1#!/usr/bin/perl -w 2# (c) 2008, Steven Rostedt <srostedt@redhat.com> 3# Licensed under the terms of the GNU GPL License version 2 4# 5# recordmcount.pl - makes a section called __mcount_loc that holds 6# all the offsets to the calls to mcount. 7# 8# 9# What we want to end up with is a section in vmlinux called 10# __mcount_loc that contains a list of pointers to all the 11# call sites in the kernel that call mcount. Later on boot up, the kernel 12# will read this list, save the locations and turn them into nops. 13# When tracing or profiling is later enabled, these locations will then 14# be converted back to pointers to some function. 15# 16# This is no easy feat. This script is called just after the original 17# object is compiled and before it is linked. 18# 19# The references to the call sites are offsets from the section of text 20# that the call site is in. Hence, all functions in a section that 21# has a call site to mcount, will have the offset from the beginning of 22# the section and not the beginning of the function. 23# 24# The trick is to find a way to record the beginning of the section. 25# The way we do this is to look at the first function in the section 26# which will also be the location of that section after final link. 27# e.g. 28# 29# .section ".sched.text", "ax" 30# .globl my_func 31# my_func: 32# [...] 33# call mcount (offset: 0x5) 34# [...] 35# ret 36# other_func: 37# [...] 38# call mcount (offset: 0x1b) 39# [...] 40# 41# Both relocation offsets for the mcounts in the above example will be 42# offset from .sched.text. If we make another file called tmp.s with: 43# 44# .section __mcount_loc 45# .quad my_func + 0x5 46# .quad my_func + 0x1b 47# 48# We can then compile this tmp.s into tmp.o, and link it to the original 49# object. 50# 51# But this gets hard if my_func is not globl (a static function). 52# In such a case we have: 53# 54# .section ".sched.text", "ax" 55# my_func: 56# [...] 57# call mcount (offset: 0x5) 58# [...] 59# ret 60# .globl my_func 61# other_func: 62# [...] 63# call mcount (offset: 0x1b) 64# [...] 65# 66# If we make the tmp.s the same as above, when we link together with 67# the original object, we will end up with two symbols for my_func: 68# one local, one global. After final compile, we will end up with 69# an undefined reference to my_func. 70# 71# Since local objects can reference local variables, we need to find 72# a way to make tmp.o reference the local objects of the original object 73# file after it is linked together. To do this, we convert the my_func 74# into a global symbol before linking tmp.o. Then after we link tmp.o 75# we will only have a single symbol for my_func that is global. 76# We can convert my_func back into a local symbol and we are done. 77# 78# Here are the steps we take: 79# 80# 1) Record all the local symbols by using 'nm' 81# 2) Use objdump to find all the call site offsets and sections for 82# mcount. 83# 3) Compile the list into its own object. 84# 4) Do we have to deal with local functions? If not, go to step 8. 85# 5) Make an object that converts these local functions to global symbols 86# with objcopy. 87# 6) Link together this new object with the list object. 88# 7) Convert the local functions back to local symbols and rename 89# the result as the original object. 90# End. 91# 8) Link the object with the list object. 92# 9) Move the result back to the original object. 93# End. 94# 95 96use strict; 97 98my $P = $0; 99$P =~ s@.*/@@g; 100 101my $V = '0.1'; 102 103if ($#ARGV < 7) { 104 print "usage: $P arch bits objdump objcopy cc ld nm rm mv is_module inputfile\n"; 105 print "version: $V\n"; 106 exit(1); 107} 108 109my ($arch, $bits, $objdump, $objcopy, $cc, 110 $ld, $nm, $rm, $mv, $is_module, $inputfile) = @ARGV; 111 112# This file refers to mcount and shouldn't be ftraced, so lets' ignore it 113if ($inputfile eq "kernel/trace/ftrace.o") { 114 exit(0); 115} 116 117# Acceptable sections to record. 118my %text_sections = ( 119 ".text" => 1, 120 ".sched.text" => 1, 121 ".spinlock.text" => 1, 122 ".irqentry.text" => 1, 123); 124 125$objdump = "objdump" if ((length $objdump) == 0); 126$objcopy = "objcopy" if ((length $objcopy) == 0); 127$cc = "gcc" if ((length $cc) == 0); 128$ld = "ld" if ((length $ld) == 0); 129$nm = "nm" if ((length $nm) == 0); 130$rm = "rm" if ((length $rm) == 0); 131$mv = "mv" if ((length $mv) == 0); 132 133#print STDERR "running: $P '$arch' '$objdump' '$objcopy' '$cc' '$ld' " . 134# "'$nm' '$rm' '$mv' '$inputfile'\n"; 135 136my %locals; # List of local (static) functions 137my %weak; # List of weak functions 138my %convert; # List of local functions used that needs conversion 139 140my $type; 141my $nm_regex; # Find the local functions (return function) 142my $section_regex; # Find the start of a section 143my $function_regex; # Find the name of a function 144 # (return offset and func name) 145my $mcount_regex; # Find the call site to mcount (return offset) 146my $alignment; # The .align value to use for $mcount_section 147my $section_type; # Section header plus possible alignment command 148 149if ($arch eq "x86") { 150 if ($bits == 64) { 151 $arch = "x86_64"; 152 } else { 153 $arch = "i386"; 154 } 155} 156 157# 158# We base the defaults off of i386, the other archs may 159# feel free to change them in the below if statements. 160# 161$nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\S+)"; 162$section_regex = "Disassembly of section\\s+(\\S+):"; 163$function_regex = "^([0-9a-fA-F]+)\\s+<(.*?)>:"; 164$mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\smcount\$"; 165$section_type = '@progbits'; 166$type = ".long"; 167 168if ($arch eq "x86_64") { 169 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\smcount([+-]0x[0-9a-zA-Z]+)?\$"; 170 $type = ".quad"; 171 $alignment = 8; 172 173 # force flags for this arch 174 $ld .= " -m elf_x86_64"; 175 $objdump .= " -M x86-64"; 176 $objcopy .= " -O elf64-x86-64"; 177 $cc .= " -m64"; 178 179} elsif ($arch eq "i386") { 180 $alignment = 4; 181 182 # force flags for this arch 183 $ld .= " -m elf_i386"; 184 $objdump .= " -M i386"; 185 $objcopy .= " -O elf32-i386"; 186 $cc .= " -m32"; 187 188} elsif ($arch eq "s390" && $bits == 32) { 189 $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_390_32\\s+_mcount\$"; 190 $alignment = 4; 191 $ld .= " -m elf_s390"; 192 $cc .= " -m31"; 193 194} elsif ($arch eq "s390" && $bits == 64) { 195 $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_390_(PC|PLT)32DBL\\s+_mcount\\+0x2\$"; 196 $alignment = 8; 197 $type = ".quad"; 198 $ld .= " -m elf64_s390"; 199 $cc .= " -m64"; 200 201} elsif ($arch eq "sh") { 202 $alignment = 2; 203 204 # force flags for this arch 205 $ld .= " -m shlelf_linux"; 206 $objcopy .= " -O elf32-sh-linux"; 207 $cc .= " -m32"; 208 209} elsif ($arch eq "powerpc") { 210 $nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\.?\\S+)"; 211 $function_regex = "^([0-9a-fA-F]+)\\s+<(\\.?.*?)>:"; 212 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s\\.?_mcount\$"; 213 214 if ($bits == 64) { 215 $type = ".quad"; 216 } 217 218} elsif ($arch eq "arm") { 219 $alignment = 2; 220 $section_type = '%progbits'; 221 222} elsif ($arch eq "ia64") { 223 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s_mcount\$"; 224 $type = "data8"; 225 226 if ($is_module eq "0") { 227 $cc .= " -mconstant-gp"; 228 } 229} elsif ($arch eq "sparc64") { 230 # In the objdump output there are giblets like: 231 # 0000000000000000 <igmp_net_exit-0x18>: 232 # As there's some data blobs that get emitted into the 233 # text section before the first instructions and the first 234 # real symbols. We don't want to match that, so to combat 235 # this we use '\w' so we'll match just plain symbol names, 236 # and not those that also include hex offsets inside of the 237 # '<>' brackets. Actually the generic function_regex setting 238 # could safely use this too. 239 $function_regex = "^([0-9a-fA-F]+)\\s+<(\\w*?)>:"; 240 241 # Sparc64 calls '_mcount' instead of plain 'mcount'. 242 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s_mcount\$"; 243 244 $alignment = 8; 245 $type = ".xword"; 246 $ld .= " -m elf64_sparc"; 247 $cc .= " -m64"; 248 $objcopy .= " -O elf64-sparc"; 249} else { 250 die "Arch $arch is not supported with CONFIG_FTRACE_MCOUNT_RECORD"; 251} 252 253my $text_found = 0; 254my $read_function = 0; 255my $opened = 0; 256my $mcount_section = "__mcount_loc"; 257 258my $dirname; 259my $filename; 260my $prefix; 261my $ext; 262 263if ($inputfile =~ m,^(.*)/([^/]*)$,) { 264 $dirname = $1; 265 $filename = $2; 266} else { 267 $dirname = "."; 268 $filename = $inputfile; 269} 270 271if ($filename =~ m,^(.*)(\.\S),) { 272 $prefix = $1; 273 $ext = $2; 274} else { 275 $prefix = $filename; 276 $ext = ""; 277} 278 279my $mcount_s = $dirname . "/.tmp_mc_" . $prefix . ".s"; 280my $mcount_o = $dirname . "/.tmp_mc_" . $prefix . ".o"; 281 282# 283# --globalize-symbols came out in 2.17, we must test the version 284# of objcopy, and if it is less than 2.17, then we can not 285# record local functions. 286my $use_locals = 01; 287my $local_warn_once = 0; 288my $found_version = 0; 289 290open (IN, "$objcopy --version |") || die "error running $objcopy"; 291while (<IN>) { 292 if (/objcopy.*\s(\d+)\.(\d+)/) { 293 my $major = $1; 294 my $minor = $2; 295 296 $found_version = 1; 297 if ($major < 2 || 298 ($major == 2 && $minor < 17)) { 299 $use_locals = 0; 300 } 301 last; 302 } 303} 304close (IN); 305 306if (!$found_version) { 307 print STDERR "WARNING: could not find objcopy version.\n" . 308 "\tDisabling local function references.\n"; 309} 310 311# 312# Step 1: find all the local (static functions) and weak symbols. 313# 't' is local, 'w/W' is weak (we never use a weak function) 314# 315open (IN, "$nm $inputfile|") || die "error running $nm"; 316while (<IN>) { 317 if (/$nm_regex/) { 318 $locals{$1} = 1; 319 } elsif (/^[0-9a-fA-F]+\s+([wW])\s+(\S+)/) { 320 $weak{$2} = $1; 321 } 322} 323close(IN); 324 325my @offsets; # Array of offsets of mcount callers 326my $ref_func; # reference function to use for offsets 327my $offset = 0; # offset of ref_func to section beginning 328 329## 330# update_funcs - print out the current mcount callers 331# 332# Go through the list of offsets to callers and write them to 333# the output file in a format that can be read by an assembler. 334# 335sub update_funcs 336{ 337 return if ($#offsets < 0); 338 339 defined($ref_func) || die "No function to reference"; 340 341 # A section only had a weak function, to represent it. 342 # Unfortunately, a weak function may be overwritten by another 343 # function of the same name, making all these offsets incorrect. 344 # To be safe, we simply print a warning and bail. 345 if (defined $weak{$ref_func}) { 346 print STDERR 347 "$inputfile: WARNING: referencing weak function" . 348 " $ref_func for mcount\n"; 349 return; 350 } 351 352 # is this function static? If so, note this fact. 353 if (defined $locals{$ref_func}) { 354 355 # only use locals if objcopy supports globalize-symbols 356 if (!$use_locals) { 357 return; 358 } 359 $convert{$ref_func} = 1; 360 } 361 362 # Loop through all the mcount caller offsets and print a reference 363 # to the caller based from the ref_func. 364 for (my $i=0; $i <= $#offsets; $i++) { 365 if (!$opened) { 366 open(FILE, ">$mcount_s") || die "can't create $mcount_s\n"; 367 $opened = 1; 368 print FILE "\t.section $mcount_section,\"a\",$section_type\n"; 369 print FILE "\t.align $alignment\n" if (defined($alignment)); 370 } 371 printf FILE "\t%s %s + %d\n", $type, $ref_func, $offsets[$i] - $offset; 372 } 373} 374 375# 376# Step 2: find the sections and mcount call sites 377# 378open(IN, "$objdump -hdr $inputfile|") || die "error running $objdump"; 379 380my $text; 381 382my $read_headers = 1; 383 384while (<IN>) { 385 # is it a section? 386 if (/$section_regex/) { 387 $read_headers = 0; 388 389 # Only record text sections that we know are safe 390 if (defined($text_sections{$1})) { 391 $read_function = 1; 392 } else { 393 $read_function = 0; 394 } 395 # print out any recorded offsets 396 update_funcs() if (defined($ref_func)); 397 398 # reset all markers and arrays 399 $text_found = 0; 400 undef($ref_func); 401 undef(@offsets); 402 403 # section found, now is this a start of a function? 404 } elsif ($read_function && /$function_regex/) { 405 $text_found = 1; 406 $text = $2; 407 408 # if this is either a local function or a weak function 409 # keep looking for functions that are global that 410 # we can use safely. 411 if (!defined($locals{$text}) && !defined($weak{$text})) { 412 $ref_func = $text; 413 $read_function = 0; 414 $offset = hex $1; 415 } else { 416 # if we already have a function, and this is weak, skip it 417 if (!defined($ref_func) && !defined($weak{$text}) && 418 # PPC64 can have symbols that start with .L and 419 # gcc considers these special. Don't use them! 420 $text !~ /^\.L/) { 421 $ref_func = $text; 422 $offset = hex $1; 423 } 424 } 425 } elsif ($read_headers && /$mcount_section/) { 426 # 427 # Somehow the make process can execute this script on an 428 # object twice. If it does, we would duplicate the mcount 429 # section and it will cause the function tracer self test 430 # to fail. Check if the mcount section exists, and if it does, 431 # warn and exit. 432 # 433 print STDERR "ERROR: $mcount_section already in $inputfile\n" . 434 "\tThis may be an indication that your build is corrupted.\n" . 435 "\tDelete $inputfile and try again. If the same object file\n" . 436 "\tstill causes an issue, then disable CONFIG_DYNAMIC_FTRACE.\n"; 437 exit(-1); 438 } 439 440 # is this a call site to mcount? If so, record it to print later 441 if ($text_found && /$mcount_regex/) { 442 $offsets[$#offsets + 1] = hex $1; 443 } 444} 445 446# dump out anymore offsets that may have been found 447update_funcs() if (defined($ref_func)); 448 449# If we did not find any mcount callers, we are done (do nothing). 450if (!$opened) { 451 exit(0); 452} 453 454close(FILE); 455 456# 457# Step 3: Compile the file that holds the list of call sites to mcount. 458# 459`$cc -o $mcount_o -c $mcount_s`; 460 461my @converts = keys %convert; 462 463# 464# Step 4: Do we have sections that started with local functions? 465# 466if ($#converts >= 0) { 467 my $globallist = ""; 468 my $locallist = ""; 469 470 foreach my $con (@converts) { 471 $globallist .= " --globalize-symbol $con"; 472 $locallist .= " --localize-symbol $con"; 473 } 474 475 my $globalobj = $dirname . "/.tmp_gl_" . $filename; 476 my $globalmix = $dirname . "/.tmp_mx_" . $filename; 477 478 # 479 # Step 5: set up each local function as a global 480 # 481 `$objcopy $globallist $inputfile $globalobj`; 482 483 # 484 # Step 6: Link the global version to our list. 485 # 486 `$ld -r $globalobj $mcount_o -o $globalmix`; 487 488 # 489 # Step 7: Convert the local functions back into local symbols 490 # 491 `$objcopy $locallist $globalmix $inputfile`; 492 493 # Remove the temp files 494 `$rm $globalobj $globalmix`; 495 496} else { 497 498 my $mix = $dirname . "/.tmp_mx_" . $filename; 499 500 # 501 # Step 8: Link the object with our list of call sites object. 502 # 503 `$ld -r $inputfile $mcount_o -o $mix`; 504 505 # 506 # Step 9: Move the result back to the original object. 507 # 508 `$mv $mix $inputfile`; 509} 510 511# Clean up the temp files 512`$rm $mcount_o $mcount_s`; 513 514exit(0); 515