main::opt_test, 419 "debug!" => \$main::opt_debug, 420 # Undocumented flags used only by unittests: 421 "test_stride=i" => \$main::opt_test_stride, 422 ) || usage("Invalid option(s)"); 423 424 # Deal with the standard --help and --version 425 if ($main::opt_help) { 426 print usage_string(); 427 exit(0); 428 } 429 430 if ($main::opt_version) { 431 print version_string(); 432 exit(0); 433 } 434 435 # Disassembly/listing/symbols mode requires address-level info 436 if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) { 437 $main::opt_functions = 0; 438 $main::opt_lines = 0; 439 $main::opt_addresses = 1; 440 $main::opt_files = 0; 441 } 442 443 # Check heap-profiling flags 444 if ($main::opt_inuse_space + 445 $main::opt_inuse_objects + 446 $main::opt_alloc_space + 447 $main::opt_alloc_objects > 1) { 448 usage("Specify at most on of --inuse/--alloc options"); 449 } 450 451 # Check output granularities 452 my $grains = 453 $main::opt_functions + 454 $main::opt_lines + 455 $main::opt_addresses + 456 $main::opt_files + 457 0; 458 if ($grains > 1) { 459 usage("Only specify one output granularity option"); 460 } 461 if ($grains == 0) { 462 $main::opt_functions = 1; 463 } 464 465 # Check output modes 466 my $modes = 467 $main::opt_text + 468 $main::opt_callgrind + 469 ($main::opt_list eq '' ? 0 : 1) + 470 ($main::opt_disasm eq '' ? 0 : 1) + 471 ($main::opt_symbols == 0 ? 0 : 1) + 472 $main::opt_gv + 473 $main::opt_evince + 474 $main::opt_web + 475 $main::opt_dot + 476 $main::opt_ps + 477 $main::opt_pdf + 478 $main::opt_svg + 479 $main::opt_gif + 480 $main::opt_raw + 481 $main::opt_interactive + 482 0; 483 if ($modes > 1) { 484 usage("Only specify one output mode"); 485 } 486 if ($modes == 0) { 487 if (-t STDOUT) { # If STDOUT is a tty, activate interactive mode 488 $main::opt_interactive = 1; 489 } else { 490 $main::opt_text = 1; 491 } 492 } 493 494 if ($main::opt_test) { 495 RunUnitTests(); 496 # Should not return 497 exit(1); 498 } 499 500 # Binary name and profile arguments list 501 $main::prog = ""; 502 @main::pfile_args = (); 503 504 # Remote profiling without a binary (using $SYMBOL_PAGE instead) 505 if (@ARGV > 0) { 506 if (IsProfileURL($ARGV[0])) { 507 $main::use_symbol_page = 1; 508 } elsif (IsSymbolizedProfileFile($ARGV[0])) { 509 $main::use_symbolized_profile = 1; 510 $main::prog = $UNKNOWN_BINARY; # will be set later from the profile file 511 } 512 } 513 514 if ($main::use_symbol_page || $main::use_symbolized_profile) { 515 # We don't need a binary! 516 my %disabled = ('--lines' => $main::opt_lines, 517 '--disasm' => $main::opt_disasm); 518 for my $option (keys %disabled) { 519 usage("$option cannot be used without a binary") if $disabled{$option}; 520 } 521 # Set $main::prog later... 522 scalar(@ARGV) || usage("Did not specify profile file"); 523 } elsif ($main::opt_symbols) { 524 # --symbols needs a binary-name (to run nm on, etc) but not profiles 525 $main::prog = shift(@ARGV) || usage("Did not specify program"); 526 } else { 527 $main::prog = shift(@ARGV) || usage("Did not specify program"); 528 scalar(@ARGV) || usage("Did not specify profile file"); 529 } 530 531 # Parse profile file/location arguments 532 foreach my $farg (@ARGV) { 533 if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) { 534 my $machine = $1; 535 my $num_machines = $2; 536 my $path = $3; 537 for (my $i = 0; $i < $num_machines; $i++) { 538 unshift(@main::pfile_args, "$i.$machine$path"); 539 } 540 } else { 541 unshift(@main::pfile_args, $farg); 542 } 543 } 544 545 if ($main::use_symbol_page) { 546 unless (IsProfileURL($main::pfile_args[0])) { 547 error("The first profile should be a remote form to use $SYMBOL_PAGE\n"); 548 } 549 CheckSymbolPage(); 550 $main::prog = FetchProgramName(); 551 } elsif (!$main::use_symbolized_profile) { # may not need objtools! 552 ConfigureObjTools($main::prog) 553 } 554 555 # Break the opt_lib_prefix into the prefix_list array 556 @prefix_list = split (',', $main::opt_lib_prefix); 557 558 # Remove trailing / from the prefixes, in the list to prevent 559 # searching things like /my/path//lib/mylib.so 560 foreach (@prefix_list) { 561 s|/+$||; 562 } 563 } 564 565 sub Main() { 566 Init(); 567 $main::collected_profile = undef; 568 @main::profile_files = (); 569 $main::op_time = time(); 570 571 # Printing symbols is special and requires a lot less info that most. 572 if ($main::opt_symbols) { 573 PrintSymbols(*STDIN); # Get /proc/maps and symbols output from stdin 574 return; 575 } 576 577 # Fetch all profile data 578 FetchDynamicProfiles(); 579 580 # this will hold symbols that we read from the profile files 581 my $symbol_map = {}; 582 583 # Read one profile, pick the last item on the list 584 my $data = ReadProfile($main::prog, pop(@main::profile_files)); 585 my $profile = $data->{profile}; 586 my $pcs = $data->{pcs}; 587 my $libs = $data->{libs}; # Info about main program and shared libraries 588 $symbol_map = MergeSymbols($symbol_map, $data->{symbols}); 589 590 # Add additional profiles, if available. 591 if (scalar(@main::profile_files) > 0) { 592 foreach my $pname (@main::profile_files) { 593 my $data2 = ReadProfile($main::prog, $pname); 594 $profile = AddProfile($profile, $data2->{profile}); 595 $pcs = AddPcs($pcs, $data2->{pcs}); 596 $symbol_map = MergeSymbols($symbol_map, $data2->{symbols}); 597 } 598 } 599 600 # Subtract base from profile, if specified 601 if ($main::opt_base ne '') { 602 my $base = ReadProfile($main::prog, $main::opt_base); 603 $profile = SubtractProfile($profile, $base->{profile}); 604 $pcs = AddPcs($pcs, $base->{pcs}); 605 $symbol_map = MergeSymbols($symbol_map, $base->{symbols}); 606 } 607 608 # Get total data in profile 609 my $total = TotalProfile($profile); 610 611 # Collect symbols 612 my $symbols; 613 if ($main::use_symbolized_profile) { 614 $symbols = FetchSymbols($pcs, $symbol_map); 615 } elsif ($main::use_symbol_page) { 616 $symbols = FetchSymbols($pcs); 617 } else { 618 # TODO(csilvers): $libs uses the /proc/self/maps data from profile1, 619 # which may differ from the data from subsequent profiles, especially 620 # if they were run on different machines. Use appropriate libs for 621 # each pc somehow. 622 $symbols = ExtractSymbols($libs, $pcs); 623 } 624 625 # Remove uniniteresting stack items 626 $profile = RemoveUninterestingFrames($symbols, $profile); 627 628 # Focus? 629 if ($main::opt_focus ne '') { 630 $profile = FocusProfile($symbols, $profile, $main::opt_focus); 631 } 632 633 # Ignore? 634 if ($main::opt_ignore ne '') { 635 $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore); 636 } 637 638 my $calls = ExtractCalls($symbols, $profile); 639 640 # Reduce profiles to required output granularity, and also clean 641 # each stack trace so a given entry exists at most once. 642 my $reduced = ReduceProfile($symbols, $profile); 643 644 # Get derived profiles 645 my $flat = FlatProfile($reduced); 646 my $cumulative = CumulativeProfile($reduced); 647 648 # Print 649 if (!$main::opt_interactive) { 650 if ($main::opt_disasm) { 651 PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm); 652 } elsif ($main::opt_list) { 653 PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0); 654 } elsif ($main::opt_text) { 655 # Make sure the output is empty when have nothing to report 656 # (only matters when --heapcheck is given but we must be 657 # compatible with old branches that did not pass --heapcheck always): 658 if ($total != 0) { 659 printf("Total: %s %s\n", Unparse($total), Units()); 660 } 661 PrintText($symbols, $flat, $cumulative, -1); 662 } elsif ($main::opt_raw) { 663 PrintSymbolizedProfile($symbols, $profile, $main::prog); 664 } elsif ($main::opt_callgrind) { 665 PrintCallgrind($calls); 666 } else { 667 if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) { 668 if ($main::opt_gv) { 669 RunGV(TempName($main::next_tmpfile, "ps"), ""); 670 } elsif ($main::opt_evince) { 671 RunEvince(TempName($main::next_tmpfile, "pdf"), ""); 672 } elsif ($main::opt_web) { 673 my $tmp = TempName($main::next_tmpfile, "svg"); 674 RunWeb($tmp); 675 # The command we run might hand the file name off 676 # to an already running browser instance and then exit. 677 # Normally, we'd remove $tmp on exit (right now), 678 # but fork a child to remove $tmp a little later, so that the 679 # browser has time to load it first. 680 delete $main::tempnames{$tmp}; 681 if (fork() == 0) { 682 sleep 5; 683 unlink($tmp); 684 exit(0); 685 } 686 } 687 } else { 688 cleanup(); 689 exit(1); 690 } 691 } 692 } else { 693 InteractiveMode($profile, $symbols, $libs, $total); 694 } 695 696 cleanup(); 697 exit(0); 698 } 699 700 ##### Entry Point ##### 701 702 Main(); 703 704 # Temporary code to detect if we're running on a Goobuntu system. 705 # These systems don't have the right stuff installed for the special 706 # Readline libraries to work, so as a temporary workaround, we default 707 # to using the normal stdio code, rather than the fancier readline-based 708 # code 709 sub ReadlineMightFail { 710 if (-e '/lib/libtermcap.so.2') { 711 return 0; # libtermcap exists, so readline should be okay 712 } else { 713 return 1; 714 } 715 } 716 717 sub RunGV { 718 my $fname = shift; 719 my $bg = shift; # "" or " &" if we should run in background 720 if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) { 721 # Options using double dash are supported by this gv version. 722 # Also, turn on noantialias to better handle bug in gv for 723 # postscript files with large dimensions. 724 # TODO: Maybe we should not pass the --noantialias flag 725 # if the gv version is known to work properly without the flag. 726 system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname) 727 . $bg); 728 } else { 729 # Old gv version - only supports options that use single dash. 730 print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n"; 731 system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg); 732 } 733 } 734 735 sub RunEvince { 736 my $fname = shift; 737 my $bg = shift; # "" or " &" if we should run in background 738 system(ShellEscape(@EVINCE, $fname) . $bg); 739 } 740 741 sub RunWeb { 742 my $fname = shift; 743 print STDERR "Loading web page file:///$fname\n"; 744 745 if (`uname` =~ /Darwin/) { 746 # OS X: open will use standard preference for SVG files. 747 system("/usr/bin/open", $fname); 748 return; 749 } 750 751 # Some kind of Unix; try generic symlinks, then specific browsers. 752 # (Stop once we find one.) 753 # Works best if the browser is already running. 754 my @alt = ( 755 "/etc/alternatives/gnome-www-browser", 756 "/etc/alternatives/x-www-browser", 757 "google-chrome", 758 "firefox", 759 ); 760 foreach my $b (@alt) { 761 if (system($b, $fname) == 0) { 762 return; 763 } 764 } 765 766 print STDERR "Could not load web browser.\n"; 767 } 768 769 sub RunKcachegrind { 770 my $fname = shift; 771 my $bg = shift; # "" or " &" if we should run in background 772 print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n"; 773 system(ShellEscape(@KCACHEGRIND, $fname) . $bg); 774 } 775 776 777 ##### Interactive helper routines ##### 778 779 sub InteractiveMode { 780 $| = 1; # Make output unbuffered for interactive mode 781 my ($orig_profile, $symbols, $libs, $total) = @_; 782 783 print STDERR "Welcome to pprof! For help, type 'help'.\n"; 784 785 # Use ReadLine if it's installed and input comes from a console. 786 if ( -t STDIN && 787 !ReadlineMightFail() && 788 defined(eval {require Term::ReadLine}) ) { 789 my $term = new Term::ReadLine 'pprof'; 790 while ( defined ($_ = $term->readline('(pprof) '))) { 791 $term->addhistory($_) if /\S/; 792 if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) { 793 last; # exit when we get an interactive command to quit 794 } 795 } 796 } else { # don't have readline 797 while (1) { 798 print STDERR "(pprof) "; 799 $_ = <STDIN>; 800 last if ! defined $_ ; 801 s/\r//g; # turn windows-looking lines into unix-looking lines 802 803 # Save some flags that might be reset by InteractiveCommand() 804 my $save_opt_lines = $main::opt_lines; 805 806 if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) { 807 last; # exit when we get an interactive command to quit 808 } 809 810 # Restore flags 811 $main::opt_lines = $save_opt_lines; 812 } 813 } 814 } 815 816 # Takes two args: orig profile, and command to run. 817 # Returns 1 if we should keep going, or 0 if we were asked to quit 818 sub InteractiveCommand { 819 my($orig_profile, $symbols, $libs, $total, $command) = @_; 820 $_ = $command; # just to make future m//'s easier 821 if (!defined($_)) { 822 print STDERR "\n"; 823 return 0; 824 } 825 if (m/^\s*quit/) { 826 return 0; 827 } 828 if (m/^\s*help/) { 829 InteractiveHelpMessage(); 830 return 1; 831 } 832 # Clear all the mode options -- mode is controlled by "$command" 833 $main::opt_text = 0; 834 $main::opt_callgrind = 0; 835 $main::opt_disasm = 0; 836 $main::opt_list = 0; 837 $main::opt_gv = 0; 838 $main::opt_evince = 0; 839 $main::opt_cum = 0; 840 841 if (m/^\s*(text|top)(\d*)\s*(.*)/) { 842 $main::opt_text = 1; 843 844 my $line_limit = ($2 ne "") ? int($2) : 10; 845 846 my $routine; 847 my $ignore; 848 ($routine, $ignore) = ParseInteractiveArgs($3); 849 850 my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore); 851 my $reduced = ReduceProfile($symbols, $profile); 852 853 # Get derived profiles 854 my $flat = FlatProfile($reduced); 855 my $cumulative = CumulativeProfile($reduced); 856 857 PrintText($symbols, $flat, $cumulative, $line_limit); 858 return 1; 859 } 860 if (m/^\s*callgrind\s*([^ \n]*)/) { 861 $main::opt_callgrind = 1; 862 863 # Get derived profiles 864 my $calls = ExtractCalls($symbols, $orig_profile); 865 my $filename = $1; 866 if ( $1 eq '' ) { 867 $filename = TempName($main::next_tmpfile, "callgrind"); 868 } 869 PrintCallgrind($calls, $filename); 870 if ( $1 eq '' ) { 871 RunKcachegrind($filename, " & "); 872 $main::next_tmpfile++; 873 } 874 875 return 1; 876 } 877 if (m/^\s*(web)?list\s*(.+)/) { 878 my $html = (defined($1) && ($1 eq "web")); 879 $main::opt_list = 1; 880 881 my $routine; 882 my $ignore; 883 ($routine, $ignore) = ParseInteractiveArgs($2); 884 885 my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore); 886 my $reduced = ReduceProfile($symbols, $profile); 887 888 # Get derived profiles 889 my $flat = FlatProfile($reduced); 890 my $cumulative = CumulativeProfile($reduced); 891 892 PrintListing($total, $libs, $flat, $cumulative, $routine, $html); 893 return 1; 894 } 895 if (m/^\s*disasm\s*(.+)/) { 896 $main::opt_disasm = 1; 897 898 my $routine; 899 my $ignore; 900 ($routine, $ignore) = ParseInteractiveArgs($1); 901 902 # Process current profile to account for various settings 903 my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore); 904 my $reduced = ReduceProfile($symbols, $profile); 905 906 # Get derived profiles 907 my $flat = FlatProfile($reduced); 908 my $cumulative = CumulativeProfile($reduced); 909 910 PrintDisassembly($libs, $flat, $cumulative, $routine); 911 return 1; 912 } 913 if (m/^\s*(gv|web|evince)\s*(.*)/) { 914 $main::opt_gv = 0; 915 $main::opt_evince = 0; 916 $main::opt_web = 0; 917 if ($1 eq "gv") { 918 $main::opt_gv = 1; 919 } elsif ($1 eq "evince") { 920 $main::opt_evince = 1; 921 } elsif ($1 eq "web") { 922 $main::opt_web = 1; 923 } 924 925 my $focus; 926 my $ignore; 927 ($focus, $ignore) = ParseInteractiveArgs($2); 928 929 # Process current profile to account for various settings 930 my $profile = ProcessProfile($total, $orig_profile, $symbols, 931 $focus, $ignore); 932 my $reduced = ReduceProfile($symbols, $profile); 933 934 # Get derived profiles 935 my $flat = FlatProfile($reduced); 936 my $cumulative = CumulativeProfile($reduced); 937 938 if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) { 939 if ($main::opt_gv) { 940 RunGV(TempName($main::next_tmpfile, "ps"), " &"); 941 } elsif ($main::opt_evince) { 942 RunEvince(TempName($main::next_tmpfile, "pdf"), " &"); 943 } elsif ($main::opt_web) { 944 RunWeb(TempName($main::next_tmpfile, "svg")); 945 } 946 $main::next_tmpfile++; 947 } 948 return 1; 949 } 950 if (m/^\s*$/) { 951 return 1; 952 } 953 print STDERR "Unknown command: try 'help'.\n"; 954 return 1; 955 } 956 957 958 sub ProcessProfile { 959 my $total_count = shift; 960 my $orig_profile = shift; 961 my $symbols = shift; 962 my $focus = shift; 963 my $ignore = shift; 964 965 # Process current profile to account for various settings 966 my $profile = $orig_profile; 967 printf("Total: %s %s\n", Unparse($total_count), Units()); 968 if ($focus ne '') { 969 $profile = FocusProfile($symbols, $profile, $focus); 970 my $focus_count = TotalProfile($profile); 971 printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n", 972 $focus, 973 Unparse($focus_count), Units(), 974 Unparse($total_count), ($focus_count*100.0) / $total_count); 975 } 976 if ($ignore ne '') { 977 $profile = IgnoreProfile($symbols, $profile, $ignore); 978 my $ignore_count = TotalProfile($profile); 979 printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n", 980 $ignore, 981 Unparse($ignore_count), Units(), 982 Unparse($total_count), 983 ($ignore_count*100.0) / $total_count); 984 } 985 986 return $profile; 987 } 988 989 sub InteractiveHelpMessage { 990 print STDERR <<ENDOFHELP; 991 Interactive pprof mode 992 993 Commands: 994 gv 995 gv [focus] [-ignore1] [-ignore2] 996 Show graphical hierarchical display of current profile. Without 997 any arguments, shows all samples in the profile. With the optional 998 "focus" argument, restricts the samples shown to just those where 999 the "focus" regular expression matches a routine name on the stack 1000 trace. 1001 1002 web 1003 web [focus] [-ignore1] [-ignore2] 1004 Like GV, but displays profile in your web browser instead of using 1005 Ghostview. Works best if your web browser is already running. 1006 To change the browser that gets used: 1007 On Linux, set the /etc/alternatives/gnome-www-browser symlink. 1008 On OS X, change the Finder association for SVG files. 1009 1010 list [routine_regexp] [-ignore1] [-ignore2] 1011 Show source listing of routines whose names match "routine_regexp" 1012 1013 weblist [routine_regexp] [-ignore1] [-ignore2] 1014 Displays a source listing of routines whose names match "routine_regexp" 1015 in a web browser. You can click on source lines to view the 1016 corresponding disassembly. 1017 1018 top [--cum] [-ignore1] [-ignore2] 1019 top20 [--cum] [-ignore1] [-ignore2] 1020 top37 [--cum] [-ignore1] [-ignore2] 1021 Show top lines ordered by flat profile count, or cumulative count 1022 if --cum is specified. If a number is present after 'top', the 1023 top K routines will be shown (defaults to showing the top 10) 1024 1025 disasm [routine_regexp] [-ignore1] [-ignore2] 1026 Show disassembly of routines whose names match "routine_regexp", 1027 annotated with sample counts. 1028 1029 callgrind 1030 callgrind [filename] 1031 Generates callgrind file. If no filename is given, kcachegrind is called. 1032 1033 help - This listing 1034 quit or ^D - End pprof 1035 1036 For commands that accept optional -ignore tags, samples where any routine in 1037 the stack trace matches the regular expression in any of the -ignore 1038 parameters will be ignored. 1039 1040 Further pprof details are available at this location (or one similar): 1041 1042 /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html 1043 /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html 1044 1045 ENDOFHELP 1046 } 1047 sub ParseInteractiveArgs { 1048 my $args = shift; 1049 my $focus = ""; 1050 my $ignore = ""; 1051 my @x = split(/ +/, $args); 1052 foreach $a (@x) { 1053 if ($a =~ m/^(--|-)lines$/) { 1054 $main::opt_lines = 1; 1055 } elsif ($a =~ m/^(--|-)cum$/) { 1056 $main::opt_cum = 1; 1057 } elsif ($a =~ m/^-(.*)/) { 1058 $ignore .= (($ignore ne "") ? "|" : "" ) . $1; 1059 } else { 1060 $focus .= (($focus ne "") ? "|" : "" ) . $a; 1061 } 1062 } 1063 if ($ignore ne "") { 1064 print STDERR "Ignoring samples in call stacks that match '$ignore'\n"; 1065 } 1066 return ($focus, $ignore); 1067 } 1068 1069 ##### Output code ##### 1070 1071 sub TempName { 1072 my $fnum = shift; 1073 my $ext = shift; 1074 my $file = "$main::tmpfile_ps.$fnum.$ext"; 1075 $main::tempnames{$file} = 1; 1076 return $file; 1077 } 1078 1079 # Print profile data in packed binary format (64-bit) to standard out 1080 sub PrintProfileData { 1081 my $profile = shift; 1082 1083 # print header (64-bit style) 1084 # (zero) (header-size) (version) (sample-period) (zero) 1085 print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0); 1086 1087 foreach my $k (keys(%{$profile})) { 1088 my $count = $profile->{$k}; 1089 my @addrs = split(/\n/, $k); 1090 if ($#addrs >= 0) { 1091 my $depth = $#addrs + 1; 1092 # int(foo / 2**32) is the only reliable way to get rid of bottom 1093 # 32 bits on both 32- and 64-bit systems. 1094 print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32)); 1095 print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32)); 1096 1097 foreach my $full_addr (@addrs) { 1098 my $addr = $full_addr; 1099 $addr =~ s/0x0*//; # strip off leading 0x, zeroes 1100 if (length($addr) > 16) { 1101 print STDERR "Invalid address in profile: $full_addr\n"; 1102 next; 1103 } 1104 my $low_addr = substr($addr, -8); # get last 8 hex chars 1105 my $high_addr = substr($addr, -16, 8); # get up to 8 more hex chars 1106 print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr)); 1107 } 1108 } 1109 } 1110 } 1111 1112 # Print symbols and profile data 1113 sub PrintSymbolizedProfile { 1114 my $symbols = shift; 1115 my $profile = shift; 1116 my $prog = shift; 1117 1118 $SYMBOL_PAGE =~ m,[^/]+$,; # matches everything after the last slash 1119 my $symbol_marker = $&; 1120 1121 print '--- ', $symbol_marker, "\n"; 1122 if (defined($prog)) { 1123 print 'binary=', $prog, "\n"; 1124 } 1125 while (my ($pc, $name) = each(%{$symbols})) { 1126 my $sep = ' '; 1127 print '0x', $pc; 1128 # We have a list of function names, which include the inlined 1129 # calls. They are separated (and terminated) by --, which is 1130 # illegal in function names. 1131 for (my $j = 2; $j <= $#{$name}; $j += 3) { 1132 print $sep, $name->[$j]; 1133 $sep = '--'; 1134 } 1135 print "\n"; 1136 } 1137 print '---', "\n"; 1138 1139 $PROFILE_PAGE =~ m,[^/]+$,; # matches everything after the last slash 1140 my $profile_marker = $&; 1141 print '--- ', $profile_marker, "\n"; 1142 if (defined($main::collected_profile)) { 1143 # if used with remote fetch, simply dump the collected profile to output. 1144 open(SRC, "<$main::collected_profile"); 1145 while (<SRC>) { 1146 print $_; 1147 } 1148 close(SRC); 1149 } else { 1150 # dump a cpu-format profile to standard out 1151 PrintProfileData($profile); 1152 } 1153 } 1154 1155 # Print text output 1156 sub PrintText { 1157 my $symbols = shift; 1158 my $flat = shift; 1159 my $cumulative = shift; 1160 my $line_limit = shift; 1161 1162 my $total = TotalProfile($flat); 1163 1164 # Which profile to sort by? 1165 my $s = $main::opt_cum ? $cumulative : $flat; 1166 1167 my $running_sum = 0; 1168 my $lines = 0; 1169 foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b } 1170 keys(%{$cumulative})) { 1171 my $f = GetEntry($flat, $k); 1172 my $c = GetEntry($cumulative, $k); 1173 $running_sum += $f; 1174 1175 my $sym = $k; 1176 if (exists($symbols->{$k})) { 1177 $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1]; 1178 if ($main::opt_addresses) { 1179 $sym = $k . " " . $sym; 1180 } 1181 } 1182 1183 if ($f != 0 || $c != 0) { 1184 printf("%8s %6s %6s %8s %6s %s\n", 1185 Unparse($f), 1186 Percent($f, $total), 1187 Percent($running_sum, $total), 1188 Unparse($c), 1189 Percent($c, $total), 1190 $sym); 1191 } 1192 $lines++; 1193 last if ($line_limit >= 0 && $lines >= $line_limit); 1194 } 1195 } 1196 1197 # Callgrind format has a compression for repeated function and file 1198 # names. You show the name the first time, and just use its number 1199 # subsequently. This can cut down the file to about a third or a 1200 # quarter of its uncompressed size. $key and $val are the key/value 1201 # pair that would normally be printed by callgrind; $map is a map from 1202 # value to number. 1203 sub CompressedCGName { 1204 my($key, $val, $map) = @_; 1205 my $idx = $map->{$val}; 1206 # For very short keys, providing an index hurts rather than helps. 1207 if (length($val) <= 3) { 1208 return "$key=$val\n"; 1209 } elsif (defined($idx)) { 1210 return "$key=($idx)\n"; 1211 } else { 1212 # scalar(keys $map) gives the number of items in the map. 1213 $idx = scalar(keys(%{$map})) + 1; 1214 $map->{$val} = $idx; 1215 return "$key=($idx) $val\n"; 1216 } 1217 } 1218 1219 # Print the call graph in a way that's suiteable for callgrind. 1220 sub PrintCallgrind { 1221 my $calls = shift; 1222 my $filename; 1223 my %filename_to_index_map; 1224 my %fnname_to_index_map; 1225 1226 if ($main::opt_interactive) { 1227 $filename = shift; 1228 print STDERR "Writing callgrind file to '$filename'.\n" 1229 } else { 1230 $filename = "&STDOUT"; 1231 } 1232 open(CG, ">$filename"); 1233 printf CG ("events: Hits\n\n"); 1234 foreach my $call ( map { $_->[0] } 1235 sort { $a->[1] cmp $b ->[1] || 1236 $a->[2] <=> $b->[2] } 1237 map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/; 1238 [$_, $1, $2] } 1239 keys %$calls ) { 1240 my $count = int($calls->{$call}); 1241 $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/; 1242 my ( $caller_file, $caller_line, $caller_function, 1243 $callee_file, $callee_line, $callee_function ) = 1244 ( $1, $2, $3, $5, $6, $7 ); 1245 1246 # TODO(csilvers): for better compression, collect all the 1247 # caller/callee_files and functions first, before printing 1248 # anything, and only compress those referenced more than once. 1249 printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map); 1250 printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map); 1251 if (defined $6) { 1252 printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map); 1253 printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map); 1254 printf CG ("calls=$count $callee_line\n"); 1255 } 1256 printf CG ("$caller_line $count\n\n"); 1257 } 1258 } 1259 1260 # Print disassembly for all all routines that match $main::opt_disasm 1261 sub PrintDisassembly { 1262 my $libs = shift; 1263 my $flat = shift; 1264 my $cumulative = shift; 1265 my $disasm_opts = shift; 1266 1267 my $total = TotalProfile($flat); 1268 1269 foreach my $lib (@{$libs}) { 1270 my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts); 1271 my $offset = AddressSub($lib->[1], $lib->[3]); 1272 foreach my $routine (sort ByName keys(%{$symbol_table})) { 1273 my $start_addr = $symbol_table->{$routine}->[0]; 1274 my $end_addr = $symbol_table->{$routine}->[1]; 1275 # See if there are any samples in this routine 1276 my $length = hex(AddressSub($end_addr, $start_addr)); 1277 my $addr = AddressAdd($start_addr, $offset); 1278 for (my $i = 0; $i < $length; $i++) { 1279 if (defined($cumulative->{$addr})) { 1280 PrintDisassembledFunction($lib->[0], $offset, 1281 $routine, $flat, $cumulative, 1282 $start_addr, $end_addr, $total); 1283 last; 1284 } 1285 $addr = AddressInc($addr); 1286 } 1287 } 1288 } 1289 } 1290 1291 # Return reference to array of tuples of the form: 1292 # [start_address, filename, linenumber, instruction, limit_address] 1293 # E.g., 1294 # ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"] 1295 sub Disassemble { 1296 my $prog = shift; 1297 my $offset = shift; 1298 my $start_addr = shift; 1299 my $end_addr = shift; 1300 1301 my $objdump = $obj_tool_map{"objdump"}; 1302 my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn", 1303 "--start-address=0x$start_addr", 1304 "--stop-address=0x$end_addr", $prog); 1305 open(OBJDUMP, "$cmd |") || error("$cmd: $!\n"); 1306 my @result = (); 1307 my $filename = ""; 1308 my $linenumber = -1; 1309 my $last = ["", "", "", ""]; 1310 while (<OBJDUMP>) { 1311 s/\r//g; # turn windows-looking lines into unix-looking lines 1312 chop; 1313 if (m|\s*([^:\s]+):(\d+)\s*$|) { 1314 # Location line of the form: 1315 # <filename>:<linenumber> 1316 $filename = $1; 1317 $linenumber = $2; 1318 } elsif (m/^ +([0-9a-f]+):\s*(.*)/) { 1319 # Disassembly line -- zero-extend address to full length 1320 my $addr = HexExtend($1); 1321 my $k = AddressAdd($addr, $offset); 1322 $last->[4] = $k; # Store ending address for previous instruction 1323 $last = [$k, $filename, $linenumber, $2, $end_addr]; 1324 push(@result, $last); 1325 } 1326 } 1327 close(OBJDUMP); 1328 return @result; 1329 } 1330 1331 # The input file should contain lines of the form /proc/maps-like 1332 # output (same format as expected from the profiles) or that looks 1333 # like hex addresses (like "0xDEADBEEF"). We will parse all 1334 # /proc/maps output, and for all the hex addresses, we will output 1335 # "short" symbol names, one per line, in the same order as the input. 1336 sub PrintSymbols { 1337 my $maps_and_symbols_file = shift; 1338 1339 # ParseLibraries expects pcs to be in a set. Fine by us... 1340 my @pclist = (); # pcs in sorted order 1341 my $pcs = {}; 1342 my $map = ""; 1343 foreach my $line (<$maps_and_symbols_file>) { 1344 $line =~ s/\r//g; # turn windows-looking lines into unix-looking lines 1345 if ($line =~ /\b(0x[0-9a-f]+)\b/i) { 1346 push(@pclist, HexExtend($1)); 1347 $pcs->{$pclist[-1]} = 1; 1348 } else { 1349 $map .= $line; 1350 } 1351 } 1352 1353 my $libs = ParseLibraries($main::prog, $map, $pcs); 1354 my $symbols = ExtractSymbols($libs, $pcs); 1355 1356 foreach my $pc (@pclist) { 1357 # ->[0] is the shortname, ->[2] is the full name 1358 print(($symbols->{$pc}->[0] || "??") . "\n"); 1359 } 1360 } 1361 1362 1363 # For sorting functions by name 1364 sub ByName { 1365 return ShortFunctionName($a) cmp ShortFunctionName($b); 1366 } 1367 1368 # Print source-listing for all all routines that match $list_opts 1369 sub PrintListing { 1370 my $total = shift; 1371 my $libs = shift; 1372 my $flat = shift; 1373 my $cumulative = shift; 1374 my $list_opts = shift; 1375 my $html = shift; 1376 1377 my $output = \*STDOUT; 1378 my $fname = ""; 1379 1380 if ($html) { 1381 # Arrange to write the output to a temporary file 1382 $fname = TempName($main::next_tmpfile, "html"); 1383 $main::next_tmpfile++; 1384 if (!open(TEMP, ">$fname")) { 1385 print STDERR "$fname: $!\n"; 1386 return; 1387 } 1388 $output = \*TEMP; 1389 print $output HtmlListingHeader(); 1390 printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n", 1391 $main::prog, Unparse($total), Units()); 1392 } 1393 1394 my $listed = 0; 1395 foreach my $lib (@{$libs}) { 1396 my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts); 1397 my $offset = AddressSub($lib->[1], $lib->[3]); 1398 foreach my $routine (sort ByName keys(%{$symbol_table})) { 1399 # Print if there are any samples in this routine 1400 my $start_addr = $symbol_table->{$routine}->[0]; 1401 my $end_addr = $symbol_table->{$routine}->[1]; 1402 my $length = hex(AddressSub($end_addr, $start_addr)); 1403 my $addr = AddressAdd($start_addr, $offset); 1404 for (my $i = 0; $i < $length; $i++) { 1405 if (defined($cumulative->{$addr})) { 1406 $listed += PrintSource( 1407 $lib->[0], $offset, 1408 $routine, $flat, $cumulative, 1409 $start_addr, $end_addr, 1410 $html, 1411 $output); 1412 last; 1413 } 1414 $addr = AddressInc($addr); 1415 } 1416 } 1417 } 1418 1419 if ($html) { 1420 if ($listed > 0) { 1421 print $output HtmlListingFooter(); 1422 close($output); 1423 RunWeb($fname); 1424 } else { 1425 close($output); 1426 unlink($fname); 1427 } 1428 } 1429 } 1430 1431 sub HtmlListingHeader { 1432 return <<'EOF'; 1433 <DOCTYPE html> 1434 <html> 1435 <head> 1436 <title>Pprof listing</title> 1437 <style type="text/css"> 1438 body { 1439 font-family: sans-serif; 1440 } 1441 h1 { 1442 font-size: 1.5em; 1443 margin-bottom: 4px; 1444 } 1445 .legend { 1446 font-size: 1.25em; 1447 } 1448 .line { 1449 color: #aaaaaa; 1450 } 1451 .nop { 1452 color: #aaaaaa; 1453 } 1454 .unimportant { 1455 color: #cccccc; 1456 } 1457 .disasmloc { 1458 color: #000000; 1459 } 1460 .deadsrc { 1461 cursor: pointer; 1462 } 1463 .deadsrc:hover { 1464 background-color: #eeeeee; 1465 } 1466 .livesrc { 1467 color: #0000ff; 1468 cursor: pointer; 1469 } 1470 .livesrc:hover { 1471 background-color: #eeeeee; 1472 } 1473 .asm { 1474 color: #008800; 1475 display: none; 1476 } 1477 </style> 1478 <script type="text/javascript"> 1479 function pprof_toggle_asm(e) { 1480 var target; 1481 if (!e) e = window.event; 1482 if (e.target) target = e.target; 1483 else if (e.srcElement) target = e.srcElement; 1484 1485 if (target) { 1486 var asm = target.nextSibling; 1487 if (asm && asm.className == "asm") { 1488 asm.style.display = (asm.style.display == "block" ? "" : "block"); 1489 e.preventDefault(); 1490 return false; 1491 } 1492 } 1493 } 1494 </script> 1495 </head> 1496 <body> 1497 EOF 1498 } 1499 1500 sub HtmlListingFooter { 1501 return <<'EOF'; 1502 </body> 1503 </html> 1504 EOF 1505 } 1506 1507 sub HtmlEscape { 1508 my $text = shift; 1509 $text =~ s/&/&/g; 1510 $text =~ s/</</g; 1511 $text =~ s/>/>/g; 1512 return $text; 1513 } 1514 1515 # Returns the indentation of the line, if it has any non-whitespace 1516 # characters. Otherwise, returns -1. 1517 sub Indentation { 1518 my $line = shift; 1519 if (m/^(\s*)\S/) { 1520 return length($1); 1521 } else { 1522 return -1; 1523 } 1524 } 1525 1526 # If the symbol table contains inlining info, Disassemble() may tag an 1527 # instruction with a location inside an inlined function. But for 1528 # source listings, we prefer to use the location in the function we 1529 # are listing. So use MapToSymbols() to fetch full location 1530 # information for each instruction and then pick out the first 1531 # location from a location list (location list contains callers before 1532 # callees in case of inlining). 1533 # 1534 # After this routine has run, each entry in $instructions contains: 1535 # [0] start address 1536 # [1] filename for function we are listing 1537 # [2] line number for function we are listing 1538 # [3] disassembly 1539 # [4] limit address 1540 # [5] most specific filename (may be different from [1] due to inlining) 1541 # [6] most specific line number (may be different from [2] due to inlining) 1542 sub GetTopLevelLineNumbers { 1543 my ($lib, $offset, $instructions) = @_; 1544 my $pcs = []; 1545 for (my $i = 0; $i <= $#{$instructions}; $i++) { 1546 push(@{$pcs}, $instructions->[$i]->[0]); 1547 } 1548 my $symbols = {}; 1549 MapToSymbols($lib, $offset, $pcs, $symbols); 1550 for (my $i = 0; $i <= $#{$instructions}; $i++) { 1551 my $e = $instructions->[$i]; 1552 push(@{$e}, $e->[1]); 1553 push(@{$e}, $e->[2]); 1554 my $addr = $e->[0]; 1555 my $sym = $symbols->{$addr}; 1556 if (defined($sym)) { 1557 if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) { 1558 $e->[1] = $1; # File name 1559 $e->[2] = $2; # Line number 1560 } 1561 } 1562 } 1563 } 1564 1565 # Print source-listing for one routine 1566 sub PrintSource { 1567 my $prog = shift; 1568 my $offset = shift; 1569 my $routine = shift; 1570 my $flat = shift; 1571 my $cumulative = shift; 1572 my $start_addr = shift; 1573 my $end_addr = shift; 1574 my $html = shift; 1575 my $output = shift; 1576 1577 # Disassemble all instructions (just to get line numbers) 1578 my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr); 1579 GetTopLevelLineNumbers($prog, $offset, \@instructions); 1580 1581 # Hack 1: assume that the first source file encountered in the 1582 # disassembly contains the routine 1583 my $filename = undef; 1584 for (my $i = 0; $i <= $#instructions; $i++) { 1585 if ($instructions[$i]->[2] >= 0) { 1586 $filename = $instructions[$i]->[1]; 1587 last; 1588 } 1589 } 1590 if (!defined($filename)) { 1591 print STDERR "no filename found in $routine\n"; 1592 return 0; 1593 } 1594 1595 # Hack 2: assume that the largest line number from $filename is the 1596 # end of the procedure. This is typically safe since if P1 contains 1597 # an inlined call to P2, then P2 usually occurs earlier in the 1598 # source file. If this does not work, we might have to compute a 1599 # density profile or just print all regions we find. 1600 my $lastline = 0; 1601 for (my $i = 0; $i <= $#instructions; $i++) { 1602 my $f = $instructions[$i]->[1]; 1603 my $l = $instructions[$i]->[2]; 1604 if (($f eq $filename) && ($l > $lastline)) { 1605 $lastline = $l; 1606 } 1607 } 1608 1609 # Hack 3: assume the first source location from "filename" is the start of 1610 # the source code. 1611 my $firstline = 1; 1612 for (my $i = 0; $i <= $#instructions; $i++) { 1613 if ($instructions[$i]->[1] eq $filename) { 1614 $firstline = $instructions[$i]->[2]; 1615 last; 1616 } 1617 } 1618 1619 # Hack 4: Extend last line forward until its indentation is less than 1620 # the indentation we saw on $firstline 1621 my $oldlastline = $lastline; 1622 { 1623 if (!open(FILE, "<$filename")) { 1624 print STDERR "$filename: $!\n"; 1625 return 0; 1626 } 1627 my $l = 0; 1628 my $first_indentation = -1; 1629 while (<FILE>) { 1630 s/\r//g; # turn windows-looking lines into unix-looking lines 1631 $l++; 1632 my $indent = Indentation($_); 1633 if ($l >= $firstline) { 1634 if ($first_indentation < 0 && $indent >= 0) { 1635 $first_indentation = $indent; 1636 last if ($first_indentation == 0); 1637 } 1638 } 1639 if ($l >= $lastline && $indent >= 0) { 1640 if ($indent >= $first_indentation) { 1641 $lastline = $l+1; 1642 } else { 1643 last; 1644 } 1645 } 1646 } 1647 close(FILE); 1648 } 1649 1650 # Assign all samples to the range $firstline,$lastline, 1651 # Hack 4: If an instruction does not occur in the range, its samples 1652 # are moved to the next instruction that occurs in the range. 1653 my $samples1 = {}; # Map from line number to flat count 1654 my $samples2 = {}; # Map from line number to cumulative count 1655 my $running1 = 0; # Unassigned flat counts 1656 my $running2 = 0; # Unassigned cumulative counts 1657 my $total1 = 0; # Total flat counts 1658 my $total2 = 0; # Total cumulative counts 1659 my %disasm = (); # Map from line number to disassembly 1660 my $running_disasm = ""; # Unassigned disassembly 1661 my $skip_marker = "---\n"; 1662 if ($html) { 1663 $skip_marker = ""; 1664 for (my $l = $firstline; $l <= $lastline; $l++) { 1665 $disasm{$l} = ""; 1666 } 1667 } 1668 my $last_dis_filename = ''; 1669 my $last_dis_linenum = -1; 1670 my $last_touched_line = -1; # To detect gaps in disassembly for a line 1671 foreach my $e (@instructions) { 1672 # Add up counts for all address that fall inside this instruction 1673 my $c1 = 0; 1674 my $c2 = 0; 1675 for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) { 1676 $c1 += GetEntry($flat, $a); 1677 $c2 += GetEntry($cumulative, $a); 1678 } 1679 1680 if ($html) { 1681 my $dis = sprintf(" %6s %6s \t\t%8s: %s ", 1682 HtmlPrintNumber($c1), 1683 HtmlPrintNumber($c2), 1684 UnparseAddress($offset, $e->[0]), 1685 CleanDisassembly($e->[3])); 1686 1687 # Append the most specific source line associated with this instruction 1688 if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) }; 1689 $dis = HtmlEscape($dis); 1690 my $f = $e->[5]; 1691 my $l = $e->[6]; 1692 if ($f ne $last_dis_filename) { 1693 $dis .= sprintf("<span class=disasmloc>%s:%d</span>", 1694 HtmlEscape(CleanFileName($f)), $l); 1695 } elsif ($l ne $last_dis_linenum) { 1696 # De-emphasize the unchanged file name portion 1697 $dis .= sprintf("<span class=unimportant>%s</span>" . 1698 "<span class=disasmloc>:%d</span>", 1699 HtmlEscape(CleanFileName($f)), $l); 1700 } else { 1701 # De-emphasize the entire location 1702 $dis .= sprintf("<span class=unimportant>%s:%d</span>", 1703 HtmlEscape(CleanFileName($f)), $l); 1704 } 1705 $last_dis_filename = $f; 1706 $last_dis_linenum = $l; 1707 $running_disasm .= $dis; 1708 $running_disasm .= "\n"; 1709 } 1710 1711 $running1 += $c1; 1712 $running2 += $c2; 1713 $total1 += $c1; 1714 $total2 += $c2; 1715 my $file = $e->[1]; 1716 my $line = $e->[2]; 1717 if (($file eq $filename) && 1718 ($line >= $firstline) && 1719 ($line <= $lastline)) { 1720 # Assign all accumulated samples to this line 1721 AddEntry($samples1, $line, $running1); 1722 AddEntry($samples2, $line, $running2); 1723 $running1 = 0; 1724 $running2 = 0; 1725 if ($html) { 1726 if ($line != $last_touched_line && $disasm{$line} ne '') { 1727 $disasm{$line} .= "\n"; 1728 } 1729 $disasm{$line} .= $running_disasm; 1730 $running_disasm = ''; 1731 $last_touched_line = $line; 1732 } 1733 } 1734 } 1735 1736 # Assign any leftover samples to $lastline 1737 AddEntry($samples1, $lastline, $running1); 1738 AddEntry($samples2, $lastline, $running2); 1739 if ($html) { 1740 if ($lastline != $last_touched_line && $disasm{$lastline} ne '') { 1741 $disasm{$lastline} .= "\n"; 1742 } 1743 $disasm{$lastline} .= $running_disasm; 1744 } 1745 1746 if ($html) { 1747 printf $output ( 1748 "<h1>%s</h1>%s\n<pre onClick=\"pprof_toggle_asm()\">\n" . 1749 "Total:%6s %6s (flat / cumulative %s)\n", 1750 HtmlEscape(ShortFunctionName($routine)), 1751 HtmlEscape(CleanFileName($filename)), 1752 Unparse($total1), 1753 Unparse($total2), 1754 Units()); 1755 } else { 1756 printf $output ( 1757 "ROUTINE ====================== %s in %s\n" . 1758 "%6s %6s Total %s (flat / cumulative)\n", 1759 ShortFunctionName($routine), 1760 CleanFileName($filename), 1761 Unparse($total1), 1762 Unparse($total2), 1763 Units()); 1764 } 1765 if (!open(FILE, "<$filename")) { 1766 print STDERR "$filename: $!\n"; 1767 return 0; 1768 } 1769 my $l = 0; 1770 while (<FILE>) { 1771 s/\r//g; # turn windows-looking lines into unix-looking lines 1772 $l++; 1773 if ($l >= $firstline - 5 && 1774 (($l <= $oldlastline + 5) || ($l <= $lastline))) { 1775 chop; 1776 my $text = $_; 1777 if ($l == $firstline) { print $output $skip_marker; } 1778 my $n1 = GetEntry($samples1, $l); 1779 my $n2 = GetEntry($samples2, $l); 1780 if ($html) { 1781 # Emit a span that has one of the following classes: 1782 # livesrc -- has samples 1783 # deadsrc -- has disassembly, but with no samples 1784 # nop -- has no matching disasembly 1785 # Also emit an optional span containing disassembly. 1786 my $dis = $disasm{$l}; 1787 my $asm = ""; 1788 if (defined($dis) && $dis ne '') { 1789 $asm = "<span class=\"asm\">" . $dis . "</span>"; 1790 } 1791 my $source_class = (($n1 + $n2 > 0) 1792 ? "livesrc" 1793 : (($asm ne "") ? "deadsrc" : "nop")); 1794 printf $output ( 1795 "<span class=\"line\">%5d</span> " . 1796 "<span class=\"%s\">%6s %6s %s</span>%s\n", 1797 $l, $source_class, 1798 HtmlPrintNumber($n1), 1799 HtmlPrintNumber($n2), 1800 HtmlEscape($text), 1801 $asm); 1802 } else { 1803 printf $output( 1804 "%6s %6s %4d: %s\n", 1805 UnparseAlt($n1), 1806 UnparseAlt($n2), 1807 $l, 1808 $text); 1809 } 1810 if ($l == $lastline) { print $output $skip_marker; } 1811 }; 1812 } 1813 close(FILE); 1814 if ($html) { 1815 print $output "</pre>\n"; 1816 } 1817 return 1; 1818 } 1819 1820 # Return the source line for the specified file/linenumber. 1821 # Returns undef if not found. 1822 sub SourceLine { 1823 my $file = shift; 1824 my $line = shift; 1825 1826 # Look in cache 1827 if (!defined($main::source_cache{$file})) { 1828 if (100 < scalar keys(%main::source_cache)) { 1829 # Clear the cache when it gets too big 1830 $main::source_cache = (); 1831 } 1832 1833 # Read all lines from the file 1834 if (!open(FILE, "<$file")) { 1835 print STDERR "$file: $!\n"; 1836 $main::source_cache{$file} = []; # Cache the negative result 1837 return undef; 1838 } 1839 my $lines = []; 1840 push(@{$lines}, ""); # So we can use 1-based line numbers as indices 1841 while (<FILE>) { 1842 push(@{$lines}, $_); 1843 } 1844 close(FILE); 1845 1846 # Save the lines in the cache 1847 $main::source_cache{$file} = $lines; 1848 } 1849 1850 my $lines = $main::source_cache{$file}; 1851 if (($line < 0) || ($line > $#{$lines})) { 1852 return undef; 1853 } else { 1854 return $lines->[$line]; 1855 } 1856 } 1857 1858 # Print disassembly for one routine with interspersed source if available 1859 sub PrintDisassembledFunction { 1860 my $prog = shift; 1861 my $offset = shift; 1862 my $routine = shift; 1863 my $flat = shift; 1864 my $cumulative = shift; 1865 my $start_addr = shift; 1866 my $end_addr = shift; 1867 my $total = shift; 1868 1869 # Disassemble all instructions 1870 my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr); 1871 1872 # Make array of counts per instruction 1873 my @flat_count = (); 1874 my @cum_count = (); 1875 my $flat_total = 0; 1876 my $cum_total = 0; 1877 foreach my $e (@instructions) { 1878 # Add up counts for all address that fall inside this instruction 1879 my $c1 = 0; 1880 my $c2 = 0; 1881 for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) { 1882 $c1 += GetEntry($flat, $a); 1883 $c2 += GetEntry($cumulative, $a); 1884 } 1885 push(@flat_count, $c1); 1886 push(@cum_count, $c2); 1887 $flat_total += $c1; 1888 $cum_total += $c2; 1889 } 1890 1891 # Print header with total counts 1892 printf("ROUTINE ====================== %s\n" . 1893 "%6s %6s %s (flat, cumulative) %.1f%% of total\n", 1894 ShortFunctionName($routine), 1895 Unparse($flat_total), 1896 Unparse($cum_total), 1897 Units(), 1898 ($cum_total * 100.0) / $total); 1899 1900 # Process instructions in order 1901 my $current_file = ""; 1902 for (my $i = 0; $i <= $#instructions; ) { 1903 my $e = $instructions[$i]; 1904 1905 # Print the new file name whenever we switch files 1906 if ($e->[1] ne $current_file) { 1907 $current_file = $e->[1]; 1908 my $fname = $current_file; 1909 $fname =~ s|^\./||; # Trim leading "./" 1910 1911 # Shorten long file names 1912 if (length($fname) >= 58) { 1913 $fname = "..." . substr($fname, -55); 1914 } 1915 printf("-------------------- %s\n", $fname); 1916 } 1917 1918 # TODO: Compute range of lines to print together to deal with 1919 # small reorderings. 1920 my $first_line = $e->[2]; 1921 my $last_line = $first_line; 1922 my %flat_sum = (); 1923 my %cum_sum = (); 1924 for (my $l = $first_line; $l <= $last_line; $l++) { 1925 $flat_sum{$l} = 0; 1926 $cum_sum{$l} = 0; 1927 } 1928 1929 # Find run of instructions for this range of source lines 1930 my $first_inst = $i; 1931 while (($i <= $#instructions) && 1932 ($instructions[$i]->[2] >= $first_line) && 1933 ($instructions[$i]->[2] <= $last_line)) { 1934 $e = $instructions[$i]; 1935 $flat_sum{$e->[2]} += $flat_count[$i]; 1936 $cum_sum{$e->[2]} += $cum_count[$i]; 1937 $i++; 1938 } 1939 my $last_inst = $i - 1; 1940 1941 # Print source lines 1942 for (my $l = $first_line; $l <= $last_line; $l++) { 1943 my $line = SourceLine($current_file, $l); 1944 if (!defined($line)) { 1945 $line = "?\n"; 1946 next; 1947 } else { 1948 $line =~ s/^\s+//; 1949 } 1950 printf("%6s %6s %5d: %s", 1951 UnparseAlt($flat_sum{$l}), 1952 UnparseAlt($cum_sum{$l}), 1953 $l, 1954 $line); 1955 } 1956 1957 # Print disassembly 1958 for (my $x = $first_inst; $x <= $last_inst; $x++) { 1959 my $e = $instructions[$x]; 1960 printf("%6s %6s %8s: %6s\n", 1961 UnparseAlt($flat_count[$x]), 1962 UnparseAlt($cum_count[$x]), 1963 UnparseAddress($offset, $e->[0]), 1964 CleanDisassembly($e->[3])); 1965 } 1966 } 1967 } 1968 1969 # Print DOT graph 1970 sub PrintDot { 1971 my $prog = shift; 1972 my $symbols = shift; 1973 my $raw = shift; 1974 my $flat = shift; 1975 my $cumulative = shift; 1976 my $overall_total = shift; 1977 1978 # Get total 1979 my $local_total = TotalProfile($flat); 1980 my $nodelimit = int($main::opt_nodefraction * $local_total); 1981 my $edgelimit = int($main::opt_edgefraction * $local_total); 1982 my $nodecount = $main::opt_nodecount; 1983 1984 # Find nodes to include 1985 my @list = (sort { abs(GetEntry($cumulative, $b)) <=> 1986 abs(GetEntry($cumulative, $a)) 1987 || $a cmp $b } 1988 keys(%{$cumulative})); 1989 my $last = $nodecount - 1; 1990 if ($last > $#list) { 1991 $last = $#list; 1992 } 1993 while (($last >= 0) && 1994 (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) { 1995 $last--; 1996 } 1997 if ($last < 0) { 1998 print STDERR "No nodes to print\n"; 1999 return 0; 2000 } 2001 2002 if ($nodelimit > 0 || $edgelimit > 0) { 2003 printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n", 2004 Unparse($nodelimit), Units(), 2005 Unparse($edgelimit), Units()); 2006 } 2007 2008 # Open DOT output file 2009 my $output; 2010 my $escaped_dot = ShellEscape(@DOT); 2011 my $escaped_ps2pdf = ShellEscape(@PS2PDF); 2012 if ($main::opt_gv) { 2013 my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps")); 2014 $output = "| $escaped_dot -Tps2 >$escaped_outfile"; 2015 } elsif ($main::opt_evince) { 2016 my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf")); 2017 $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile"; 2018 } elsif ($main::opt_ps) { 2019 $output = "| $escaped_dot -Tps2"; 2020 } elsif ($main::opt_pdf) { 2021 $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -"; 2022 } elsif ($main::opt_web || $main::opt_svg) { 2023 # We need to post-process the SVG, so write to a temporary file always. 2024 my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg")); 2025 $output = "| $escaped_dot -Tsvg >$escaped_outfile"; 2026 } elsif ($main::opt_gif) { 2027 $output = "| $escaped_dot -Tgif"; 2028 } else { 2029 $output = ">&STDOUT"; 2030 } 2031 open(DOT, $output) || error("$output: $!\n"); 2032 2033 # Title 2034 printf DOT ("digraph \"%s; %s %s\" {\n", 2035 $prog, 2036 Unparse($overall_total), 2037 Units()); 2038 if ($main::opt_pdf) { 2039 # The output is more printable if we set the page size for dot. 2040 printf DOT ("size=\"8,11\"\n"); 2041 } 2042 printf DOT ("node [width=0.375,height=0.25];\n"); 2043 2044 # Print legend 2045 printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," . 2046 "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n", 2047 $prog, 2048 sprintf("Total %s: %s", Units(), Unparse($overall_total)), 2049 sprintf("Focusing on: %s", Unparse($local_total)), 2050 sprintf("Dropped nodes with <= %s abs(%s)", 2051 Unparse($nodelimit), Units()), 2052 sprintf("Dropped edges with <= %s %s", 2053 Unparse($edgelimit), Units()) 2054 ); 2055 2056 # Print nodes 2057 my %node = (); 2058 my $nextnode = 1; 2059 foreach my $a (@list[0..$last]) { 2060 # Pick font size 2061 my $f = GetEntry($flat, $a); 2062 my $c = GetEntry($cumulative, $a); 2063 2064 my $fs = 8; 2065 if ($local_total > 0) { 2066 $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total))); 2067 } 2068 2069 $node{$a} = $nextnode++; 2070 my $sym = $a; 2071 $sym =~ s/\s+/\\n/g; 2072 $sym =~ s/::/\\n/g; 2073 2074 # Extra cumulative info to print for non-leaves 2075 my $extra = ""; 2076 if ($f != $c) { 2077 $extra = sprintf("\\rof %s (%s)", 2078 Unparse($c), 2079 Percent($c, $local_total)); 2080 } 2081 my $style = ""; 2082 if ($main::opt_heapcheck) { 2083 if ($f > 0) { 2084 # make leak-causing nodes more visible (add a background) 2085 $style = ",style=filled,fillcolor=gray" 2086 } elsif ($f < 0) { 2087 # make anti-leak-causing nodes (which almost never occur) 2088 # stand out as well (triple border) 2089 $style = ",peripheries=3" 2090 } 2091 } 2092 2093 printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" . 2094 "\",shape=box,fontsize=%.1f%s];\n", 2095 $node{$a}, 2096 $sym, 2097 Unparse($f), 2098 Percent($f, $local_total), 2099 $extra, 2100 $fs, 2101 $style, 2102 ); 2103 } 2104 2105 # Get edges and counts per edge 2106 my %edge = (); 2107 my $n; 2108 my $fullname_to_shortname_map = {}; 2109 FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map); 2110 foreach my $k (keys(%{$raw})) { 2111 # TODO: omit low %age edges 2112 $n = $raw->{$k}; 2113 my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k); 2114 for (my $i = 1; $i <= $#translated; $i++) { 2115 my $src = $translated[$i]; 2116 my $dst = $translated[$i-1]; 2117 #next if ($src eq $dst); # Avoid self-edges? 2118 if (exists($node{$src}) && exists($node{$dst})) { 2119 my $edge_label = "$src\001$dst"; 2120 if (!exists($edge{$edge_label})) { 2121 $edge{$edge_label} = 0; 2122 } 2123 $edge{$edge_label} += $n; 2124 } 2125 } 2126 } 2127 2128 # Print edges (process in order of decreasing counts) 2129 my %indegree = (); # Number of incoming edges added per node so far 2130 my %outdegree = (); # Number of outgoing edges added per node so far 2131 foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) { 2132 my @x = split(/\001/, $e); 2133 $n = $edge{$e}; 2134 2135 # Initialize degree of kept incoming and outgoing edges if necessary 2136 my $src = $x[0]; 2137 my $dst = $x[1]; 2138 if (!exists($outdegree{$src})) { $outdegree{$src} = 0; } 2139 if (!exists($indegree{$dst})) { $indegree{$dst} = 0; } 2140 2141 my $keep; 2142 if ($indegree{$dst} == 0) { 2143 # Keep edge if needed for reachability 2144 $keep = 1; 2145 } elsif (abs($n) <= $edgelimit) { 2146 # Drop if we are below --edgefraction 2147 $keep = 0; 2148 } elsif ($outdegree{$src} >= $main::opt_maxdegree || 2149 $indegree{$dst} >= $main::opt_maxdegree) { 2150 # Keep limited number of in/out edges per node 2151 $keep = 0; 2152 } else { 2153 $keep = 1; 2154 } 2155 2156 if ($keep) { 2157 $outdegree{$src}++; 2158 $indegree{$dst}++; 2159 2160 # Compute line width based on edge count 2161 my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0); 2162 if ($fraction > 1) { $fraction = 1; } 2163 my $w = $fraction * 2; 2164 if ($w < 1 && ($main::opt_web || $main::opt_svg)) { 2165 # SVG output treats line widths < 1 poorly. 2166 $w = 1; 2167 } 2168 2169 # Dot sometimes segfaults if given edge weights that are too large, so 2170 # we cap the weights at a large value 2171 my $edgeweight = abs($n) ** 0.7; 2172 if ($edgeweight > 100000) { $edgeweight = 100000; } 2173 $edgeweight = int($edgeweight); 2174 2175 my $style = sprintf("setlinewidth(%f)", $w); 2176 if ($x[1] =~ m/\(inline\)/) { 2177 $style .= ",dashed"; 2178 } 2179 2180 # Use a slightly squashed function of the edge count as the weight 2181 printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n", 2182 $node{$x[0]}, 2183 $node{$x[1]}, 2184 Unparse($n), 2185 $edgeweight, 2186 $style); 2187 } 2188 } 2189 2190 print DOT ("}\n"); 2191 close(DOT); 2192 2193 if ($main::opt_web || $main::opt_svg) { 2194 # Rewrite SVG to be more usable inside web browser. 2195 RewriteSvg(TempName($main::next_tmpfile, "svg")); 2196 } 2197 2198 return 1; 2199 } 2200 2201 sub RewriteSvg { 2202 my $svgfile = shift; 2203 2204 open(SVG, $svgfile) || die "open temp svg: $!"; 2205 my @svg = <SVG>; 2206 close(SVG); 2207 unlink $svgfile; 2208 my $svg = join('', @svg); 2209 2210 # Dot's SVG output is 2211 # 2212 # <svg width="___" height="___" 2213 # viewBox="___" xmlns=...> 2214 # <g id="graph0" transform="..."> 2215 # ... 2216 # </g> 2217 # </svg> 2218 # 2219 # Change it to 2220 # 2221 # <svg width="100%" height="100%" 2222 # xmlns=...> 2223 # $svg_javascript 2224 # <g id="viewport" transform="translate(0,0)"> 2225 # <g id="graph0" transform="..."> 2226 # ... 2227 # </g> 2228 # </g> 2229 # </svg> 2230 2231 # Fix width, height; drop viewBox. 2232 $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/; 2233 2234 # Insert script, viewport <g> above first <g> 2235 my $svg_javascript = SvgJavascript(); 2236 my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n"; 2237 $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/; 2238 2239 # Insert final </g> above </svg>. 2240 $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/; 2241 $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/; 2242 2243 if ($main::opt_svg) { 2244 # --svg: write to standard output. 2245 print $svg; 2246 } else { 2247 # Write back to temporary file. 2248 open(SVG, ">$svgfile") || die "open $svgfile: $!"; 2249 print SVG $svg; 2250 close(SVG); 2251 } 2252 } 2253 2254 sub SvgJavascript { 2255 return <<'EOF'; 2256 <script type="text/ecmascript"><![CDATA[ 2257 // SVGPan 2258 // http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/ 2259 // Local modification: if(true || ...) below to force panning, never moving. 2260 2261 /** 2262 * SVGPan library 1.2 2263 * ==================== 2264 * 2265 * Given an unique existing element with id "viewport", including the 2266 * the library into any SVG adds the following capabilities: 2267 * 2268 * - Mouse panning 2269 * - Mouse zooming (using the wheel) 2270 * - Object dargging 2271 * 2272 * Known issues: 2273 * 2274 * - Zooming (while panning) on Safari has still some issues 2275 * 2276 * Releases: 2277 * 2278 * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui 2279 * Fixed a bug with browser mouse handler interaction 2280 * 2281 * 1.1, Wed Feb 3 17:39:33 GMT 2010, Zeng Xiaohui 2282 * Updated the zoom code to support the mouse wheel on Safari/Chrome 2283 * 2284 * 1.0, Andrea Leofreddi 2285 * First release 2286 * 2287 * This code is licensed under the following BSD license: 2288 * 2289 * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved. 2290 * 2291 * Redistribution and use in source and binary forms, with or without modification, are 2292 * permitted provided that the following conditions are met: 2293 * 2294 * 1. Redistributions of source code must retain the above copyright notice, this list of 2295 * conditions and the following disclaimer. 2296 * 2297 * 2. Redistributions in binary form must reproduce the above copyright notice, this list 2298 * of conditions and the following disclaimer in the documentation and/or other materials 2299 * provided with the distribution. 2300 * 2301 * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED 2302 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND 2303 * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR 2304 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 2305 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 2306 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON 2307 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 2308 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 2309 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 2310 * 2311 * The views and conclusions contained in the software and documentation are those of the 2312 * authors and should not be interpreted as representing official policies, either expressed 2313 * or implied, of Andrea Leofreddi. 2314 */ 2315 2316 var root = document.documentElement; 2317 2318 var state = 'none', stateTarget, stateOrigin, stateTf; 2319 2320 setupHandlers(root); 2321 2322 /** 2323 * Register handlers 2324 */ 2325 function setupHandlers(root){ 2326 setAttributes(root, { 2327 "onmouseup" : "add(evt)", 2328 "onmousedown" : "handleMouseDown(evt)", 2329 "onmousemove" : "handleMouseMove(evt)", 2330 "onmouseup" : "handleMouseUp(evt)", 2331 //"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element 2332 }); 2333 2334 if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0) 2335 window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari 2336 else 2337 window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others 2338 2339 var g = svgDoc.getElementById("svg"); 2340 g.width = "100%"; 2341 g.height = "100%"; 2342 } 2343 2344 /** 2345 * Instance an SVGPoint object with given event coordinates. 2346 */ 2347 function getEventPoint(evt) { 2348 var p = root.createSVGPoint(); 2349 2350 p.x = evt.clientX; 2351 p.y = evt.clientY; 2352 2353 return p; 2354 } 2355 2356 /** 2357 * Sets the current transform matrix of an element. 2358 */ 2359 function setCTM(element, matrix) { 2360 var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")"; 2361 2362 element.setAttribute("transform", s); 2363 } 2364 2365 /** 2366 * Dumps a matrix to a string (useful for debug). 2367 */ 2368 function dumpMatrix(matrix) { 2369 var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n 0, 0, 1 ]"; 2370 2371 return s; 2372 } 2373 2374 /** 2375 * Sets attributes of an element. 2376 */ 2377 function setAttributes(element, attributes){ 2378 for (i in attributes) 2379 element.setAttributeNS(null, i, attributes[i]); 2380 } 2381 2382 /** 2383 * Handle mouse move event. 2384 */ 2385 function handleMouseWheel(evt) { 2386 if(evt.preventDefault) 2387 evt.preventDefault(); 2388 2389 evt.returnValue = false; 2390 2391 var svgDoc = evt.target.ownerDocument; 2392 2393 var delta; 2394 2395 if(evt.wheelDelta) 2396 delta = evt.wheelDelta / 3600; // Chrome/Safari 2397 else 2398 delta = evt.detail / -90; // Mozilla 2399 2400 var z = 1 + delta; // Zoom factor: 0.9/1.1 2401 2402 var g = svgDoc.getElementById("viewport"); 2403 2404 var p = getEventPoint(evt); 2405 2406 p = p.matrixTransform(g.getCTM().inverse()); 2407 2408 // Compute new scale matrix in current mouse position 2409 var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y); 2410 2411 setCTM(g, g.getCTM().multiply(k)); 2412 2413 stateTf = stateTf.multiply(k.inverse()); 2414 } 2415 2416 /** 2417 * Handle mouse move event. 2418 */ 2419 function handleMouseMove(evt) { 2420 if(evt.preventDefault) 2421 evt.preventDefault(); 2422 2423 evt.returnValue = false; 2424 2425 var svgDoc = evt.target.ownerDocument; 2426 2427 var g = svgDoc.getElementById("viewport"); 2428 2429 if(state == 'pan') { 2430 // Pan mode 2431 var p = getEventPoint(evt).matrixTransform(stateTf); 2432 2433 setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y)); 2434 } else if(state == 'move') { 2435 // Move mode 2436 var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse()); 2437 2438 setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM())); 2439 2440 stateOrigin = p; 2441 } 2442 } 2443 2444 /** 2445 * Handle click event. 2446 */ 2447 function handleMouseDown(evt) { 2448 if(evt.preventDefault) 2449 evt.preventDefault(); 2450 2451 evt.returnValue = false; 2452 2453 var svgDoc = evt.target.ownerDocument; 2454 2455 var g = svgDoc.getElementById("viewport"); 2456 2457 if(true || evt.target.tagName == "svg") { 2458 // Pan mode 2459 state = 'pan'; 2460 2461 stateTf = g.getCTM().inverse(); 2462 2463 stateOrigin = getEventPoint(evt).matrixTransform(stateTf); 2464 } else { 2465 // Move mode 2466 state = 'move'; 2467 2468 stateTarget = evt.target; 2469 2470 stateTf = g.getCTM().inverse(); 2471 2472 stateOrigin = getEventPoint(evt).matrixTransform(stateTf); 2473 } 2474 } 2475 2476 /** 2477 * Handle mouse button release event. 2478 */ 2479 function handleMouseUp(evt) { 2480 if(evt.preventDefault) 2481 evt.preventDefault(); 2482 2483 evt.returnValue = false; 2484 2485 var svgDoc = evt.target.ownerDocument; 2486 2487 if(state == 'pan' || state == 'move') { 2488 // Quit pan mode 2489 state = ''; 2490 } 2491 } 2492 2493 ]]></script> 2494 EOF 2495 } 2496 2497 # Provides a map from fullname to shortname for cases where the 2498 # shortname is ambiguous. The symlist has both the fullname and 2499 # shortname for all symbols, which is usually fine, but sometimes -- 2500 # such as overloaded functions -- two different fullnames can map to 2501 # the same shortname. In that case, we use the address of the 2502 # function to disambiguate the two. This function fills in a map that 2503 # maps fullnames to modified shortnames in such cases. If a fullname 2504 # is not present in the map, the 'normal' shortname provided by the 2505 # symlist is the appropriate one to use. 2506 sub FillFullnameToShortnameMap { 2507 my $symbols = shift; 2508 my $fullname_to_shortname_map = shift; 2509 my $shortnames_seen_once = {}; 2510 my $shortnames_seen_more_than_once = {}; 2511 2512 foreach my $symlist (values(%{$symbols})) { 2513 # TODO(csilvers): deal with inlined symbols too. 2514 my $shortname = $symlist->[0]; 2515 my $fullname = $symlist->[2]; 2516 if ($fullname !~ /<[0-9a-fA-F]+>$/) { # fullname doesn't end in an address 2517 next; # the only collisions we care about are when addresses differ 2518 } 2519 if (defined($shortnames_seen_once->{$shortname}) && 2520 $shortnames_seen_once->{$shortname} ne $fullname) { 2521 $shortnames_seen_more_than_once->{$shortname} = 1; 2522 } else { 2523 $shortnames_seen_once->{$shortname} = $fullname; 2524 } 2525 } 2526 2527 foreach my $symlist (values(%{$symbols})) { 2528 my $shortname = $symlist->[0]; 2529 my $fullname = $symlist->[2]; 2530 # TODO(csilvers): take in a list of addresses we care about, and only 2531 # store in the map if $symlist->[1] is in that list. Saves space. 2532 next if defined($fullname_to_shortname_map->{$fullname}); 2533 if (defined($shortnames_seen_more_than_once->{$shortname})) { 2534 if ($fullname =~ /<0*([^>]*)>$/) { # fullname has address at end of it 2535 $fullname_to_shortname_map->{$fullname} = "$shortname\@$1"; 2536 } 2537 } 2538 } 2539 } 2540 2541 # Return a small number that identifies the argument. 2542 # Multiple calls with the same argument will return the same number. 2543 # Calls with different arguments will return different numbers. 2544 sub ShortIdFor { 2545 my $key = shift; 2546 my $id = $main::uniqueid{$key}; 2547 if (!defined($id)) { 2548 $id = keys(%main::uniqueid) + 1; 2549 $main::uniqueid{$key} = $id; 2550 } 2551 return $id; 2552 } 2553 2554 # Translate a stack of addresses into a stack of symbols 2555 sub TranslateStack { 2556 my $symbols = shift; 2557 my $fullname_to_shortname_map = shift; 2558 my $k = shift; 2559 2560 my @addrs = split(/\n/, $k); 2561 my @result = (); 2562 for (my $i = 0; $i <= $#addrs; $i++) { 2563 my $a = $addrs[$i]; 2564 2565 # Skip large addresses since they sometimes show up as fake entries on RH9 2566 if (length($a) > 8 && $a gt "7fffffffffffffff") { 2567 next; 2568 } 2569 2570 if ($main::opt_disasm || $main::opt_list) { 2571 # We want just the address for the key 2572 push(@result, $a); 2573 next; 2574 } 2575 2576 my $symlist = $symbols->{$a}; 2577 if (!defined($symlist)) { 2578 $symlist = [$a, "", $a]; 2579 } 2580 2581 # We can have a sequence of symbols for a particular entry 2582 # (more than one symbol in the case of inlining). Callers 2583 # come before callees in symlist, so walk backwards since 2584 # the translated stack should contain callees before callers. 2585 for (my $j = $#{$symlist}; $j >= 2; $j -= 3) { 2586 my $func = $symlist->[$j-2]; 2587 my $fileline = $symlist->[$j-1]; 2588 my $fullfunc = $symlist->[$j]; 2589 if (defined($fullname_to_shortname_map->{$fullfunc})) { 2590 $func = $fullname_to_shortname_map->{$fullfunc}; 2591 } 2592 if ($j > 2) { 2593 $func = "$func (inline)"; 2594 } 2595 2596 # Do not merge nodes corresponding to Callback::Run since that 2597 # causes confusing cycles in dot display. Instead, we synthesize 2598 # a unique name for this frame per caller. 2599 if ($func =~ m/Callback.*::Run$/) { 2600 my $caller = ($i > 0) ? $addrs[$i-1] : 0; 2601 $func = "Run#" . ShortIdFor($caller); 2602 } 2603 2604 if ($main::opt_addresses) { 2605 push(@result, "$a $func $fileline"); 2606 } elsif ($main::opt_lines) { 2607 if ($func eq '??' && $fileline eq '??:0') { 2608 push(@result, "$a"); 2609 } else { 2610 push(@result, "$func $fileline"); 2611 } 2612 } elsif ($main::opt_functions) { 2613 if ($func eq '??') { 2614 push(@result, "$a"); 2615 } else { 2616 push(@result, $func); 2617 } 2618 } elsif ($main::opt_files) { 2619 if ($fileline eq '??:0' || $fileline eq '') { 2620 push(@result, "$a"); 2621 } else { 2622 my $f = $fileline; 2623 $f =~ s/:\d+$//; 2624 push(@result, $f); 2625 } 2626 } else { 2627 push(@result, $a); 2628 last; # Do not print inlined info 2629 } 2630 } 2631 } 2632 2633 # print join(",", @addrs), " => ", join(",", @result), "\n"; 2634 return @result; 2635 } 2636 2637 # Generate percent string for a number and a total 2638 sub Percent { 2639 my $num = shift; 2640 my $tot = shift; 2641 if ($tot != 0) { 2642 return sprintf("%.1f%%", $num * 100.0 / $tot); 2643 } else { 2644 return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf"); 2645 } 2646 } 2647 2648 # Generate pretty-printed form of number 2649 sub Unparse { 2650 my $num = shift; 2651 if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') { 2652 if ($main::opt_inuse_objects || $main::opt_alloc_objects) { 2653 return sprintf("%d", $num); 2654 } else { 2655 if ($main::opt_show_bytes) { 2656 return sprintf("%d", $num); 2657 } else { 2658 return sprintf("%.1f", $num / 1048576.0); 2659 } 2660 } 2661 } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) { 2662 return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds 2663 } else { 2664 return sprintf("%d", $num); 2665 } 2666 } 2667 2668 # Alternate pretty-printed form: 0 maps to "." 2669 sub UnparseAlt { 2670 my $num = shift; 2671 if ($num == 0) { 2672 return "."; 2673 } else { 2674 return Unparse($num); 2675 } 2676 } 2677 2678 # Alternate pretty-printed form: 0 maps to "" 2679 sub HtmlPrintNumber { 2680 my $num = shift; 2681 if ($num == 0) { 2682 return ""; 2683 } else { 2684 return Unparse($num); 2685 } 2686 } 2687 2688 # Return output units 2689 sub Units { 2690 if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') { 2691 if ($main::opt_inuse_objects || $main::opt_alloc_objects) { 2692 return "objects"; 2693 } else { 2694 if ($main::opt_show_bytes) { 2695 return "B"; 2696 } else { 2697 return "MB"; 2698 } 2699 } 2700 } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) { 2701 return "seconds"; 2702 } else { 2703 return "samples"; 2704 } 2705 } 2706 2707 ##### Profile manipulation code ##### 2708 2709 # Generate flattened profile: 2710 # If count is charged to stack [a,b,c,d], in generated profile, 2711 # it will be charged to [a] 2712 sub FlatProfile { 2713 my $profile = shift; 2714 my $result = {}; 2715 foreach my $k (keys(%{$profile})) { 2716 my $count = $profile->{$k}; 2717 my @addrs = split(/\n/, $k); 2718 if ($#addrs >= 0) { 2719 AddEntry($result, $addrs[0], $count); 2720 } 2721 } 2722 return $result; 2723 } 2724 2725 # Generate cumulative profile: 2726 # If count is charged to stack [a,b,c,d], in generated profile, 2727 # it will be charged to [a], [b], [c], [d] 2728 sub CumulativeProfile { 2729 my $profile = shift; 2730 my $result = {}; 2731 foreach my $k (keys(%{$profile})) { 2732 my $count = $profile->{$k}; 2733 my @addrs = split(/\n/, $k); 2734 foreach my $a (@addrs) { 2735 AddEntry($result, $a, $count); 2736 } 2737 } 2738 return $result; 2739 } 2740 2741 # If the second-youngest PC on the stack is always the same, returns 2742 # that pc. Otherwise, returns undef. 2743 sub IsSecondPcAlwaysTheSame { 2744 my $profile = shift; 2745 2746 my $second_pc = undef; 2747 foreach my $k (keys(%{$profile})) { 2748 my @addrs = split(/\n/, $k); 2749 if ($#addrs < 1) { 2750 return undef; 2751 } 2752 if (not defined $second_pc) { 2753 $second_pc = $addrs[1]; 2754 } else { 2755 if ($second_pc ne $addrs[1]) { 2756 return undef; 2757 } 2758 } 2759 } 2760 return $second_pc; 2761 } 2762 2763 sub ExtractSymbolLocation { 2764 my $symbols = shift; 2765 my $address = shift; 2766 # 'addr2line' outputs "??:0" for unknown locations; we do the 2767 # same to be consistent. 2768 my $location = "??:0:unknown"; 2769 if (exists $symbols->{$address}) { 2770 my $file = $symbols->{$address}->[1]; 2771 if ($file eq "?") { 2772 $file = "??:0" 2773 } 2774 $location = $file . ":" . $symbols->{$address}->[0]; 2775 } 2776 return $location; 2777 } 2778 2779 # Extracts a graph of calls. 2780 sub ExtractCalls { 2781 my $symbols = shift; 2782 my $profile = shift; 2783 2784 my $calls = {}; 2785 while( my ($stack_trace, $count) = each %$profile ) { 2786 my @address = split(/\n/, $stack_trace); 2787 my $destination = ExtractSymbolLocation($symbols, $address[0]); 2788 AddEntry($calls, $destination, $count); 2789 for (my $i = 1; $i <= $#address; $i++) { 2790 my $source = ExtractSymbolLocation($symbols, $address[$i]); 2791 my $call = "$source -> $destination"; 2792 AddEntry($calls, $call, $count); 2793 $destination = $source; 2794 } 2795 } 2796 2797 return $calls; 2798 } 2799 2800 sub RemoveUninterestingFrames { 2801 my $symbols = shift; 2802 my $profile = shift; 2803 2804 # List of function names to skip 2805 my %skip = (); 2806 my $skip_regexp = 'NOMATCH'; 2807 if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') { 2808 foreach my $name ('calloc', 2809 'cfree', 2810 'malloc', 2811 'free', 2812 'memalign', 2813 'posix_memalign', 2814 'aligned_alloc', 2815 'pvalloc', 2816 'valloc', 2817 'realloc', 2818 'mallocx', # jemalloc 2819 'rallocx', # jemalloc 2820 'xallocx', # jemalloc 2821 'dallocx', # jemalloc 2822 'tc_calloc', 2823 'tc_cfree', 2824 'tc_malloc', 2825 'tc_free', 2826 'tc_memalign', 2827 'tc_posix_memalign', 2828 'tc_pvalloc', 2829 'tc_valloc', 2830 'tc_realloc', 2831 'tc_new', 2832 'tc_delete', 2833 'tc_newarray', 2834 'tc_deletearray', 2835 'tc_new_nothrow', 2836 'tc_newarray_nothrow', 2837 'do_malloc', 2838 '::do_malloc', # new name -- got moved to an unnamed ns 2839 '::do_malloc_or_cpp_alloc', 2840 'DoSampledAllocation', 2841 'simple_alloc::allocate', 2842 '__malloc_alloc_template::allocate', 2843 '__builtin_delete', 2844 '__builtin_new', 2845 '__builtin_vec_delete', 2846 '__builtin_vec_new', 2847 'operator new', 2848 'operator new[]', 2849 # The entry to our memory-allocation routines on OS X 2850 'malloc_zone_malloc', 2851 'malloc_zone_calloc', 2852 'malloc_zone_valloc', 2853 'malloc_zone_realloc', 2854 'malloc_zone_memalign', 2855 'malloc_zone_free', 2856 # These mark the beginning/end of our custom sections 2857 '__start_google_malloc', 2858 '__stop_google_malloc', 2859 '__start_malloc_hook', 2860 '__stop_malloc_hook') { 2861 $skip{$name} = 1; 2862 $skip{"_" . $name} = 1; # Mach (OS X) adds a _ prefix to everything 2863 } 2864 # TODO: Remove TCMalloc once everything has been 2865 # moved into the tcmalloc:: namespace and we have flushed 2866 # old code out of the system. 2867 $skip_regexp = "TCMalloc|^tcmalloc::"; 2868 } elsif ($main::profile_type eq 'contention') { 2869 foreach my $vname ('base::RecordLockProfileData', 2870 'base::SubmitMutexProfileData', 2871 'base::SubmitSpinLockProfileData', 2872 'Mutex::Unlock', 2873 'Mutex::UnlockSlow', 2874 'Mutex::ReaderUnlock', 2875 'MutexLock::~MutexLock', 2876 'SpinLock::Unlock', 2877 'SpinLock::SlowUnlock', 2878 'SpinLockHolder::~SpinLockHolder') { 2879 $skip{$vname} = 1; 2880 } 2881 } elsif ($main::profile_type eq 'cpu') { 2882 # Drop signal handlers used for CPU profile collection 2883 # TODO(dpeng): this should not be necessary; it's taken 2884 # care of by the general 2nd-pc mechanism below. 2885 foreach my $name ('ProfileData::Add', # historical 2886 'ProfileData::prof_handler', # historical 2887 'CpuProfiler::prof_handler', 2888 '__FRAME_END__', 2889 '__pthread_sighandler', 2890 '__restore') { 2891 $skip{$name} = 1; 2892 } 2893 } else { 2894 # Nothing skipped for unknown types 2895 } 2896 2897 if ($main::profile_type eq 'cpu') { 2898 # If all the second-youngest program counters are the same, 2899 # this STRONGLY suggests that it is an artifact of measurement, 2900 # i.e., stack frames pushed by the CPU profiler signal handler. 2901 # Hence, we delete them. 2902 # (The topmost PC is read from the signal structure, not from 2903 # the stack, so it does not get involved.) 2904 while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) { 2905 my $result = {}; 2906 my $func = ''; 2907 if (exists($symbols->{$second_pc})) { 2908 $second_pc = $symbols->{$second_pc}->[0]; 2909 } 2910 print STDERR "Removing $second_pc from all stack traces.\n"; 2911 foreach my $k (keys(%{$profile})) { 2912 my $count = $profile->{$k}; 2913 my @addrs = split(/\n/, $k); 2914 splice @addrs, 1, 1; 2915 my $reduced_path = join("\n", @addrs); 2916 AddEntry($result, $reduced_path, $count); 2917 } 2918 $profile = $result; 2919 } 2920 } 2921 2922 my $result = {}; 2923 foreach my $k (keys(%{$profile})) { 2924 my $count = $profile->{$k}; 2925 my @addrs = split(/\n/, $k); 2926 my @path = (); 2927 foreach my $a (@addrs) { 2928 if (exists($symbols->{$a})) { 2929 my $func = $symbols->{$a}->[0]; 2930 if ($skip{$func} || ($func =~ m/$skip_regexp/)) { 2931 # Throw away the portion of the backtrace seen so far, under the 2932 # assumption that previous frames were for functions internal to the 2933 # allocator. 2934 @path = (); 2935 next; 2936 } 2937 } 2938 push(@path, $a); 2939 } 2940 my $reduced_path = join("\n", @path); 2941 AddEntry($result, $reduced_path, $count); 2942 } 2943 return $result; 2944 } 2945 2946 # Reduce profile to granularity given by user 2947 sub ReduceProfile { 2948 my $symbols = shift; 2949 my $profile = shift; 2950 my $result = {}; 2951 my $fullname_to_shortname_map = {}; 2952 FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map); 2953 foreach my $k (keys(%{$profile})) { 2954 my $count = $profile->{$k}; 2955 my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k); 2956 my @path = (); 2957 my %seen = (); 2958 $seen{''} = 1; # So that empty keys are skipped 2959 foreach my $e (@translated) { 2960 # To avoid double-counting due to recursion, skip a stack-trace 2961 # entry if it has already been seen 2962 if (!$seen{$e}) { 2963 $seen{$e} = 1; 2964 push(@path, $e); 2965 } 2966 } 2967 my $reduced_path = join("\n", @path); 2968 AddEntry($result, $reduced_path, $count); 2969 } 2970 return $result; 2971 } 2972 2973 # Does the specified symbol array match the regexp? 2974 sub SymbolMatches { 2975 my $sym = shift; 2976 my $re = shift; 2977 if (defined($sym)) { 2978 for (my $i = 0; $i < $#{$sym}; $i += 3) { 2979 if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) { 2980 return 1; 2981 } 2982 } 2983 } 2984 return 0; 2985 } 2986 2987 # Focus only on paths involving specified regexps 2988 sub FocusProfile { 2989 my $symbols = shift; 2990 my $profile = shift; 2991 my $focus = shift; 2992 my $result = {}; 2993 foreach my $k (keys(%{$profile})) { 2994 my $count = $profile->{$k}; 2995 my @addrs = split(/\n/, $k); 2996 foreach my $a (@addrs) { 2997 # Reply if it matches either the address/shortname/fileline 2998 if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) { 2999 AddEntry($result, $k, $count); 3000 last; 3001 } 3002 } 3003 } 3004 return $result; 3005 } 3006 3007 # Focus only on paths not involving specified regexps 3008 sub IgnoreProfile { 3009 my $symbols = shift; 3010 my $profile = shift; 3011 my $ignore = shift; 3012 my $result = {}; 3013 foreach my $k (keys(%{$profile})) { 3014 my $count = $profile->{$k}; 3015 my @addrs = split(/\n/, $k); 3016 my $matched = 0; 3017 foreach my $a (@addrs) { 3018 # Reply if it matches either the address/shortname/fileline 3019 if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) { 3020 $matched = 1; 3021 last; 3022 } 3023 } 3024 if (!$matched) { 3025 AddEntry($result, $k, $count); 3026 } 3027 } 3028 return $result; 3029 } 3030 3031 # Get total count in profile 3032 sub TotalProfile { 3033 my $profile = shift; 3034 my $result = 0; 3035 foreach my $k (keys(%{$profile})) { 3036 $result += $profile->{$k}; 3037 } 3038 return $result; 3039 } 3040 3041 # Add A to B 3042 sub AddProfile { 3043 my $A = shift; 3044 my $B = shift; 3045 3046 my $R = {}; 3047 # add all keys in A 3048 foreach my $k (keys(%{$A})) { 3049 my $v = $A->{$k}; 3050 AddEntry($R, $k, $v); 3051 } 3052 # add all keys in B 3053 foreach my $k (keys(%{$B})) { 3054 my $v = $B->{$k}; 3055 AddEntry($R, $k, $v); 3056 } 3057 return $R; 3058 } 3059 3060 # Merges symbol maps 3061 sub MergeSymbols { 3062 my $A = shift; 3063 my $B = shift; 3064 3065 my $R = {}; 3066 foreach my $k (keys(%{$A})) { 3067 $R->{$k} = $A->{$k}; 3068 } 3069 if (defined($B)) { 3070 foreach my $k (keys(%{$B})) { 3071 $R->{$k} = $B->{$k}; 3072 } 3073 } 3074 return $R; 3075 } 3076 3077 3078 # Add A to B 3079 sub AddPcs { 3080 my $A = shift; 3081 my $B = shift; 3082 3083 my $R = {}; 3084 # add all keys in A 3085 foreach my $k (keys(%{$A})) { 3086 $R->{$k} = 1 3087 } 3088 # add all keys in B 3089 foreach my $k (keys(%{$B})) { 3090 $R->{$k} = 1 3091 } 3092 return $R; 3093 } 3094 3095 # Subtract B from A 3096 sub SubtractProfile { 3097 my $A = shift; 3098 my $B = shift; 3099 3100 my $R = {}; 3101 foreach my $k (keys(%{$A})) { 3102 my $v = $A->{$k} - GetEntry($B, $k); 3103 if ($v < 0 && $main::opt_drop_negative) { 3104 $v = 0; 3105 } 3106 AddEntry($R, $k, $v); 3107 } 3108 if (!$main::opt_drop_negative) { 3109 # Take care of when subtracted profile has more entries 3110 foreach my $k (keys(%{$B})) { 3111 if (!exists($A->{$k})) { 3112 AddEntry($R, $k, 0 - $B->{$k}); 3113 } 3114 } 3115 } 3116 return $R; 3117 } 3118 3119 # Get entry from profile; zero if not present 3120 sub GetEntry { 3121 my $profile = shift; 3122 my $k = shift; 3123 if (exists($profile->{$k})) { 3124 return $profile->{$k}; 3125 } else { 3126 return 0; 3127 } 3128 } 3129 3130 # Add entry to specified profile 3131 sub AddEntry { 3132 my $profile = shift; 3133 my $k = shift; 3134 my $n = shift; 3135 if (!exists($profile->{$k})) { 3136 $profile->{$k} = 0; 3137 } 3138 $profile->{$k} += $n; 3139 } 3140 3141 # Add a stack of entries to specified profile, and add them to the $pcs 3142 # list. 3143 sub AddEntries { 3144 my $profile = shift; 3145 my $pcs = shift; 3146 my $stack = shift; 3147 my $count = shift; 3148 my @k = (); 3149 3150 foreach my $e (split(/\s+/, $stack)) { 3151 my $pc = HexExtend($e); 3152 $pcs->{$pc} = 1; 3153 push @k, $pc; 3154 } 3155 AddEntry($profile, (join "\n", @k), $count); 3156 } 3157 3158 ##### Code to profile a server dynamically ##### 3159 3160 sub CheckSymbolPage { 3161 my $url = SymbolPageURL(); 3162 my $command = ShellEscape(@URL_FETCHER, $url); 3163 open(SYMBOL, "$command |") or error($command); 3164 my $line = <SYMBOL>; 3165 $line =~ s/\r//g; # turn windows-looking lines into unix-looking lines 3166 close(SYMBOL); 3167 unless (defined($line)) { 3168 error("$url doesn't exist\n"); 3169 } 3170 3171 if ($line =~ /^num_symbols:\s+(\d+)$/) { 3172 if ($1 == 0) { 3173 error("Stripped binary. No symbols available.\n"); 3174 } 3175 } else { 3176 error("Failed to get the number of symbols from $url\n"); 3177 } 3178 } 3179 3180 sub IsProfileURL { 3181 my $profile_name = shift; 3182 if (-f $profile_name) { 3183 printf STDERR "Using local file $profile_name.\n"; 3184 return 0; 3185 } 3186 return 1; 3187 } 3188 3189 sub ParseProfileURL { 3190 my $profile_name = shift; 3191 3192 if (!defined($profile_name) || $profile_name eq "") { 3193 return (); 3194 } 3195 3196 # Split profile URL - matches all non-empty strings, so no test. 3197 $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,; 3198 3199 my $proto = $1 || "http://"; 3200 my $hostport = $2; 3201 my $prefix = $3; 3202 my $profile = $4 || "/"; 3203 3204 my $host = $hostport; 3205 $host =~ s/:.*//; 3206 3207 my $baseurl = "$proto$hostport$prefix"; 3208 return ($host, $baseurl, $profile); 3209 } 3210 3211 # We fetch symbols from the first profile argument. 3212 sub SymbolPageURL { 3213 my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]); 3214 return "$baseURL$SYMBOL_PAGE"; 3215 } 3216 3217 sub FetchProgramName() { 3218 my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]); 3219 my $url = "$baseURL$PROGRAM_NAME_PAGE"; 3220 my $command_line = ShellEscape(@URL_FETCHER, $url); 3221 open(CMDLINE, "$command_line |") or error($command_line); 3222 my $cmdline = <CMDLINE>; 3223 $cmdline =~ s/\r//g; # turn windows-looking lines into unix-looking lines 3224 close(CMDLINE); 3225 error("Failed to get program name from $url\n") unless defined($cmdline); 3226 $cmdline =~ s/\x00.+//; # Remove argv[1] and latters. 3227 $cmdline =~ s!\n!!g; # Remove LFs. 3228 return $cmdline; 3229 } 3230 3231 # Gee, curl's -L (--location) option isn't reliable at least 3232 # with its 7.12.3 version. Curl will forget to post data if 3233 # there is a redirection. This function is a workaround for 3234 # curl. Redirection happens on borg hosts. 3235 sub ResolveRedirectionForCurl { 3236 my $url = shift; 3237 my $command_line = ShellEscape(@URL_FETCHER, "--head", $url); 3238 open(CMDLINE, "$command_line |") or error($command_line); 3239 while (<CMDLINE>) { 3240 s/\r//g; # turn windows-looking lines into unix-looking lines 3241 if (/^Location: (.*)/) { 3242 $url = $1; 3243 } 3244 } 3245 close(CMDLINE); 3246 return $url; 3247 } 3248 3249 # Add a timeout flat to URL_FETCHER. Returns a new list. 3250 sub AddFetchTimeout { 3251 my $timeout = shift; 3252 my @fetcher = shift; 3253 if (defined($timeout)) { 3254 if (join(" ", @fetcher) =~ m/\bcurl -s/) { 3255 push(@fetcher, "--max-time", sprintf("%d", $timeout)); 3256 } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) { 3257 push(@fetcher, sprintf("--deadline=%d", $timeout)); 3258 } 3259 } 3260 return @fetcher; 3261 } 3262 3263 # Reads a symbol map from the file handle name given as $1, returning 3264 # the resulting symbol map. Also processes variables relating to symbols. 3265 # Currently, the only variable processed is 'binary=<value>' which updates 3266 # $main::prog to have the correct program name. 3267 sub ReadSymbols { 3268 my $in = shift; 3269 my $map = {}; 3270 while (<$in>) { 3271 s/\r//g; # turn windows-looking lines into unix-looking lines 3272 # Removes all the leading zeroes from the symbols, see comment below. 3273 if (m/^0x0*([0-9a-f]+)\s+(.+)/) { 3274 $map->{$1} = $2; 3275 } elsif (m/^---/) { 3276 last; 3277 } elsif (m/^([a-z][^=]*)=(.*)$/ ) { 3278 my ($variable, $value) = ($1, $2); 3279 for ($variable, $value) { 3280 s/^\s+//; 3281 s/\s+$//; 3282 } 3283 if ($variable eq "binary") { 3284 if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) { 3285 printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n", 3286 $main::prog, $value); 3287 } 3288 $main::prog = $value; 3289 } else { 3290 printf STDERR ("Ignoring unknown variable in symbols list: " . 3291 "'%s' = '%s'\n", $variable, $value); 3292 } 3293 } 3294 } 3295 return $map; 3296 } 3297 3298 # Fetches and processes symbols to prepare them for use in the profile output 3299 # code. If the optional 'symbol_map' arg is not given, fetches symbols from 3300 # $SYMBOL_PAGE for all PC values found in profile. Otherwise, the raw symbols 3301 # are assumed to have already been fetched into 'symbol_map' and are simply 3302 # extracted and processed. 3303 sub FetchSymbols { 3304 my $pcset = shift; 3305 my $symbol_map = shift; 3306 3307 my %seen = (); 3308 my @pcs = grep { !$seen{$_}++ } keys(%$pcset); # uniq 3309 3310 if (!defined($symbol_map)) { 3311 my $post_data = join("+", sort((map {"0x" . "$_"} @pcs))); 3312 3313 open(POSTFILE, ">$main::tmpfile_sym"); 3314 print POSTFILE $post_data; 3315 close(POSTFILE); 3316 3317 my $url = SymbolPageURL(); 3318 3319 my $command_line; 3320 if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) { 3321 $url = ResolveRedirectionForCurl($url); 3322 $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym", 3323 $url); 3324 } else { 3325 $command_line = (ShellEscape(@URL_FETCHER, "--post", $url) 3326 . " < " . ShellEscape($main::tmpfile_sym)); 3327 } 3328 # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols. 3329 my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"}); 3330 open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line); 3331 $symbol_map = ReadSymbols(*SYMBOL{IO}); 3332 close(SYMBOL); 3333 } 3334 3335 my $symbols = {}; 3336 foreach my $pc (@pcs) { 3337 my $fullname; 3338 # For 64 bits binaries, symbols are extracted with 8 leading zeroes. 3339 # Then /symbol reads the long symbols in as uint64, and outputs 3340 # the result with a "0x%08llx" format which get rid of the zeroes. 3341 # By removing all the leading zeroes in both $pc and the symbols from 3342 # /symbol, the symbols match and are retrievable from the map. 3343 my $shortpc = $pc; 3344 $shortpc =~ s/^0*//; 3345 # Each line may have a list of names, which includes the function 3346 # and also other functions it has inlined. They are separated (in 3347 # PrintSymbolizedProfile), by --, which is illegal in function names. 3348 my $fullnames; 3349 if (defined($symbol_map->{$shortpc})) { 3350 $fullnames = $symbol_map->{$shortpc}; 3351 } else { 3352 $fullnames = "0x" . $pc; # Just use addresses 3353 } 3354 my $sym = []; 3355 $symbols->{$pc} = $sym; 3356 foreach my $fullname (split("--", $fullnames)) { 3357 my $name = ShortFunctionName($fullname); 3358 push(@{$sym}, $name, "?", $fullname); 3359 } 3360 } 3361 return $symbols; 3362 } 3363 3364 sub BaseName { 3365 my $file_name = shift; 3366 $file_name =~ s!^.*/!!; # Remove directory name 3367 return $file_name; 3368 } 3369 3370 sub MakeProfileBaseName { 3371 my ($binary_name, $profile_name) = @_; 3372 my ($host, $baseURL, $path) = ParseProfileURL($profile_name); 3373 my $binary_shortname = BaseName($binary_name); 3374 return sprintf("%s.%s.%s", 3375 $binary_shortname, $main::op_time, $host); 3376 } 3377 3378 sub FetchDynamicProfile { 3379 my $binary_name = shift; 3380 my $profile_name = shift; 3381 my $fetch_name_only = shift; 3382 my $encourage_patience = shift; 3383 3384 if (!IsProfileURL($profile_name)) { 3385 return $profile_name; 3386 } else { 3387 my ($host, $baseURL, $path) = ParseProfileURL($profile_name); 3388 if ($path eq "" || $path eq "/") { 3389 # Missing type specifier defaults to cpu-profile 3390 $path = $PROFILE_PAGE; 3391 } 3392 3393 my $profile_file = MakeProfileBaseName($binary_name, $profile_name); 3394 3395 my $url = "$baseURL$path"; 3396 my $fetch_timeout = undef; 3397 if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) { 3398 if ($path =~ m/[?]/) { 3399 $url .= "&"; 3400 } else { 3401 $url .= "?"; 3402 } 3403 $url .= sprintf("seconds=%d", $main::opt_seconds); 3404 $fetch_timeout = $main::opt_seconds * 1.01 + 60; 3405 } else { 3406 # For non-CPU profiles, we add a type-extension to 3407 # the target profile file name. 3408 my $suffix = $path; 3409 $suffix =~ s,/,.,g; 3410 $profile_file .= $suffix; 3411 } 3412 3413 my $profile_dir = $ENV{"PPROF_TMPDIR"} || ($ENV{HOME} . "/pprof"); 3414 if (! -d $profile_dir) { 3415 mkdir($profile_dir) 3416 || die("Unable to create profile directory $profile_dir: $!\n"); 3417 } 3418 my $tmp_profile = "$profile_dir/.tmp.$profile_file"; 3419 my $real_profile = "$profile_dir/$profile_file"; 3420 3421 if ($fetch_name_only > 0) { 3422 return $real_profile; 3423 } 3424 3425 my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER); 3426 my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile); 3427 if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){ 3428 print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n ${real_profile}\n"; 3429 if ($encourage_patience) { 3430 print STDERR "Be patient...\n"; 3431 } 3432 } else { 3433 print STDERR "Fetching $path profile from $url to\n ${real_profile}\n"; 3434 } 3435 3436 (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n"); 3437 (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n"); 3438 print STDERR "Wrote profile to $real_profile\n"; 3439 $main::collected_profile = $real_profile; 3440 return $main::collected_profile; 3441 } 3442 } 3443 3444 # Collect profiles in parallel 3445 sub FetchDynamicProfiles { 3446 my $items = scalar(@main::pfile_args); 3447 my $levels = log($items) / log(2); 3448 3449 if ($items == 1) { 3450 $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1); 3451 } else { 3452 # math rounding issues 3453 if ((2 ** $levels) < $items) { 3454 $levels++; 3455 } 3456 my $count = scalar(@main::pfile_args); 3457 for (my $i = 0; $i < $count; $i++) { 3458 $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0); 3459 } 3460 print STDERR "Fetching $count profiles, Be patient...\n"; 3461 FetchDynamicProfilesRecurse($levels, 0, 0); 3462 $main::collected_profile = join(" \\\n ", @main::profile_files); 3463 } 3464 } 3465 3466 # Recursively fork a process to get enough processes 3467 # collecting profiles 3468 sub FetchDynamicProfilesRecurse { 3469 my $maxlevel = shift; 3470 my $level = shift; 3471 my $position = shift; 3472 3473 if (my $pid = fork()) { 3474 $position = 0 | ($position << 1); 3475 TryCollectProfile($maxlevel, $level, $position); 3476 wait; 3477 } else { 3478 $position = 1 | ($position << 1); 3479 TryCollectProfile($maxlevel, $level, $position); 3480 cleanup(); 3481 exit(0); 3482 } 3483 } 3484 3485 # Collect a single profile 3486 sub TryCollectProfile { 3487 my $maxlevel = shift; 3488 my $level = shift; 3489 my $position = shift; 3490 3491 if ($level >= ($maxlevel - 1)) { 3492 if ($position < scalar(@main::pfile_args)) { 3493 FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0); 3494 } 3495 } else { 3496 FetchDynamicProfilesRecurse($maxlevel, $level+1, $position); 3497 } 3498 } 3499 3500 ##### Parsing code ##### 3501 3502 # Provide a small streaming-read module to handle very large 3503 # cpu-profile files. Stream in chunks along a sliding window. 3504 # Provides an interface to get one 'slot', correctly handling 3505 # endian-ness differences. A slot is one 32-bit or 64-bit word 3506 # (depending on the input profile). We tell endianness and bit-size 3507 # for the profile by looking at the first 8 bytes: in cpu profiles, 3508 # the second slot is always 3 (we'll accept anything that's not 0). 3509 BEGIN { 3510 package CpuProfileStream; 3511 3512 sub new { 3513 my ($class, $file, $fname) = @_; 3514 my $self = { file => $file, 3515 base => 0, 3516 stride => 512 * 1024, # must be a multiple of bitsize/8 3517 slots => [], 3518 unpack_code => "", # N for big-endian, V for little 3519 perl_is_64bit => 1, # matters if profile is 64-bit 3520 }; 3521 bless $self, $class; 3522 # Let unittests adjust the stride 3523 if ($main::opt_test_stride > 0) { 3524 $self->{stride} = $main::opt_test_stride; 3525 } 3526 # Read the first two slots to figure out bitsize and endianness. 3527 my $slots = $self->{slots}; 3528 my $str; 3529 read($self->{file}, $str, 8); 3530 # Set the global $address_length based on what we see here. 3531 # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars). 3532 $address_length = ($str eq (chr(0)x8)) ? 16 : 8; 3533 if ($address_length == 8) { 3534 if (substr($str, 6, 2) eq chr(0)x2) { 3535 $self->{unpack_code} = 'V'; # Little-endian. 3536 } elsif (substr($str, 4, 2) eq chr(0)x2) { 3537 $self->{unpack_code} = 'N'; # Big-endian 3538 } else { 3539 ::error("$fname: header size >= 2**16\n"); 3540 } 3541 @$slots = unpack($self->{unpack_code} . "*", $str); 3542 } else { 3543 # If we're a 64-bit profile, check if we're a 64-bit-capable 3544 # perl. Otherwise, each slot will be represented as a float 3545 # instead of an int64, losing precision and making all the 3546 # 64-bit addresses wrong. We won't complain yet, but will 3547 # later if we ever see a value that doesn't fit in 32 bits. 3548 my $has_q = 0; 3549 eval { $has_q = pack("Q", "1") ? 1 : 1; }; 3550 if (!$has_q) { 3551 $self->{perl_is_64bit} = 0; 3552 } 3553 read($self->{file}, $str, 8); 3554 if (substr($str, 4, 4) eq chr(0)x4) { 3555 # We'd love to use 'Q', but it's a) not universal, b) not endian-proof. 3556 $self->{unpack_code} = 'V'; # Little-endian. 3557 } elsif (substr($str, 0, 4) eq chr(0)x4) { 3558 $self->{unpack_code} = 'N'; # Big-endian 3559 } else { 3560 ::error("$fname: header size >= 2**32\n"); 3561 } 3562 my @pair = unpack($self->{unpack_code} . "*", $str); 3563 # Since we know one of the pair is 0, it's fine to just add them. 3564 @$slots = (0, $pair[0] + $pair[1]); 3565 } 3566 return $self; 3567 } 3568 3569 # Load more data when we access slots->get(X) which is not yet in memory. 3570 sub overflow { 3571 my ($self) = @_; 3572 my $slots = $self->{slots}; 3573 $self->{base} += $#$slots + 1; # skip over data we're replacing 3574 my $str; 3575 read($self->{file}, $str, $self->{stride}); 3576 if ($address_length == 8) { # the 32-bit case 3577 # This is the easy case: unpack provides 32-bit unpacking primitives. 3578 @$slots = unpack($self->{unpack_code} . "*", $str); 3579 } else { 3580 # We need to unpack 32 bits at a time and combine. 3581 my @b32_values = unpack($self->{unpack_code} . "*", $str); 3582 my @b64_values = (); 3583 for (my $i = 0; $i < $#b32_values; $i += 2) { 3584 # TODO(csilvers): if this is a 32-bit perl, the math below 3585 # could end up in a too-large int, which perl will promote 3586 # to a double, losing necessary precision. Deal with that. 3587 # Right now, we just die. 3588 my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]); 3589 if ($self->{unpack_code} eq 'N') { # big-endian 3590 ($lo, $hi) = ($hi, $lo); 3591 } 3592 my $value = $lo + $hi * (2**32); 3593 if (!$self->{perl_is_64bit} && # check value is exactly represented 3594 (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) { 3595 ::error("Need a 64-bit perl to process this 64-bit profile.\n"); 3596 } 3597 push(@b64_values, $value); 3598 } 3599 @$slots = @b64_values; 3600 } 3601 } 3602 3603 # Access the i-th long in the file (logically), or -1 at EOF. 3604 sub get { 3605 my ($self, $idx) = @_; 3606 my $slots = $self->{slots}; 3607 while ($#$slots >= 0) { 3608 if ($idx < $self->{base}) { 3609 # The only time we expect a reference to $slots[$i - something] 3610 # after referencing $slots[$i] is reading the very first header. 3611 # Since $stride > |header|, that shouldn't cause any lookback 3612 # errors. And everything after the header is sequential. 3613 print STDERR "Unexpected look-back reading CPU profile"; 3614 return -1; # shrug, don't know what better to return 3615 } elsif ($idx > $self->{base} + $#$slots) { 3616 $self->overflow(); 3617 } else { 3618 return $slots->[$idx - $self->{base}]; 3619 } 3620 } 3621 # If we get here, $slots is [], which means we've reached EOF 3622 return -1; # unique since slots is supposed to hold unsigned numbers 3623 } 3624 } 3625 3626 # Reads the top, 'header' section of a profile, and returns the last 3627 # line of the header, commonly called a 'header line'. The header 3628 # section of a profile consists of zero or more 'command' lines that 3629 # are instructions to pprof, which pprof executes when reading the 3630 # header. All 'command' lines start with a %. After the command 3631 # lines is the 'header line', which is a profile-specific line that 3632 # indicates what type of profile it is, and perhaps other global 3633 # information about the profile. For instance, here's a header line 3634 # for a heap profile: 3635 # heap profile: 53: 38236 [ 5525: 1284029] @ heapprofile 3636 # For historical reasons, the CPU profile does not contain a text- 3637 # readable header line. If the profile looks like a CPU profile, 3638 # this function returns "". If no header line could be found, this 3639 # function returns undef. 3640 # 3641 # The following commands are recognized: 3642 # %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:' 3643 # 3644 # The input file should be in binmode. 3645 sub ReadProfileHeader { 3646 local *PROFILE = shift; 3647 my $firstchar = ""; 3648 my $line = ""; 3649 read(PROFILE, $firstchar, 1); 3650 seek(PROFILE, -1, 1); # unread the firstchar 3651 if ($firstchar !~ /[[:print:]]/) { # is not a text character 3652 return ""; 3653 } 3654 while (defined($line = <PROFILE>)) { 3655 $line =~ s/\r//g; # turn windows-looking lines into unix-looking lines 3656 if ($line =~ /^%warn\s+(.*)/) { # 'warn' command 3657 # Note this matches both '%warn blah\n' and '%warn\n'. 3658 print STDERR "WARNING: $1\n"; # print the rest of the line 3659 } elsif ($line =~ /^%/) { 3660 print STDERR "Ignoring unknown command from profile header: $line"; 3661 } else { 3662 # End of commands, must be the header line. 3663 return $line; 3664 } 3665 } 3666 return undef; # got to EOF without seeing a header line 3667 } 3668 3669 sub IsSymbolizedProfileFile { 3670 my $file_name = shift; 3671 if (!(-e $file_name) || !(-r $file_name)) { 3672 return 0; 3673 } 3674 # Check if the file contains a symbol-section marker. 3675 open(TFILE, "<$file_name"); 3676 binmode TFILE; 3677 my $firstline = ReadProfileHeader(*TFILE); 3678 close(TFILE); 3679 if (!$firstline) { 3680 return 0; 3681 } 3682 $SYMBOL_PAGE =~ m,[^/]+$,; # matches everything after the last slash 3683 my $symbol_marker = $&; 3684 return $firstline =~ /^--- *$symbol_marker/; 3685 } 3686 3687 # Parse profile generated by common/profiler.cc and return a reference 3688 # to a map: 3689 # $result->{version} Version number of profile file 3690 # $result->{period} Sampling period (in microseconds) 3691 # $result->{profile} Profile object 3692 # $result->{map} Memory map info from profile 3693 # $result->{pcs} Hash of all PC values seen, key is hex address 3694 sub ReadProfile { 3695 my $prog = shift; 3696 my $fname = shift; 3697 my $result; # return value 3698 3699 $CONTENTION_PAGE =~ m,[^/]+$,; # matches everything after the last slash 3700 my $contention_marker = $&; 3701 $GROWTH_PAGE =~ m,[^/]+$,; # matches everything after the last slash 3702 my $growth_marker = $&; 3703 $SYMBOL_PAGE =~ m,[^/]+$,; # matches everything after the last slash 3704 my $symbol_marker = $&; 3705 $PROFILE_PAGE =~ m,[^/]+$,; # matches everything after the last slash 3706 my $profile_marker = $&; 3707 3708 # Look at first line to see if it is a heap or a CPU profile. 3709 # CPU profile may start with no header at all, and just binary data 3710 # (starting with \0\0\0\0) -- in that case, don't try to read the 3711 # whole firstline, since it may be gigabytes(!) of data. 3712 open(PROFILE, "<$fname") || error("$fname: $!\n"); 3713 binmode PROFILE; # New perls do UTF-8 processing 3714 my $header = ReadProfileHeader(*PROFILE); 3715 if (!defined($header)) { # means "at EOF" 3716 error("Profile is empty.\n"); 3717 } 3718 3719 my $symbols; 3720 if ($header =~ m/^--- *$symbol_marker/o) { 3721 # Verify that the user asked for a symbolized profile 3722 if (!$main::use_symbolized_profile) { 3723 # we have both a binary and symbolized profiles, abort 3724 error("FATAL ERROR: Symbolized profile\n $fname\ncannot be used with " . 3725 "a binary arg. Try again without passing\n $prog\n"); 3726 } 3727 # Read the symbol section of the symbolized profile file. 3728 $symbols = ReadSymbols(*PROFILE{IO}); 3729 # Read the next line to get the header for the remaining profile. 3730 $header = ReadProfileHeader(*PROFILE) || ""; 3731 } 3732 3733 $main::profile_type = ''; 3734 if ($header =~ m/^heap profile:.*$growth_marker/o) { 3735 $main::profile_type = 'growth'; 3736 $result = ReadHeapProfile($prog, *PROFILE, $header); 3737 } elsif ($header =~ m/^heap profile:/) { 3738 $main::profile_type = 'heap'; 3739 $result = ReadHeapProfile($prog, *PROFILE, $header); 3740 } elsif ($header =~ m/^--- *$contention_marker/o) { 3741 $main::profile_type = 'contention'; 3742 $result = ReadSynchProfile($prog, *PROFILE); 3743 } elsif ($header =~ m/^--- *Stacks:/) { 3744 print STDERR 3745 "Old format contention profile: mistakenly reports " . 3746 "condition variable signals as lock contentions.\n"; 3747 $main::profile_type = 'contention'; 3748 $result = ReadSynchProfile($prog, *PROFILE); 3749 } elsif ($header =~ m/^--- *$profile_marker/) { 3750 # the binary cpu profile data starts immediately after this line 3751 $main::profile_type = 'cpu'; 3752 $result = ReadCPUProfile($prog, $fname, *PROFILE); 3753 } else { 3754 if (defined($symbols)) { 3755 # a symbolized profile contains a format we don't recognize, bail out 3756 error("$fname: Cannot recognize profile section after symbols.\n"); 3757 } 3758 # no ascii header present -- must be a CPU profile 3759 $main::profile_type = 'cpu'; 3760 $result = ReadCPUProfile($prog, $fname, *PROFILE); 3761 } 3762 3763 close(PROFILE); 3764 3765 # if we got symbols along with the profile, return those as well 3766 if (defined($symbols)) { 3767 $result->{symbols} = $symbols; 3768 } 3769 3770 return $result; 3771 } 3772 3773 # Subtract one from caller pc so we map back to call instr. 3774 # However, don't do this if we're reading a symbolized profile 3775 # file, in which case the subtract-one was done when the file 3776 # was written. 3777 # 3778 # We apply the same logic to all readers, though ReadCPUProfile uses an 3779 # independent implementation. 3780 sub FixCallerAddresses { 3781 my $stack = shift; 3782 if ($main::use_symbolized_profile) { 3783 return $stack; 3784 } else { 3785 $stack =~ /(\s)/; 3786 my $delimiter = $1; 3787 my @addrs = split(' ', $stack); 3788 my @fixedaddrs; 3789 $#fixedaddrs = $#addrs; 3790 if ($#addrs >= 0) { 3791 $fixedaddrs[0] = $addrs[0]; 3792 } 3793 for (my $i = 1; $i <= $#addrs; $i++) { 3794 $fixedaddrs[$i] = AddressSub($addrs[$i], "0x1"); 3795 } 3796 return join $delimiter, @fixedaddrs; 3797 } 3798 } 3799 3800 # CPU profile reader 3801 sub ReadCPUProfile { 3802 my $prog = shift; 3803 my $fname = shift; # just used for logging 3804 local *PROFILE = shift; 3805 my $version; 3806 my $period; 3807 my $i; 3808 my $profile = {}; 3809 my $pcs = {}; 3810 3811 # Parse string into array of slots. 3812 my $slots = CpuProfileStream->new(*PROFILE, $fname); 3813 3814 # Read header. The current header version is a 5-element structure 3815 # containing: 3816 # 0: header count (always 0) 3817 # 1: header "words" (after this one: 3) 3818 # 2: format version (0) 3819 # 3: sampling period (usec) 3820 # 4: unused padding (always 0) 3821 if ($slots->get(0) != 0 ) { 3822 error("$fname: not a profile file, or old format profile file\n"); 3823 } 3824 $i = 2 + $slots->get(1); 3825 $version = $slots->get(2); 3826 $period = $slots->get(3); 3827 # Do some sanity checking on these header values. 3828 if ($version > (2**32) || $period > (2**32) || $i > (2**32) || $i < 5) { 3829 error("$fname: not a profile file, or corrupted profile file\n"); 3830 } 3831 3832 # Parse profile 3833 while ($slots->get($i) != -1) { 3834 my $n = $slots->get($i++); 3835 my $d = $slots->get($i++); 3836 if ($d > (2**16)) { # TODO(csilvers): what's a reasonable max-stack-depth? 3837 my $addr = sprintf("0%o", $i * ($address_length == 8 ? 4 : 8)); 3838 print STDERR "At index $i (address $addr):\n"; 3839 error("$fname: stack trace depth >= 2**32\n"); 3840 } 3841 if ($slots->get($i) == 0) { 3842 # End of profile data marker 3843 $i += $d; 3844 last; 3845 } 3846 3847 # Make key out of the stack entries 3848 my @k = (); 3849 for (my $j = 0; $j < $d; $j++) { 3850 my $pc = $slots->get($i+$j); 3851 # Subtract one from caller pc so we map back to call instr. 3852 # However, don't do this if we're reading a symbolized profile 3853 # file, in which case the subtract-one was done when the file 3854 # was written. 3855 if ($j > 0 && !$main::use_symbolized_profile) { 3856 $pc--; 3857 } 3858 $pc = sprintf("%0*x", $address_length, $pc); 3859 $pcs->{$pc} = 1; 3860 push @k, $pc; 3861 } 3862 3863 AddEntry($profile, (join "\n", @k), $n); 3864 $i += $d; 3865 } 3866 3867 # Parse map 3868 my $map = ''; 3869 seek(PROFILE, $i * 4, 0); 3870 read(PROFILE, $map, (stat PROFILE)[7]); 3871 3872 my $r = {}; 3873 $r->{version} = $version; 3874 $r->{period} = $period; 3875 $r->{profile} = $profile; 3876 $r->{libs} = ParseLibraries($prog, $map, $pcs); 3877 $r->{pcs} = $pcs; 3878 3879 return $r; 3880 } 3881 3882 sub ReadHeapProfile { 3883 my $prog = shift; 3884 local *PROFILE = shift; 3885 my $header = shift; 3886 3887 my $index = 1; 3888 if ($main::opt_inuse_space) { 3889 $index = 1; 3890 } elsif ($main::opt_inuse_objects) { 3891 $index = 0; 3892 } elsif ($main::opt_alloc_space) { 3893 $index = 3; 3894 } elsif ($main::opt_alloc_objects) { 3895 $index = 2; 3896 } 3897 3898 # Find the type of this profile. The header line looks like: 3899 # heap profile: 1246: 8800744 [ 1246: 8800744] @ <heap-url>/266053 3900 # There are two pairs <count: size>, the first inuse objects/space, and the 3901 # second allocated objects/space. This is followed optionally by a profile 3902 # type, and if that is present, optionally by a sampling frequency. 3903 # For remote heap profiles (v1): 3904 # The interpretation of the sampling frequency is that the profiler, for 3905 # each sample, calculates a uniformly distributed random integer less than 3906 # the given value, and records the next sample after that many bytes have 3907 # been allocated. Therefore, the expected sample interval is half of the 3908 # given frequency. By default, if not specified, the expected sample 3909 # interval is 128KB. Only remote-heap-page profiles are adjusted for 3910 # sample size. 3911 # For remote heap profiles (v2): 3912 # The sampling frequency is the rate of a Poisson process. This means that 3913 # the probability of sampling an allocation of size X with sampling rate Y 3914 # is 1 - exp(-X/Y) 3915 # For version 2, a typical header line might look like this: 3916 # heap profile: 1922: 127792360 [ 1922: 127792360] @ <heap-url>_v2/524288 3917 # the trailing number (524288) is the sampling rate. (Version 1 showed 3918 # double the 'rate' here) 3919 my $sampling_algorithm = 0; 3920 my $sample_adjustment = 0; 3921 chomp($header); 3922 my $type = "unknown"; 3923 if ($header =~ m"^heap profile:\s*(\d+):\s+(\d+)\s+\[\s*(\d+):\s+(\d+)\](\s*@\s*([^/]*)(/(\d+))?)?") { 3924 if (defined($6) && ($6 ne '')) { 3925 $type = $6; 3926 my $sample_period = $8; 3927 # $type is "heapprofile" for profiles generated by the 3928 # heap-profiler, and either "heap" or "heap_v2" for profiles 3929 # generated by sampling directly within tcmalloc. It can also 3930 # be "growth" for heap-growth profiles. The first is typically 3931 # found for profiles generated locally, and the others for 3932 # remote profiles. 3933 if (($type eq "heapprofile") || ($type !~ /heap/) ) { 3934 # No need to adjust for the sampling rate with heap-profiler-derived data 3935 $sampling_algorithm = 0; 3936 } elsif ($type =~ /_v2/) { 3937 $sampling_algorithm = 2; # version 2 sampling 3938 if (defined($sample_period) && ($sample_period ne '')) { 3939 $sample_adjustment = int($sample_period); 3940 } 3941 } else { 3942 $sampling_algorithm = 1; # version 1 sampling 3943 if (defined($sample_period) && ($sample_period ne '')) { 3944 $sample_adjustment = int($sample_period)/2; 3945 } 3946 } 3947 } else { 3948 # We detect whether or not this is a remote-heap profile by checking 3949 # that the total-allocated stats ($n2,$s2) are exactly the 3950 # same as the in-use stats ($n1,$s1). It is remotely conceivable 3951 # that a non-remote-heap profile may pass this check, but it is hard 3952 # to imagine how that could happen. 3953 # In this case it's so old it's guaranteed to be remote-heap version 1. 3954 my ($n1, $s1, $n2, $s2) = ($1, $2, $3, $4); 3955 if (($n1 == $n2) && ($s1 == $s2)) { 3956 # This is likely to be a remote-heap based sample profile 3957 $sampling_algorithm = 1; 3958 } 3959 } 3960 } 3961 3962 if ($sampling_algorithm > 0) { 3963 # For remote-heap generated profiles, adjust the counts and sizes to 3964 # account for the sample rate (we sample once every 128KB by default). 3965 if ($sample_adjustment == 0) { 3966 # Turn on profile adjustment. 3967 $sample_adjustment = 128*1024; 3968 print STDERR "Adjusting heap profiles for 1-in-128KB sampling rate\n"; 3969 } else { 3970 printf STDERR ("Adjusting heap profiles for 1-in-%d sampling rate\n", 3971 $sample_adjustment); 3972 } 3973 if ($sampling_algorithm > 1) { 3974 # We don't bother printing anything for the original version (version 1) 3975 printf STDERR "Heap version $sampling_algorithm\n"; 3976 } 3977 } 3978 3979 my $profile = {}; 3980 my $pcs = {}; 3981 my $map = ""; 3982 3983 while (<PROFILE>) { 3984 s/\r//g; # turn windows-looking lines into unix-looking lines 3985 if (/^MAPPED_LIBRARIES:/) { 3986 # Read the /proc/self/maps data 3987 while (<PROFILE>) { 3988 s/\r//g; # turn windows-looking lines into unix-looking lines 3989 $map .= $_; 3990 } 3991 last; 3992 } 3993 3994 if (/^--- Memory map:/) { 3995 # Read /proc/self/maps data as formatted by DumpAddressMap() 3996 my $buildvar = ""; 3997 while (<PROFILE>) { 3998 s/\r//g; # turn windows-looking lines into unix-looking lines 3999 # Parse "build=<dir>" specification if supplied 4000 if (m/^\s*build=(.*)\n/) { 4001 $buildvar = $1; 4002 } 4003 4004 # Expand "$build" variable if available 4005 $_ =~ s/\$build\b/$buildvar/g; 4006 4007 $map .= $_; 4008 } 4009 last; 4010 } 4011 4012 # Read entry of the form: 4013 # <count1>: <bytes1> [<count2>: <bytes2>] @ a1 a2 a3 ... an 4014 s/^\s*//; 4015 s/\s*$//; 4016 if (m/^\s*(\d+):\s+(\d+)\s+\[\s*(\d+):\s+(\d+)\]\s+@\s+(.*)$/) { 4017 my $stack = $5; 4018 my ($n1, $s1, $n2, $s2) = ($1, $2, $3, $4); 4019 4020 if ($sample_adjustment) { 4021 if ($sampling_algorithm == 2) { 4022 # Remote-heap version 2 4023 # The sampling frequency is the rate of a Poisson process. 4024 # This means that the probability of sampling an allocation of 4025 # size X with sampling rate Y is 1 - exp(-X/Y) 4026 if ($n1 != 0) { 4027 my $ratio = (($s1*1.0)/$n1)/($sample_adjustment); 4028 my $scale_factor = 1/(1 - exp(-$ratio)); 4029 $n1 *= $scale_factor; 4030 $s1 *= $scale_factor; 4031 } 4032 if ($n2 != 0) { 4033 my $ratio = (($s2*1.0)/$n2)/($sample_adjustment); 4034 my $scale_factor = 1/(1 - exp(-$ratio)); 4035 $n2 *= $scale_factor; 4036 $s2 *= $scale_factor; 4037 } 4038 } else { 4039 # Remote-heap version 1 4040 my $ratio; 4041 $ratio = (($s1*1.0)/$n1)/($sample_adjustment); 4042 if ($ratio < 1) { 4043 $n1 /= $ratio; 4044 $s1 /= $ratio; 4045 } 4046 $ratio = (($s2*1.0)/$n2)/($sample_adjustment); 4047 if ($ratio < 1) { 4048 $n2 /= $ratio; 4049 $s2 /= $ratio; 4050 } 4051 } 4052 } 4053 4054 my @counts = ($n1, $s1, $n2, $s2); 4055 AddEntries($profile, $pcs, FixCallerAddresses($stack), $counts[$index]); 4056 } 4057 } 4058 4059 my $r = {}; 4060 $r->{version} = "heap"; 4061 $r->{period} = 1; 4062 $r->{profile} = $profile; 4063 $r->{libs} = ParseLibraries($prog, $map, $pcs); 4064 $r->{pcs} = $pcs; 4065 return $r; 4066 } 4067 4068 sub ReadSynchProfile { 4069 my $prog = shift; 4070 local *PROFILE = shift; 4071 my $header = shift; 4072 4073 my $map = ''; 4074 my $profile = {}; 4075 my $pcs = {}; 4076 my $sampling_period = 1; 4077 my $cyclespernanosec = 2.8; # Default assumption for old binaries 4078 my $seen_clockrate = 0; 4079 my $line; 4080 4081 my $index = 0; 4082 if ($main::opt_total_delay) { 4083 $index = 0; 4084 } elsif ($main::opt_contentions) { 4085 $index = 1; 4086 } elsif ($main::opt_mean_delay) { 4087 $index = 2; 4088 } 4089 4090 while ( $line = <PROFILE> ) { 4091 $line =~ s/\r//g; # turn windows-looking lines into unix-looking lines 4092 if ( $line =~ /^\s*(\d+)\s+(\d+) \@\s*(.*?)\s*$/ ) { 4093 my ($cycles, $count, $stack) = ($1, $2, $3); 4094 4095 # Convert cycles to nanoseconds 4096 $cycles /= $cyclespernanosec; 4097 4098 # Adjust for sampling done by application 4099 $cycles *= $sampling_period; 4100 $count *= $sampling_period; 4101 4102 my @values = ($cycles, $count, $cycles / $count); 4103 AddEntries($profile, $pcs, FixCallerAddresses($stack), $values[$index]); 4104 4105 } elsif ( $line =~ /^(slow release).*thread \d+ \@\s*(.*?)\s*$/ || 4106 $line =~ /^\s*(\d+) \@\s*(.*?)\s*$/ ) { 4107 my ($cycles, $stack) = ($1, $2); 4108 if ($cycles !~ /^\d+$/) { 4109 next; 4110 } 4111 4112 # Convert cycles to nanoseconds 4113 $cycles /= $cyclespernanosec; 4114 4115 # Adjust for sampling done by application 4116 $cycles *= $sampling_period; 4117 4118 AddEntries($profile, $pcs, FixCallerAddresses($stack), $cycles); 4119 4120 } elsif ( $line =~ m/^([a-z][^=]*)=(.*)$/ ) { 4121 my ($variable, $value) = ($1,$2); 4122 for ($variable, $value) { 4123 s/^\s+//; 4124 s/\s+$//; 4125 } 4126 if ($variable eq "cycles/second") { 4127 $cyclespernanosec = $value / 1e9; 4128 $seen_clockrate = 1; 4129 } elsif ($variable eq "sampling period") { 4130 $sampling_period = $value; 4131 } elsif ($variable eq "ms since reset") { 4132 # Currently nothing is done with this value in pprof 4133 # So we just silently ignore it for now 4134 } elsif ($variable eq "discarded samples") { 4135 # Currently nothing is done with this value in pprof 4136 # So we just silently ignore it for now 4137 } else { 4138 printf STDERR ("Ignoring unnknown variable in /contention output: " . 4139 "'%s' = '%s'\n",$variable,$value); 4140 } 4141 } else { 4142 # Memory map entry 4143 $map .= $line; 4144 } 4145 } 4146 4147 if (!$seen_clockrate) { 4148 printf STDERR ("No cycles/second entry in profile; Guessing %.1f GHz\n", 4149 $cyclespernanosec); 4150 } 4151 4152 my $r = {}; 4153 $r->{version} = 0; 4154 $r->{period} = $sampling_period; 4155 $r->{profile} = $profile; 4156 $r->{libs} = ParseLibraries($prog, $map, $pcs); 4157 $r->{pcs} = $pcs; 4158 return $r; 4159 } 4160 4161 # Given a hex value in the form "0x1abcd" or "1abcd", return either 4162 # "0001abcd" or "000000000001abcd", depending on the current (global) 4163 # address length. 4164 sub HexExtend { 4165 my $addr = shift; 4166 4167 $addr =~ s/^(0x)?0*//; 4168 my $zeros_needed = $address_length - length($addr); 4169 if ($zeros_needed < 0) { 4170 printf STDERR "Warning: address $addr is longer than address length $address_length\n"; 4171 return $addr; 4172 } 4173 return ("0" x $zeros_needed) . $addr; 4174 } 4175 4176 ##### Symbol extraction ##### 4177 4178 # Aggressively search the lib_prefix values for the given library 4179 # If all else fails, just return the name of the library unmodified. 4180 # If the lib_prefix is "/my/path,/other/path" and $file is "/lib/dir/mylib.so" 4181 # it will search the following locations in this order, until it finds a file: 4182 # /my/path/lib/dir/mylib.so 4183 # /other/path/lib/dir/mylib.so 4184 # /my/path/dir/mylib.so 4185 # /other/path/dir/mylib.so 4186 # /my/path/mylib.so 4187 # /other/path/mylib.so 4188 # /lib/dir/mylib.so (returned as last resort) 4189 sub FindLibrary { 4190 my $file = shift; 4191 my $suffix = $file; 4192 4193 # Search for the library as described above 4194 do { 4195 foreach my $prefix (@prefix_list) { 4196 my $fullpath = $prefix . $suffix; 4197 if (-e $fullpath) { 4198 return $fullpath; 4199 } 4200 } 4201 } while ($suffix =~ s|^/[^/]+/|/|); 4202 return $file; 4203 } 4204 4205 # Return path to library with debugging symbols. 4206 # For libc libraries, the copy in /usr/lib/debug contains debugging symbols 4207 sub DebuggingLibrary { 4208 my $file = shift; 4209 if ($file =~ m|^/|) { 4210 if (-f "/usr/lib/debug$file") { 4211 return "/usr/lib/debug$file"; 4212 } elsif (-f "/usr/lib/debug$file.debug") { 4213 return "/usr/lib/debug$file.debug"; 4214 } 4215 } 4216 return undef; 4217 } 4218 4219 # Parse text section header of a library using objdump 4220 sub ParseTextSectionHeaderFromObjdump { 4221 my $lib = shift; 4222 4223 my $size = undef; 4224 my $vma; 4225 my $file_offset; 4226 # Get objdump output from the library file to figure out how to 4227 # map between mapped addresses and addresses in the library. 4228 my $cmd = ShellEscape($obj_tool_map{"objdump"}, "-h", $lib); 4229 open(OBJDUMP, "$cmd |") || error("$cmd: $!\n"); 4230 while (<OBJDUMP>) { 4231 s/\r//g; # turn windows-looking lines into unix-looking lines 4232 # Idx Name Size VMA LMA File off Algn 4233 # 10 .text 00104b2c 420156f0 420156f0 000156f0 2**4 4234 # For 64-bit objects, VMA and LMA will be 16 hex digits, size and file 4235 # offset may still be 8. But AddressSub below will still handle that. 4236 my @x = split; 4237 if (($#x >= 6) && ($x[1] eq '.text')) { 4238 $size = $x[2]; 4239 $vma = $x[3]; 4240 $file_offset = $x[5]; 4241 last; 4242 } 4243 } 4244 close(OBJDUMP); 4245 4246 if (!defined($size)) { 4247 return undef; 4248 } 4249 4250 my $r = {}; 4251 $r->{size} = $size; 4252 $r->{vma} = $vma; 4253 $r->{file_offset} = $file_offset; 4254 4255 return $r; 4256 } 4257 4258 # Parse text section header of a library using otool (on OS X) 4259 sub ParseTextSectionHeaderFromOtool { 4260 my $lib = shift; 4261 4262 my $size = undef; 4263 my $vma = undef; 4264 my $file_offset = undef; 4265 # Get otool output from the library file to figure out how to 4266 # map between mapped addresses and addresses in the library. 4267 my $command = ShellEscape($obj_tool_map{"otool"}, "-l", $lib); 4268 open(OTOOL, "$command |") || error("$command: $!\n"); 4269 my $cmd = ""; 4270 my $sectname = ""; 4271 my $segname = ""; 4272 foreach my $line (<OTOOL>) { 4273 $line =~ s/\r//g; # turn windows-looking lines into unix-looking lines 4274 # Load command <#> 4275 # cmd LC_SEGMENT 4276 # [...] 4277 # Section 4278 # sectname __text 4279 # segname __TEXT 4280 # addr 0x000009f8 4281 # size 0x00018b9e 4282 # offset 2552 4283 # align 2^2 (4) 4284 # We will need to strip off the leading 0x from the hex addresses, 4285 # and convert the offset into hex. 4286 if ($line =~ /Load command/) { 4287 $cmd = ""; 4288 $sectname = ""; 4289 $segname = ""; 4290 } elsif ($line =~ /Section/) { 4291 $sectname = ""; 4292 $segname = ""; 4293 } elsif ($line =~ /cmd (\w+)/) { 4294 $cmd = $1; 4295 } elsif ($line =~ /sectname (\w+)/) { 4296 $sectname = $1; 4297 } elsif ($line =~ /segname (\w+)/) { 4298 $segname = $1; 4299 } elsif (!(($cmd eq "LC_SEGMENT" || $cmd eq "LC_SEGMENT_64") && 4300 $sectname eq "__text" && 4301 $segname eq "__TEXT")) { 4302 next; 4303 } elsif ($line =~ /\baddr 0x([0-9a-fA-F]+)/) { 4304 $vma = $1; 4305 } elsif ($line =~ /\bsize 0x([0-9a-fA-F]+)/) { 4306 $size = $1; 4307 } elsif ($line =~ /\boffset ([0-9]+)/) { 4308 $file_offset = sprintf("%016x", $1); 4309 } 4310 if (defined($vma) && defined($size) && defined($file_offset)) { 4311 last; 4312 } 4313 } 4314 close(OTOOL); 4315 4316 if (!defined($vma) || !defined($size) || !defined($file_offset)) { 4317 return undef; 4318 } 4319 4320 my $r = {}; 4321 $r->{size} = $size; 4322 $r->{vma} = $vma; 4323 $r->{file_offset} = $file_offset; 4324 4325 return $r; 4326 } 4327 4328 sub ParseTextSectionHeader { 4329 # obj_tool_map("otool") is only defined if we're in a Mach-O environment 4330 if (defined($obj_tool_map{"otool"})) { 4331 my $r = ParseTextSectionHeaderFromOtool(@_); 4332 if (defined($r)){ 4333 return $r; 4334 } 4335 } 4336 # If otool doesn't work, or we don't have it, fall back to objdump 4337 return ParseTextSectionHeaderFromObjdump(@_); 4338 } 4339 4340 # Split /proc/pid/maps dump into a list of libraries 4341 sub ParseLibraries { 4342 return if $main::use_symbol_page; # We don't need libraries info. 4343 my $prog = shift; 4344 my $map = shift; 4345 my $pcs = shift; 4346 4347 my $result = []; 4348 my $h = "[a-f0-9]+"; 4349 my $zero_offset = HexExtend("0"); 4350 4351 my $buildvar = ""; 4352 foreach my $l (split("\n", $map)) { 4353 if ($l =~ m/^\s*build=(.*)$/) { 4354 $buildvar = $1; 4355 } 4356 4357 my $start; 4358 my $finish; 4359 my $offset; 4360 my $lib; 4361 if ($l =~ /^($h)-($h)\s+..x.\s+($h)\s+\S+:\S+\s+\d+\s+(\S+\.(so|dll|dylib|bundle)((\.\d+)+\w*(\.\d+){0,3})?)$/i) { 4362 # Full line from /proc/self/maps. Example: 4363 # 40000000-40015000 r-xp 00000000 03:01 12845071 /lib/ld-2.3.2.so 4364 $start = HexExtend($1); 4365 $finish = HexExtend($2); 4366 $offset = HexExtend($3); 4367 $lib = $4; 4368 $lib =~ s|\\|/|g; # turn windows-style paths into unix-style paths 4369 } elsif ($l =~ /^\s*($h)-($h):\s*(\S+\.so(\.\d+)*)/) { 4370 # Cooked line from DumpAddressMap. Example: 4371 # 40000000-40015000: /lib/ld-2.3.2.so 4372 $start = HexExtend($1); 4373 $finish = HexExtend($2); 4374 $offset = $zero_offset; 4375 $lib = $3; 4376 } 4377 # FreeBSD 10.0 virtual memory map /proc/curproc/map as defined in 4378 # function procfs_doprocmap (sys/fs/procfs/procfs_map.c) 4379 # 4380 # Example: 4381 # 0x800600000 0x80061a000 26 0 0xfffff800035a0000 r-x 75 33 0x1004 COW NC vnode /libexec/ld-elf.s 4382 # o.1 NCH -1 4383 elsif ($l =~ /^(0x$h)\s(0x$h)\s\d+\s\d+\s0x$h\sr-x\s\d+\s\d+\s0x\d+\s(COW|NCO)\s(NC|NNC)\svnode\s(\S+\.so(\.\d+)*)/) { 4384 $start = HexExtend($1); 4385 $finish = HexExtend($2); 4386 $offset = $zero_offset; 4387 $lib = FindLibrary($5); 4388 4389 } else { 4390 next; 4391 } 4392 4393 # Expand "$build" variable if available 4394 $lib =~ s/\$build\b/$buildvar/g; 4395 4396 $lib = FindLibrary($lib); 4397 4398 # Check for pre-relocated libraries, which use pre-relocated symbol tables 4399 # and thus require adjusting the offset that we'll use to translate 4400 # VM addresses into symbol table addresses. 4401 # Only do this if we're not going to fetch the symbol table from a 4402 # debugging copy of the library. 4403 if (!DebuggingLibrary($lib)) { 4404 my $text = ParseTextSectionHeader($lib); 4405 if (defined($text)) { 4406 my $vma_offset = AddressSub($text->{vma}, $text->{file_offset}); 4407 $offset = AddressAdd($offset, $vma_offset); 4408 } 4409 } 4410 4411 if($main::opt_debug) { printf STDERR "$start:$finish ($offset) $lib\n"; } 4412 push(@{$result}, [$lib, $start, $finish, $offset]); 4413 } 4414 4415 # Append special entry for additional library (not relocated) 4416 if ($main::opt_lib ne "") { 4417 my $text = ParseTextSectionHeader($main::opt_lib); 4418 if (defined($text)) { 4419 my $start = $text->{vma}; 4420 my $finish = AddressAdd($start, $text->{size}); 4421 4422 push(@{$result}, [$main::opt_lib, $start, $finish, $start]); 4423 } 4424 } 4425 4426 # Append special entry for the main program. This covers 4427 # 0..max_pc_value_seen, so that we assume pc values not found in one 4428 # of the library ranges will be treated as coming from the main 4429 # program binary. 4430 my $min_pc = HexExtend("0"); 4431 my $max_pc = $min_pc; # find the maximal PC value in any sample 4432 foreach my $pc (keys(%{$pcs})) { 4433 if (HexExtend($pc) gt $max_pc) { $max_pc = HexExtend($pc); } 4434 } 4435 push(@{$result}, [$prog, $min_pc, $max_pc, $zero_offset]); 4436 4437 return $result; 4438 } 4439 4440 # Add two hex addresses of length $address_length. 4441 # Run pprof --test for unit test if this is changed. 4442 sub AddressAdd { 4443 my $addr1 = shift; 4444 my $addr2 = shift; 4445 my $sum; 4446 4447 if ($address_length == 8) { 4448 # Perl doesn't cope with wraparound arithmetic, so do it explicitly: 4449 $sum = (hex($addr1)+hex($addr2)) % (0x10000000 * 16); 4450 return sprintf("%08x", $sum); 4451 4452 } else { 4453 # Do the addition in 7-nibble chunks to trivialize carry handling. 4454 4455 if ($main::opt_debug and $main::opt_test) { 4456 print STDERR "AddressAdd $addr1 + $addr2 = "; 4457 } 4458 4459 my $a1 = substr($addr1,-7); 4460 $addr1 = substr($addr1,0,-7); 4461 my $a2 = substr($addr2,-7); 4462 $addr2 = substr($addr2,0,-7); 4463 $sum = hex($a1) + hex($a2); 4464 my $c = 0; 4465 if ($sum > 0xfffffff) { 4466 $c = 1; 4467 $sum -= 0x10000000; 4468 } 4469 my $r = sprintf("%07x", $sum); 4470 4471 $a1 = substr($addr1,-7); 4472 $addr1 = substr($addr1,0,-7); 4473 $a2 = substr($addr2,-7); 4474 $addr2 = substr($addr2,0,-7); 4475 $sum = hex($a1) + hex($a2) + $c; 4476 $c = 0; 4477 if ($sum > 0xfffffff) { 4478 $c = 1; 4479 $sum -= 0x10000000; 4480 } 4481 $r = sprintf("%07x", $sum) . $r; 4482 4483 $sum = hex($addr1) + hex($addr2) + $c; 4484 if ($sum > 0xff) { $sum -= 0x100; } 4485 $r = sprintf("%02x", $sum) . $r; 4486 4487 if ($main::opt_debug and $main::opt_test) { print STDERR "$r\n"; } 4488 4489 return $r; 4490 } 4491 } 4492 4493 4494 # Subtract two hex addresses of length $address_length. 4495 # Run pprof --test for unit test if this is changed. 4496 sub AddressSub { 4497 my $addr1 = shift; 4498 my $addr2 = shift; 4499 my $diff; 4500 4501 if ($address_length == 8) { 4502 # Perl doesn't cope with wraparound arithmetic, so do it explicitly: 4503 $diff = (hex($addr1)-hex($addr2)) % (0x10000000 * 16); 4504 return sprintf("%08x", $diff); 4505 4506 } else { 4507 # Do the addition in 7-nibble chunks to trivialize borrow handling. 4508 # if ($main::opt_debug) { print STDERR "AddressSub $addr1 - $addr2 = "; } 4509 4510 my $a1 = hex(substr($addr1,-7)); 4511 $addr1 = substr($addr1,0,-7); 4512 my $a2 = hex(substr($addr2,-7)); 4513 $addr2 = substr($addr2,0,-7); 4514 my $b = 0; 4515 if ($a2 > $a1) { 4516 $b = 1; 4517 $a1 += 0x10000000; 4518 } 4519 $diff = $a1 - $a2; 4520 my $r = sprintf("%07x", $diff); 4521 4522 $a1 = hex(substr($addr1,-7)); 4523 $addr1 = substr($addr1,0,-7); 4524 $a2 = hex(substr($addr2,-7)) + $b; 4525 $addr2 = substr($addr2,0,-7); 4526 $b = 0; 4527 if ($a2 > $a1) { 4528 $b = 1; 4529 $a1 += 0x10000000; 4530 } 4531 $diff = $a1 - $a2; 4532 $r = sprintf("%07x", $diff) . $r; 4533 4534 $a1 = hex($addr1); 4535 $a2 = hex($addr2) + $b; 4536 if ($a2 > $a1) { $a1 += 0x100; } 4537 $diff = $a1 - $a2; 4538 $r = sprintf("%02x", $diff) . $r; 4539 4540 # if ($main::opt_debug) { print STDERR "$r\n"; } 4541 4542 return $r; 4543 } 4544 } 4545 4546 # Increment a hex addresses of length $address_length. 4547 # Run pprof --test for unit test if this is changed. 4548 sub AddressInc { 4549 my $addr = shift; 4550 my $sum; 4551 4552 if ($address_length == 8) { 4553 # Perl doesn't cope with wraparound arithmetic, so do it explicitly: 4554 $sum = (hex($addr)+1) % (0x10000000 * 16); 4555 return sprintf("%08x", $sum); 4556 4557 } else { 4558 # Do the addition in 7-nibble chunks to trivialize carry handling. 4559 # We are always doing this to step through the addresses in a function, 4560 # and will almost never overflow the first chunk, so we check for this 4561 # case and exit early. 4562 4563 # if ($main::opt_debug) { print STDERR "AddressInc $addr1 = "; } 4564 4565 my $a1 = substr($addr,-7); 4566 $addr = substr($addr,0,-7); 4567 $sum = hex($a1) + 1; 4568 my $r = sprintf("%07x", $sum); 4569 if ($sum <= 0xfffffff) { 4570 $r = $addr . $r; 4571 # if ($main::opt_debug) { print STDERR "$r\n"; } 4572 return HexExtend($r); 4573 } else { 4574 $r = "0000000"; 4575 } 4576 4577 $a1 = substr($addr,-7); 4578 $addr = substr($addr,0,-7); 4579 $sum = hex($a1) + 1; 4580 $r = sprintf("%07x", $sum) . $r; 4581 if ($sum <= 0xfffffff) { 4582 $r = $addr . $r; 4583 # if ($main::opt_debug) { print STDERR "$r\n"; } 4584 return HexExtend($r); 4585 } else { 4586 $r = "00000000000000"; 4587 } 4588 4589 $sum = hex($addr) + 1; 4590 if ($sum > 0xff) { $sum -= 0x100; } 4591 $r = sprintf("%02x", $sum) . $r; 4592 4593 # if ($main::opt_debug) { print STDERR "$r\n"; } 4594 return $r; 4595 } 4596 } 4597 4598 # Extract symbols for all PC values found in profile 4599 sub ExtractSymbols { 4600 my $libs = shift; 4601 my $pcset = shift; 4602 4603 my $symbols = {}; 4604 4605 # Map each PC value to the containing library. To make this faster, 4606 # we sort libraries by their starting pc value (highest first), and 4607 # advance through the libraries as we advance the pc. Sometimes the 4608 # addresses of libraries may overlap with the addresses of the main 4609 # binary, so to make sure the libraries 'win', we iterate over the 4610 # libraries in reverse order (which assumes the binary doesn't start 4611 # in the middle of a library, which seems a fair assumption). 4612 my @pcs = (sort { $a cmp $b } keys(%{$pcset})); # pcset is 0-extended strings 4613 foreach my $lib (sort {$b->[1] cmp $a->[1]} @{$libs}) { 4614 my $libname = $lib->[0]; 4615 my $start = $lib->[1]; 4616 my $finish = $lib->[2]; 4617 my $offset = $lib->[3]; 4618 4619 # Use debug library if it exists 4620 my $debug_libname = DebuggingLibrary($libname); 4621 if ($debug_libname) { 4622 $libname = $debug_libname; 4623 } 4624 4625 # Get list of pcs that belong in this library. 4626 my $contained = []; 4627 my ($start_pc_index, $finish_pc_index); 4628 # Find smallest finish_pc_index such that $finish < $pc[$finish_pc_index]. 4629 for ($finish_pc_index = $#pcs + 1; $finish_pc_index > 0; 4630 $finish_pc_index--) { 4631 last if $pcs[$finish_pc_index - 1] le $finish; 4632 } 4633 # Find smallest start_pc_index such that $start <= $pc[$start_pc_index]. 4634 for ($start_pc_index = $finish_pc_index; $start_pc_index > 0; 4635 $start_pc_index--) { 4636 last if $pcs[$start_pc_index - 1] lt $start; 4637 } 4638 # This keeps PC values higher than $pc[$finish_pc_index] in @pcs, 4639 # in case there are overlaps in libraries and the main binary. 4640 @{$contained} = splice(@pcs, $start_pc_index, 4641 $finish_pc_index - $start_pc_index); 4642 # Map to symbols 4643 MapToSymbols($libname, AddressSub($start, $offset), $contained, $symbols); 4644 } 4645 4646 return $symbols; 4647 } 4648 4649 # Map list of PC values to symbols for a given image 4650 sub MapToSymbols { 4651 my $image = shift; 4652 my $offset = shift; 4653 my $pclist = shift; 4654 my $symbols = shift; 4655 4656 my $debug = 0; 4657 4658 # Ignore empty binaries 4659 if ($#{$pclist} < 0) { return; } 4660 4661 # Figure out the addr2line command to use 4662 my $addr2line = $obj_tool_map{"addr2line"}; 4663 my $cmd = ShellEscape($addr2line, "-f", "-C", "-e", $image); 4664 if (exists $obj_tool_map{"addr2line_pdb"}) { 4665 $addr2line = $obj_tool_map{"addr2line_pdb"}; 4666 $cmd = ShellEscape($addr2line, "--demangle", "-f", "-C", "-e", $image); 4667 } 4668 4669 # If "addr2line" isn't installed on the system at all, just use 4670 # nm to get what info we can (function names, but not line numbers). 4671 if (system(ShellEscape($addr2line, "--help") . " >$dev_null 2>&1") != 0) { 4672 MapSymbolsWithNM($image, $offset, $pclist, $symbols); 4673 return; 4674 } 4675 4676 # "addr2line -i" can produce a variable number of lines per input 4677 # address, with no separator that allows us to tell when data for 4678 # the next address starts. So we find the address for a special 4679 # symbol (_fini) and interleave this address between all real 4680 # addresses passed to addr2line. The name of this special symbol 4681 # can then be used as a separator. 4682 $sep_address = undef; # May be filled in by MapSymbolsWithNM() 4683 my $nm_symbols = {}; 4684 MapSymbolsWithNM($image, $offset, $pclist, $nm_symbols); 4685 if (defined($sep_address)) { 4686 # Only add " -i" to addr2line if the binary supports it. 4687 # addr2line --help returns 0, but not if it sees an unknown flag first. 4688 if (system("$cmd -i --help >$dev_null 2>&1") == 0) { 4689 $cmd .= " -i"; 4690 } else { 4691 $sep_address = undef; # no need for sep_address if we don't support -i 4692 } 4693 } 4694 4695 # Make file with all PC values with intervening 'sep_address' so 4696 # that we can reliably detect the end of inlined function list 4697 open(ADDRESSES, ">$main::tmpfile_sym") || error("$main::tmpfile_sym: $!\n"); 4698 if ($debug) { print("---- $image ---\n"); } 4699 for (my $i = 0; $i <= $#{$pclist}; $i++) { 4700 # addr2line always reads hex addresses, and does not need '0x' prefix. 4701 if ($debug) { printf STDERR ("%s\n", $pclist->[$i]); } 4702 printf ADDRESSES ("%s\n", AddressSub($pclist->[$i], $offset)); 4703 if (defined($sep_address)) { 4704 printf ADDRESSES ("%s\n", $sep_address); 4705 } 4706 } 4707 close(ADDRESSES); 4708 if ($debug) { 4709 print("----\n"); 4710 system("cat", $main::tmpfile_sym); 4711 print("----\n"); 4712 system("$cmd < " . ShellEscape($main::tmpfile_sym)); 4713 print("----\n"); 4714 } 4715 4716 open(SYMBOLS, "$cmd <" . ShellEscape($main::tmpfile_sym) . " |") 4717 || error("$cmd: $!\n"); 4718 my $count = 0; # Index in pclist 4719 while (<SYMBOLS>) { 4720 # Read fullfunction and filelineinfo from next pair of lines 4721 s/\r?\n$//g; 4722 my $fullfunction = $_; 4723 $_ = <SYMBOLS>; 4724 s/\r?\n$//g; 4725 my $filelinenum = $_; 4726 4727 if (defined($sep_address) && $fullfunction eq $sep_symbol) { 4728 # Terminating marker for data for this address 4729 $count++; 4730 next; 4731 } 4732 4733 $filelinenum =~ s|\\|/|g; # turn windows-style paths into unix-style paths 4734 4735 my $pcstr = $pclist->[$count]; 4736 my $function = ShortFunctionName($fullfunction); 4737 my $nms = $nm_symbols->{$pcstr}; 4738 if (defined($nms)) { 4739 if ($fullfunction eq '??') { 4740 # nm found a symbol for us. 4741 $function = $nms->[0]; 4742 $fullfunction = $nms->[2]; 4743 } else { 4744 # MapSymbolsWithNM tags each routine with its starting address, 4745 # useful in case the image has multiple occurrences of this 4746 # routine. (It uses a syntax that resembles template paramters, 4747 # that are automatically stripped out by ShortFunctionName().) 4748 # addr2line does not provide the same information. So we check 4749 # if nm disambiguated our symbol, and if so take the annotated 4750 # (nm) version of the routine-name. TODO(csilvers): this won't 4751 # catch overloaded, inlined symbols, which nm doesn't see. 4752 # Better would be to do a check similar to nm's, in this fn. 4753 if ($nms->[2] =~ m/^\Q$function\E/) { # sanity check it's the right fn 4754 $function = $nms->[0]; 4755 $fullfunction = $nms->[2]; 4756 } 4757 } 4758 } 4759 4760 # Prepend to accumulated symbols for pcstr 4761 # (so that caller comes before callee) 4762 my $sym = $symbols->{$pcstr}; 4763 if (!defined($sym)) { 4764 $sym = []; 4765 $symbols->{$pcstr} = $sym; 4766 } 4767 unshift(@{$sym}, $function, $filelinenum, $fullfunction); 4768 if ($debug) { printf STDERR ("%s => [%s]\n", $pcstr, join(" ", @{$sym})); } 4769 if (!defined($sep_address)) { 4770 # Inlining is off, so this entry ends immediately 4771 $count++; 4772 } 4773 } 4774 close(SYMBOLS); 4775 } 4776 4777 # Use nm to map the list of referenced PCs to symbols. Return true iff we 4778 # are able to read procedure information via nm. 4779 sub MapSymbolsWithNM { 4780 my $image = shift; 4781 my $offset = shift; 4782 my $pclist = shift; 4783 my $symbols = shift; 4784 4785 # Get nm output sorted by increasing address 4786 my $symbol_table = GetProcedureBoundaries($image, "."); 4787 if (!%{$symbol_table}) { 4788 return 0; 4789 } 4790 # Start addresses are already the right length (8 or 16 hex digits). 4791 my @names = sort { $symbol_table->{$a}->[0] cmp $symbol_table->{$b}->[0] } 4792 keys(%{$symbol_table}); 4793 4794 if ($#names < 0) { 4795 # No symbols: just use addresses 4796 foreach my $pc (@{$pclist}) { 4797 my $pcstr = "0x" . $pc; 4798 $symbols->{$pc} = [$pcstr, "?", $pcstr]; 4799 } 4800 return 0; 4801 } 4802 4803 # Sort addresses so we can do a join against nm output 4804 my $index = 0; 4805 my $fullname = $names[0]; 4806 my $name = ShortFunctionName($fullname); 4807 foreach my $pc (sort { $a cmp $b } @{$pclist}) { 4808 # Adjust for mapped offset 4809 my $mpc = AddressSub($pc, $offset); 4810 while (($index < $#names) && ($mpc ge $symbol_table->{$fullname}->[1])){ 4811 $index++; 4812 $fullname = $names[$index]; 4813 $name = ShortFunctionName($fullname); 4814 } 4815 if ($mpc lt $symbol_table->{$fullname}->[1]) { 4816 $symbols->{$pc} = [$name, "?", $fullname]; 4817 } else { 4818 my $pcstr = "0x" . $pc; 4819 $symbols->{$pc} = [$pcstr, "?", $pcstr]; 4820 } 4821 } 4822 return 1; 4823 } 4824 4825 sub ShortFunctionName { 4826 my $function = shift; 4827 while ($function =~ s/\([^()]*\)(\s*const)?//g) { } # Argument types 4828 while ($function =~ s/<[^<>]*>//g) { } # Remove template arguments 4829 $function =~ s/^.*\s+(\w+::)/$1/; # Remove leading type 4830 return $function; 4831 } 4832 4833 # Trim overly long symbols found in disassembler output 4834 sub CleanDisassembly { 4835 my $d = shift; 4836 while ($d =~ s/\([^()%]*\)(\s*const)?//g) { } # Argument types, not (%rax) 4837 while ($d =~ s/(\w+)<[^<>]*>/$1/g) { } # Remove template arguments 4838 return $d; 4839 } 4840 4841 # Clean file name for display 4842 sub CleanFileName { 4843 my ($f) = @_; 4844 $f =~ s|^/proc/self/cwd/||; 4845 $f =~ s|^\./||; 4846 return $f; 4847 } 4848 4849 # Make address relative to section and clean up for display 4850 sub UnparseAddress { 4851 my ($offset, $address) = @_; 4852 $address = AddressSub($address, $offset); 4853 $address =~ s/^0x//; 4854 $address =~ s/^0*//; 4855 return $address; 4856 } 4857 4858 ##### Miscellaneous ##### 4859 4860 # Find the right versions of the above object tools to use. The 4861 # argument is the program file being analyzed, and should be an ELF 4862 # 32-bit or ELF 64-bit executable file. The location of the tools 4863 # is determined by considering the following options in this order: 4864 # 1) --tools option, if set 4865 # 2) PPROF_TOOLS environment variable, if set 4866 # 3) the environment 4867 sub ConfigureObjTools { 4868 my $prog_file = shift; 4869 4870 # Check for the existence of $prog_file because /usr/bin/file does not 4871 # predictably return error status in prod. 4872 (-e $prog_file) || error("$prog_file does not exist.\n"); 4873 4874 my $file_type = undef; 4875 if (-e "/usr/bin/file") { 4876 # Follow symlinks (at least for systems where "file" supports that). 4877 my $escaped_prog_file = ShellEscape($prog_file); 4878 $file_type = `/usr/bin/file -L $escaped_prog_file 2>$dev_null || 4879 /usr/bin/file $escaped_prog_file`; 4880 } elsif ($^O == "MSWin32") { 4881 $file_type = "MS Windows"; 4882 } else { 4883 print STDERR "WARNING: Can't determine the file type of $prog_file"; 4884 } 4885 4886 if ($file_type =~ /64-bit/) { 4887 # Change $address_length to 16 if the program file is ELF 64-bit. 4888 # We can't detect this from many (most?) heap or lock contention 4889 # profiles, since the actual addresses referenced are generally in low 4890 # memory even for 64-bit programs. 4891 $address_length = 16; 4892 } 4893 4894 if ($file_type =~ /MS Windows/) { 4895 # For windows, we provide a version of nm and addr2line as part of 4896 # the opensource release, which is capable of parsing 4897 # Windows-style PDB executables. It should live in the path, or 4898 # in the same directory as pprof. 4899 $obj_tool_map{"nm_pdb"} = "nm-pdb"; 4900 $obj_tool_map{"addr2line_pdb"} = "addr2line-pdb"; 4901 } 4902 4903 if ($file_type =~ /Mach-O/) { 4904 # OS X uses otool to examine Mach-O files, rather than objdump. 4905 $obj_tool_map{"otool"} = "otool"; 4906 $obj_tool_map{"addr2line"} = "false"; # no addr2line 4907 $obj_tool_map{"objdump"} = "false"; # no objdump 4908 } 4909 4910 # Go fill in %obj_tool_map with the pathnames to use: 4911 foreach my $tool (keys %obj_tool_map) { 4912 $obj_tool_map{$tool} = ConfigureTool($obj_tool_map{$tool}); 4913 } 4914 } 4915 4916 # Returns the path of a caller-specified object tool. If --tools or 4917 # PPROF_TOOLS are specified, then returns the full path to the tool 4918 # with that prefix. Otherwise, returns the path unmodified (which 4919 # means we will look for it on PATH). 4920 sub ConfigureTool { 4921 my $tool = shift; 4922 my $path; 4923 4924 # --tools (or $PPROF_TOOLS) is a comma separated list, where each 4925 # item is either a) a pathname prefix, or b) a map of the form 4926 # <tool>:<path>. First we look for an entry of type (b) for our 4927 # tool. If one is found, we use it. Otherwise, we consider all the 4928 # pathname prefixes in turn, until one yields an existing file. If 4929 # none does, we use a default path. 4930 my $tools = $main::opt_tools || $ENV{"PPROF_TOOLS"} || ""; 4931 if ($tools =~ m/(,|^)\Q$tool\E:([^,]*)/) { 4932 $path = $2; 4933 # TODO(csilvers): sanity-check that $path exists? Hard if it's relative. 4934 } elsif ($tools ne '') { 4935 foreach my $prefix (split(',', $tools)) { 4936 next if ($prefix =~ /:/); # ignore "tool:fullpath" entries in the list 4937 if (-x $prefix . $tool) { 4938 $path = $prefix . $tool; 4939 last; 4940 } 4941 } 4942 if (!$path) { 4943 error("No '$tool' found with prefix specified by " . 4944 "--tools (or \$PPROF_TOOLS) '$tools'\n"); 4945 } 4946 } else { 4947 # ... otherwise use the version that exists in the same directory as 4948 # pprof. If there's nothing there, use $PATH. 4949 $0 =~ m,[^/]*$,; # this is everything after the last slash 4950 my $dirname = $`; # this is everything up to and including the last slash 4951 if (-x "$dirname$tool") { 4952 $path = "$dirname$tool"; 4953 } else { 4954 $path = $tool; 4955 } 4956 } 4957 if ($main::opt_debug) { print STDERR "Using '$path' for '$tool'.\n"; } 4958 return $path; 4959 } 4960 4961 sub ShellEscape { 4962 my @escaped_words = (); 4963 foreach my $word (@_) { 4964 my $escaped_word = $word; 4965 if ($word =~ m![^a-zA-Z0-9/.,_=-]!) { # check for anything not in whitelist 4966 $escaped_word =~ s/'/'\\''/; 4967 $escaped_word = "'$escaped_word'"; 4968 } 4969 push(@escaped_words, $escaped_word); 4970 } 4971 return join(" ", @escaped_words); 4972 } 4973 4974 sub cleanup { 4975 unlink($main::tmpfile_sym); 4976 unlink(keys %main::tempnames); 4977 4978 # We leave any collected profiles in $HOME/pprof in case the user wants 4979 # to look at them later. We print a message informing them of this. 4980 if ((scalar(@main::profile_files) > 0) && 4981 defined($main::collected_profile)) { 4982 if (scalar(@main::profile_files) == 1) { 4983 print STDERR "Dynamically gathered profile is in $main::collected_profile\n"; 4984 } 4985 print STDERR "If you want to investigate this profile further, you can do:\n"; 4986 print STDERR "\n"; 4987 print STDERR " pprof \\\n"; 4988 print STDERR " $main::prog \\\n"; 4989 print STDERR " $main::collected_profile\n"; 4990 print STDERR "\n"; 4991 } 4992 } 4993 4994 sub sighandler { 4995 cleanup(); 4996 exit(1); 4997 } 4998 4999 sub error { 5000 my $msg = shift; 5001 print STDERR $msg; 5002 cleanup(); 5003 exit(1); 5004 } 5005 5006 5007 # Run $nm_command and get all the resulting procedure boundaries whose 5008 # names match "$regexp" and returns them in a hashtable mapping from 5009 # procedure name to a two-element vector of [start address, end address] 5010 sub GetProcedureBoundariesViaNm { 5011 my $escaped_nm_command = shift; # shell-escaped 5012 my $regexp = shift; 5013 5014 my $symbol_table = {}; 5015 open(NM, "$escaped_nm_command |") || error("$escaped_nm_command: $!\n"); 5016 my $last_start = "0"; 5017 my $routine = ""; 5018 while (<NM>) { 5019 s/\r//g; # turn windows-looking lines into unix-looking lines 5020 if (m/^\s*([0-9a-f]+) (.) (..*)/) { 5021 my $start_val = $1; 5022 my $type = $2; 5023 my $this_routine = $3; 5024 5025 # It's possible for two symbols to share the same address, if 5026 # one is a zero-length variable (like __start_google_malloc) or 5027 # one symbol is a weak alias to another (like __libc_malloc). 5028 # In such cases, we want to ignore all values except for the 5029 # actual symbol, which in nm-speak has type "T". The logic 5030 # below does this, though it's a bit tricky: what happens when 5031 # we have a series of lines with the same address, is the first 5032 # one gets queued up to be processed. However, it won't 5033 # *actually* be processed until later, when we read a line with 5034 # a different address. That means that as long as we're reading 5035 # lines with the same address, we have a chance to replace that 5036 # item in the queue, which we do whenever we see a 'T' entry -- 5037 # that is, a line with type 'T'. If we never see a 'T' entry, 5038 # we'll just go ahead and process the first entry (which never 5039 # got touched in the queue), and ignore the others. 5040 if ($start_val eq $last_start && $type =~ /t/i) { 5041 # We are the 'T' symbol at this address, replace previous symbol. 5042 $routine = $this_routine; 5043 next; 5044 } elsif ($start_val eq $last_start) { 5045 # We're not the 'T' symbol at this address, so ignore us. 5046 next; 5047 } 5048 5049 if ($this_routine eq $sep_symbol) { 5050 $sep_address = HexExtend($start_val); 5051 } 5052 5053 # Tag this routine with the starting address in case the image 5054 # has multiple occurrences of this routine. We use a syntax 5055 # that resembles template parameters that are automatically 5056 # stripped out by ShortFunctionName() 5057 $this_routine .= "<$start_val>"; 5058 5059 if (defined($routine) && $routine =~ m/$regexp/) { 5060 $symbol_table->{$routine} = [HexExtend($last_start), 5061 HexExtend($start_val)]; 5062 } 5063 $last_start = $start_val; 5064 $routine = $this_routine; 5065 } elsif (m/^Loaded image name: (.+)/) { 5066 # The win32 nm workalike emits information about the binary it is using. 5067 if ($main::opt_debug) { print STDERR "Using Image $1\n"; } 5068 } elsif (m/^PDB file name: (.+)/) { 5069 # The win32 nm workalike emits information about the pdb it is using. 5070 if ($main::opt_debug) { print STDERR "Using PDB $1\n"; } 5071 } 5072 } 5073 close(NM); 5074 # Handle the last line in the nm output. Unfortunately, we don't know 5075 # how big this last symbol is, because we don't know how big the file 5076 # is. For now, we just give it a size of 0. 5077 # TODO(csilvers): do better here. 5078 if (defined($routine) && $routine =~ m/$regexp/) { 5079 $symbol_table->{$routine} = [HexExtend($last_start), 5080 HexExtend($last_start)]; 5081 } 5082 return $symbol_table; 5083 } 5084 5085 # Gets the procedure boundaries for all routines in "$image" whose names 5086 # match "$regexp" and returns them in a hashtable mapping from procedure 5087 # name to a two-element vector of [start address, end address]. 5088 # Will return an empty map if nm is not installed or not working properly. 5089 sub GetProcedureBoundaries { 5090 my $image = shift; 5091 my $regexp = shift; 5092 5093 # If $image doesn't start with /, then put ./ in front of it. This works 5094 # around an obnoxious bug in our probing of nm -f behavior. 5095 # "nm -f $image" is supposed to fail on GNU nm, but if: 5096 # 5097 # a. $image starts with [BbSsPp] (for example, bin/foo/bar), AND 5098 # b. you have a.out in your current directory (a not uncommon occurence) 5099 # 5100 # then "nm -f $image" succeeds because -f only looks at the first letter of 5101 # the argument, which looks valid because it's [BbSsPp], and then since 5102 # there's no image provided, it looks for a.out and finds it. 5103 # 5104 # This regex makes sure that $image starts with . or /, forcing the -f 5105 # parsing to fail since . and / are not valid formats. 5106 $image =~ s#^[^/]#./$&#; 5107 5108 # For libc libraries, the copy in /usr/lib/debug contains debugging symbols 5109 my $debugging = DebuggingLibrary($image); 5110 if ($debugging) { 5111 $image = $debugging; 5112 } 5113 5114 my $nm = $obj_tool_map{"nm"}; 5115 my $cppfilt = $obj_tool_map{"c++filt"}; 5116 5117 # nm can fail for two reasons: 1) $image isn't a debug library; 2) nm 5118 # binary doesn't support --demangle. In addition, for OS X we need 5119 # to use the -f flag to get 'flat' nm output (otherwise we don't sort 5120 # properly and get incorrect results). Unfortunately, GNU nm uses -f 5121 # in an incompatible way. So first we test whether our nm supports 5122 # --demangle and -f. 5123 my $demangle_flag = ""; 5124 my $cppfilt_flag = ""; 5125 my $to_devnull = ">$dev_null 2>&1"; 5126 if (system(ShellEscape($nm, "--demangle", "image") . $to_devnull) == 0) { 5127 # In this mode, we do "nm --demangle <foo>" 5128 $demangle_flag = "--demangle"; 5129 $cppfilt_flag = ""; 5130 } elsif (system(ShellEscape($cppfilt, $image) . $to_devnull) == 0) { 5131 # In this mode, we do "nm <foo> | c++filt" 5132 $cppfilt_flag = " | " . ShellEscape($cppfilt); 5133 }; 5134 my $flatten_flag = ""; 5135 if (system(ShellEscape($nm, "-f", $image) . $to_devnull) == 0) { 5136 $flatten_flag = "-f"; 5137 } 5138 5139 # Finally, in the case $imagie isn't a debug library, we try again with 5140 # -D to at least get *exported* symbols. If we can't use --demangle, 5141 # we use c++filt instead, if it exists on this system. 5142 my @nm_commands = (ShellEscape($nm, "-n", $flatten_flag, $demangle_flag, 5143 $image) . " 2>$dev_null $cppfilt_flag", 5144 ShellEscape($nm, "-D", "-n", $flatten_flag, $demangle_flag, 5145 $image) . " 2>$dev_null $cppfilt_flag", 5146 # 6nm is for Go binaries 5147 ShellEscape("6nm", "$image") . " 2>$dev_null | sort", 5148 ); 5149 5150 # If the executable is an MS Windows PDB-format executable, we'll 5151 # have set up obj_tool_map("nm_pdb"). In this case, we actually 5152 # want to use both unix nm and windows-specific nm_pdb, since 5153 # PDB-format executables can apparently include dwarf .o files. 5154 if (exists $obj_tool_map{"nm_pdb"}) { 5155 push(@nm_commands, 5156 ShellEscape($obj_tool_map{"nm_pdb"}, "--demangle", $image) 5157 . " 2>$dev_null"); 5158 } 5159 5160 foreach my $nm_command (@nm_commands) { 5161 my $symbol_table = GetProcedureBoundariesViaNm($nm_command, $regexp); 5162 return $symbol_table if (%{$symbol_table}); 5163 } 5164 my $symbol_table = {}; 5165 return $symbol_table; 5166 } 5167 5168 5169 # The test vectors for AddressAdd/Sub/Inc are 8-16-nibble hex strings. 5170 # To make them more readable, we add underscores at interesting places. 5171 # This routine removes the underscores, producing the canonical representation 5172 # used by pprof to represent addresses, particularly in the tested routines. 5173 sub CanonicalHex { 5174 my $arg = shift; 5175 return join '', (split '_',$arg); 5176 } 5177 5178 5179 # Unit test for AddressAdd: 5180 sub AddressAddUnitTest { 5181 my $test_data_8 = shift; 5182 my $test_data_16 = shift; 5183 my $error_count = 0; 5184 my $fail_count = 0; 5185 my $pass_count = 0; 5186 # print STDERR "AddressAddUnitTest: ", 1+$#{$test_data_8}, " tests\n"; 5187 5188 # First a few 8-nibble addresses. Note that this implementation uses 5189 # plain old arithmetic, so a quick sanity check along with verifying what 5190 # happens to overflow (we want it to wrap): 5191 $address_length = 8; 5192 foreach my $row (@{$test_data_8}) { 5193 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5194 my $sum = AddressAdd ($row->[0], $row->[1]); 5195 if ($sum ne $row->[2]) { 5196 printf STDERR "ERROR: %s != %s + %s = %s\n", $sum, 5197 $row->[0], $row->[1], $row->[2]; 5198 ++$fail_count; 5199 } else { 5200 ++$pass_count; 5201 } 5202 } 5203 printf STDERR "AddressAdd 32-bit tests: %d passes, %d failures\n", 5204 $pass_count, $fail_count; 5205 $error_count = $fail_count; 5206 $fail_count = 0; 5207 $pass_count = 0; 5208 5209 # Now 16-nibble addresses. 5210 $address_length = 16; 5211 foreach my $row (@{$test_data_16}) { 5212 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5213 my $sum = AddressAdd (CanonicalHex($row->[0]), CanonicalHex($row->[1])); 5214 my $expected = join '', (split '_',$row->[2]); 5215 if ($sum ne CanonicalHex($row->[2])) { 5216 printf STDERR "ERROR: %s != %s + %s = %s\n", $sum, 5217 $row->[0], $row->[1], $row->[2]; 5218 ++$fail_count; 5219 } else { 5220 ++$pass_count; 5221 } 5222 } 5223 printf STDERR "AddressAdd 64-bit tests: %d passes, %d failures\n", 5224 $pass_count, $fail_count; 5225 $error_count += $fail_count; 5226 5227 return $error_count; 5228 } 5229 5230 5231 # Unit test for AddressSub: 5232 sub AddressSubUnitTest { 5233 my $test_data_8 = shift; 5234 my $test_data_16 = shift; 5235 my $error_count = 0; 5236 my $fail_count = 0; 5237 my $pass_count = 0; 5238 # print STDERR "AddressSubUnitTest: ", 1+$#{$test_data_8}, " tests\n"; 5239 5240 # First a few 8-nibble addresses. Note that this implementation uses 5241 # plain old arithmetic, so a quick sanity check along with verifying what 5242 # happens to overflow (we want it to wrap): 5243 $address_length = 8; 5244 foreach my $row (@{$test_data_8}) { 5245 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5246 my $sum = AddressSub ($row->[0], $row->[1]); 5247 if ($sum ne $row->[3]) { 5248 printf STDERR "ERROR: %s != %s - %s = %s\n", $sum, 5249 $row->[0], $row->[1], $row->[3]; 5250 ++$fail_count; 5251 } else { 5252 ++$pass_count; 5253 } 5254 } 5255 printf STDERR "AddressSub 32-bit tests: %d passes, %d failures\n", 5256 $pass_count, $fail_count; 5257 $error_count = $fail_count; 5258 $fail_count = 0; 5259 $pass_count = 0; 5260 5261 # Now 16-nibble addresses. 5262 $address_length = 16; 5263 foreach my $row (@{$test_data_16}) { 5264 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5265 my $sum = AddressSub (CanonicalHex($row->[0]), CanonicalHex($row->[1])); 5266 if ($sum ne CanonicalHex($row->[3])) { 5267 printf STDERR "ERROR: %s != %s - %s = %s\n", $sum, 5268 $row->[0], $row->[1], $row->[3]; 5269 ++$fail_count; 5270 } else { 5271 ++$pass_count; 5272 } 5273 } 5274 printf STDERR "AddressSub 64-bit tests: %d passes, %d failures\n", 5275 $pass_count, $fail_count; 5276 $error_count += $fail_count; 5277 5278 return $error_count; 5279 } 5280 5281 5282 # Unit test for AddressInc: 5283 sub AddressIncUnitTest { 5284 my $test_data_8 = shift; 5285 my $test_data_16 = shift; 5286 my $error_count = 0; 5287 my $fail_count = 0; 5288 my $pass_count = 0; 5289 # print STDERR "AddressIncUnitTest: ", 1+$#{$test_data_8}, " tests\n"; 5290 5291 # First a few 8-nibble addresses. Note that this implementation uses 5292 # plain old arithmetic, so a quick sanity check along with verifying what 5293 # happens to overflow (we want it to wrap): 5294 $address_length = 8; 5295 foreach my $row (@{$test_data_8}) { 5296 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5297 my $sum = AddressInc ($row->[0]); 5298 if ($sum ne $row->[4]) { 5299 printf STDERR "ERROR: %s != %s + 1 = %s\n", $sum, 5300 $row->[0], $row->[4]; 5301 ++$fail_count; 5302 } else { 5303 ++$pass_count; 5304 } 5305 } 5306 printf STDERR "AddressInc 32-bit tests: %d passes, %d failures\n", 5307 $pass_count, $fail_count; 5308 $error_count = $fail_count; 5309 $fail_count = 0; 5310 $pass_count = 0; 5311 5312 # Now 16-nibble addresses. 5313 $address_length = 16; 5314 foreach my $row (@{$test_data_16}) { 5315 if ($main::opt_debug and $main::opt_test) { print STDERR "@{$row}\n"; } 5316 my $sum = AddressInc (CanonicalHex($row->[0])); 5317 if ($sum ne CanonicalHex($row->[4])) { 5318 printf STDERR "ERROR: %s != %s + 1 = %s\n", $sum, 5319 $row->[0], $row->[4]; 5320 ++$fail_count; 5321 } else { 5322 ++$pass_count; 5323 } 5324 } 5325 printf STDERR "AddressInc 64-bit tests: %d passes, %d failures\n", 5326 $pass_count, $fail_count; 5327 $error_count += $fail_count; 5328 5329 return $error_count; 5330 } 5331 5332 5333 # Driver for unit tests. 5334 # Currently just the address add/subtract/increment routines for 64-bit. 5335 sub RunUnitTests { 5336 my $error_count = 0; 5337 5338 # This is a list of tuples [a, b, a+b, a-b, a+1] 5339 my $unit_test_data_8 = [ 5340 [qw(aaaaaaaa 50505050 fafafafa 5a5a5a5a aaaaaaab)], 5341 [qw(50505050 aaaaaaaa fafafafa a5a5a5a6 50505051)], 5342 [qw(ffffffff aaaaaaaa aaaaaaa9 55555555 00000000)], 5343 [qw(00000001 ffffffff 00000000 00000002 00000002)], 5344 [qw(00000001 fffffff0 fffffff1 00000011 00000002)], 5345 ]; 5346 my $unit_test_data_16 = [ 5347 # The implementation handles data in 7-nibble chunks, so those are the 5348 # interesting boundaries. 5349 [qw(aaaaaaaa 50505050 5350 00_000000f_afafafa 00_0000005_a5a5a5a 00_000000a_aaaaaab)], 5351 [qw(50505050 aaaaaaaa 5352 00_000000f_afafafa ff_ffffffa_5a5a5a6 00_0000005_0505051)], 5353 [qw(ffffffff aaaaaaaa 5354 00_000001a_aaaaaa9 00_0000005_5555555 00_0000010_0000000)], 5355 [qw(00000001 ffffffff 5356 00_0000010_0000000 ff_ffffff0_0000002 00_0000000_0000002)], 5357 [qw(00000001 fffffff0 5358 00_000000f_ffffff1 ff_ffffff0_0000011 00_0000000_0000002)], 5359 5360 [qw(00_a00000a_aaaaaaa 50505050 5361 00_a00000f_afafafa 00_a000005_a5a5a5a 00_a00000a_aaaaaab)], 5362 [qw(0f_fff0005_0505050 aaaaaaaa 5363 0f_fff000f_afafafa 0f_ffefffa_5a5a5a6 0f_fff0005_0505051)], 5364 [qw(00_000000f_fffffff 01_800000a_aaaaaaa 5365 01_800001a_aaaaaa9 fe_8000005_5555555 00_0000010_0000000)], 5366 [qw(00_0000000_0000001 ff_fffffff_fffffff 5367 00_0000000_0000000 00_0000000_0000002 00_0000000_0000002)], 5368 [qw(00_0000000_0000001 ff_fffffff_ffffff0 5369 ff_fffffff_ffffff1 00_0000000_0000011 00_0000000_0000002)], 5370 ]; 5371 5372 $error_count += AddressAddUnitTest($unit_test_data_8, $unit_test_data_16); 5373 $error_count += AddressSubUnitTest($unit_test_data_8, $unit_test_data_16); 5374 $error_count += AddressIncUnitTest($unit_test_data_8, $unit_test_data_16); 5375 if ($error_count > 0) { 5376 print STDERR $error_count, " errors: FAILED\n"; 5377 } else { 5378 print STDERR "PASS\n"; 5379 } 5380 exit ($error_count); 5381 } 5382