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 ".text.sched" 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 .text.sched. 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 ".text.sched" 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 < 6) { 104 print "usage: $P arch objdump objcopy cc ld nm rm mv inputfile\n"; 105 print "version: $V\n"; 106 exit(1); 107} 108 109my ($arch, $bits, $objdump, $objcopy, $cc, 110 $ld, $nm, $rm, $mv, $inputfile) = @ARGV; 111 112# Acceptable sections to record. 113my %text_sections = ( 114 ".text" => 1, 115); 116 117$objdump = "objdump" if ((length $objdump) == 0); 118$objcopy = "objcopy" if ((length $objcopy) == 0); 119$cc = "gcc" if ((length $cc) == 0); 120$ld = "ld" if ((length $ld) == 0); 121$nm = "nm" if ((length $nm) == 0); 122$rm = "rm" if ((length $rm) == 0); 123$mv = "mv" if ((length $mv) == 0); 124 125#print STDERR "running: $P '$arch' '$objdump' '$objcopy' '$cc' '$ld' " . 126# "'$nm' '$rm' '$mv' '$inputfile'\n"; 127 128my %locals; # List of local (static) functions 129my %weak; # List of weak functions 130my %convert; # List of local functions used that needs conversion 131 132my $type; 133my $nm_regex; # Find the local functions (return function) 134my $section_regex; # Find the start of a section 135my $function_regex; # Find the name of a function 136 # (return offset and func name) 137my $mcount_regex; # Find the call site to mcount (return offset) 138my $alignment; # The .align value to use for $mcount_section 139 140if ($arch eq "x86") { 141 if ($bits == 64) { 142 $arch = "x86_64"; 143 } else { 144 $arch = "i386"; 145 } 146} 147 148if ($arch eq "x86_64") { 149 $nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\S+)"; 150 $section_regex = "Disassembly of section\\s+(\\S+):"; 151 $function_regex = "^([0-9a-fA-F]+)\\s+<(.*?)>:"; 152 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\smcount([+-]0x[0-9a-zA-Z]+)?\$"; 153 $type = ".quad"; 154 $alignment = 8; 155 156 # force flags for this arch 157 $ld .= " -m elf_x86_64"; 158 $objdump .= " -M x86-64"; 159 $objcopy .= " -O elf64-x86-64"; 160 $cc .= " -m64"; 161 162} elsif ($arch eq "i386") { 163 $nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\S+)"; 164 $section_regex = "Disassembly of section\\s+(\\S+):"; 165 $function_regex = "^([0-9a-fA-F]+)\\s+<(.*?)>:"; 166 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\smcount\$"; 167 $type = ".long"; 168 $alignment = 4; 169 170 # force flags for this arch 171 $ld .= " -m elf_i386"; 172 $objdump .= " -M i386"; 173 $objcopy .= " -O elf32-i386"; 174 $cc .= " -m32"; 175 176} elsif ($arch eq "sh") { 177 $nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\S+)"; 178 $section_regex = "Disassembly of section\\s+(\\S+):"; 179 $function_regex = "^([0-9a-fA-F]+)\\s+<(.*?)>:"; 180 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\smcount\$"; 181 $type = ".long"; 182 183 # force flags for this arch 184 $ld .= " -m shlelf_linux"; 185 $objcopy .= " -O elf32-sh-linux"; 186 $cc .= " -m32"; 187 188} elsif ($arch eq "powerpc") { 189 $nm_regex = "^[0-9a-fA-F]+\\s+t\\s+(\\.?\\S+)"; 190 $section_regex = "Disassembly of section\\s+(\\S+):"; 191 $function_regex = "^([0-9a-fA-F]+)\\s+<(\\.?.*?)>:"; 192 $mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s\\.?_mcount\$"; 193 if ($bits == 64) { 194 $type = ".quad"; 195 } else { 196 $type = ".long"; 197 } 198 199} else { 200 die "Arch $arch is not supported with CONFIG_FTRACE_MCOUNT_RECORD"; 201} 202 203my $text_found = 0; 204my $read_function = 0; 205my $opened = 0; 206my $mcount_section = "__mcount_loc"; 207 208my $dirname; 209my $filename; 210my $prefix; 211my $ext; 212 213if ($inputfile =~ m,^(.*)/([^/]*)$,) { 214 $dirname = $1; 215 $filename = $2; 216} else { 217 $dirname = "."; 218 $filename = $inputfile; 219} 220 221if ($filename =~ m,^(.*)(\.\S),) { 222 $prefix = $1; 223 $ext = $2; 224} else { 225 $prefix = $filename; 226 $ext = ""; 227} 228 229my $mcount_s = $dirname . "/.tmp_mc_" . $prefix . ".s"; 230my $mcount_o = $dirname . "/.tmp_mc_" . $prefix . ".o"; 231 232# 233# --globalize-symbols came out in 2.17, we must test the version 234# of objcopy, and if it is less than 2.17, then we can not 235# record local functions. 236my $use_locals = 01; 237my $local_warn_once = 0; 238my $found_version = 0; 239 240open (IN, "$objcopy --version |") || die "error running $objcopy"; 241while (<IN>) { 242 if (/objcopy.*\s(\d+)\.(\d+)/) { 243 my $major = $1; 244 my $minor = $2; 245 246 $found_version = 1; 247 if ($major < 2 || 248 ($major == 2 && $minor < 17)) { 249 $use_locals = 0; 250 } 251 last; 252 } 253} 254close (IN); 255 256if (!$found_version) { 257 print STDERR "WARNING: could not find objcopy version.\n" . 258 "\tDisabling local function references.\n"; 259} 260 261 262# 263# Step 1: find all the local (static functions) and weak symbols. 264# 't' is local, 'w/W' is weak (we never use a weak function) 265# 266open (IN, "$nm $inputfile|") || die "error running $nm"; 267while (<IN>) { 268 if (/$nm_regex/) { 269 $locals{$1} = 1; 270 } elsif (/^[0-9a-fA-F]+\s+([wW])\s+(\S+)/) { 271 $weak{$2} = $1; 272 } 273} 274close(IN); 275 276my @offsets; # Array of offsets of mcount callers 277my $ref_func; # reference function to use for offsets 278my $offset = 0; # offset of ref_func to section beginning 279 280## 281# update_funcs - print out the current mcount callers 282# 283# Go through the list of offsets to callers and write them to 284# the output file in a format that can be read by an assembler. 285# 286sub update_funcs 287{ 288 return if ($#offsets < 0); 289 290 defined($ref_func) || die "No function to reference"; 291 292 # A section only had a weak function, to represent it. 293 # Unfortunately, a weak function may be overwritten by another 294 # function of the same name, making all these offsets incorrect. 295 # To be safe, we simply print a warning and bail. 296 if (defined $weak{$ref_func}) { 297 print STDERR 298 "$inputfile: WARNING: referencing weak function" . 299 " $ref_func for mcount\n"; 300 return; 301 } 302 303 # is this function static? If so, note this fact. 304 if (defined $locals{$ref_func}) { 305 306 # only use locals if objcopy supports globalize-symbols 307 if (!$use_locals) { 308 return; 309 } 310 $convert{$ref_func} = 1; 311 } 312 313 # Loop through all the mcount caller offsets and print a reference 314 # to the caller based from the ref_func. 315 for (my $i=0; $i <= $#offsets; $i++) { 316 if (!$opened) { 317 open(FILE, ">$mcount_s") || die "can't create $mcount_s\n"; 318 $opened = 1; 319 print FILE "\t.section $mcount_section,\"a\",\@progbits\n"; 320 print FILE "\t.align $alignment\n" if (defined($alignment)); 321 } 322 printf FILE "\t%s %s + %d\n", $type, $ref_func, $offsets[$i] - $offset; 323 } 324} 325 326# 327# Step 2: find the sections and mcount call sites 328# 329open(IN, "$objdump -dr $inputfile|") || die "error running $objdump"; 330 331my $text; 332 333while (<IN>) { 334 # is it a section? 335 if (/$section_regex/) { 336 337 # Only record text sections that we know are safe 338 if (defined($text_sections{$1})) { 339 $read_function = 1; 340 } else { 341 $read_function = 0; 342 } 343 # print out any recorded offsets 344 update_funcs() if ($text_found); 345 346 # reset all markers and arrays 347 $text_found = 0; 348 undef($ref_func); 349 undef(@offsets); 350 351 # section found, now is this a start of a function? 352 } elsif ($read_function && /$function_regex/) { 353 $text_found = 1; 354 $offset = hex $1; 355 $text = $2; 356 357 # if this is either a local function or a weak function 358 # keep looking for functions that are global that 359 # we can use safely. 360 if (!defined($locals{$text}) && !defined($weak{$text})) { 361 $ref_func = $text; 362 $read_function = 0; 363 } else { 364 # if we already have a function, and this is weak, skip it 365 if (!defined($ref_func) || !defined($weak{$text})) { 366 $ref_func = $text; 367 } 368 } 369 } 370 371 # is this a call site to mcount? If so, record it to print later 372 if ($text_found && /$mcount_regex/) { 373 $offsets[$#offsets + 1] = hex $1; 374 } 375} 376 377# dump out anymore offsets that may have been found 378update_funcs() if ($text_found); 379 380# If we did not find any mcount callers, we are done (do nothing). 381if (!$opened) { 382 exit(0); 383} 384 385close(FILE); 386 387# 388# Step 3: Compile the file that holds the list of call sites to mcount. 389# 390`$cc -o $mcount_o -c $mcount_s`; 391 392my @converts = keys %convert; 393 394# 395# Step 4: Do we have sections that started with local functions? 396# 397if ($#converts >= 0) { 398 my $globallist = ""; 399 my $locallist = ""; 400 401 foreach my $con (@converts) { 402 $globallist .= " --globalize-symbol $con"; 403 $locallist .= " --localize-symbol $con"; 404 } 405 406 my $globalobj = $dirname . "/.tmp_gl_" . $filename; 407 my $globalmix = $dirname . "/.tmp_mx_" . $filename; 408 409 # 410 # Step 5: set up each local function as a global 411 # 412 `$objcopy $globallist $inputfile $globalobj`; 413 414 # 415 # Step 6: Link the global version to our list. 416 # 417 `$ld -r $globalobj $mcount_o -o $globalmix`; 418 419 # 420 # Step 7: Convert the local functions back into local symbols 421 # 422 `$objcopy $locallist $globalmix $inputfile`; 423 424 # Remove the temp files 425 `$rm $globalobj $globalmix`; 426 427} else { 428 429 my $mix = $dirname . "/.tmp_mx_" . $filename; 430 431 # 432 # Step 8: Link the object with our list of call sites object. 433 # 434 `$ld -r $inputfile $mcount_o -o $mix`; 435 436 # 437 # Step 9: Move the result back to the original object. 438 # 439 `$mv $mix $inputfile`; 440} 441 442# Clean up the temp files 443`$rm $mcount_o $mcount_s`; 444 445exit(0); 446