weidendo | a17f2a3 | 2006-03-20 10:27:30 +0000 | [diff] [blame] | 1 | /*--------------------------------------------------------------------*/ |
| 2 | /*--- Callgrind ---*/ |
| 3 | /*--- ct_threads.c ---*/ |
| 4 | /*--------------------------------------------------------------------*/ |
| 5 | |
| 6 | /* |
| 7 | This file is part of Callgrind, a Valgrind tool for call tracing. |
| 8 | |
sewardj | 4505790 | 2006-06-05 23:27:18 +0000 | [diff] [blame] | 9 | Copyright (C) 2002-2006, Josef Weidendorfer (Josef.Weidendorfer@gmx.de) |
weidendo | a17f2a3 | 2006-03-20 10:27:30 +0000 | [diff] [blame] | 10 | |
| 11 | This program is free software; you can redistribute it and/or |
| 12 | modify it under the terms of the GNU General Public License as |
| 13 | published by the Free Software Foundation; either version 2 of the |
| 14 | License, or (at your option) any later version. |
| 15 | |
| 16 | This program is distributed in the hope that it will be useful, but |
| 17 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 18 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 19 | General Public License for more details. |
| 20 | |
| 21 | You should have received a copy of the GNU General Public License |
| 22 | along with this program; if not, write to the Free Software |
| 23 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA |
| 24 | 02111-1307, USA. |
| 25 | |
| 26 | The GNU General Public License is contained in the file COPYING. |
| 27 | */ |
| 28 | |
| 29 | #include "global.h" |
| 30 | |
| 31 | #include <pub_tool_threadstate.h> |
| 32 | |
| 33 | /* forward decls */ |
| 34 | static exec_state* exec_state_save(void); |
| 35 | static exec_state* exec_state_restore(void); |
| 36 | static exec_state* push_exec_state(int); |
| 37 | static exec_state* top_exec_state(void); |
| 38 | |
| 39 | static exec_stack current_states; |
| 40 | |
| 41 | |
| 42 | /*------------------------------------------------------------*/ |
| 43 | /*--- Support for multi-threading ---*/ |
| 44 | /*------------------------------------------------------------*/ |
| 45 | |
| 46 | |
| 47 | /* |
| 48 | * For Valgrind, MT is cooperative (no preemting in our code), |
| 49 | * so we don't need locks... |
| 50 | * |
| 51 | * Per-thread data: |
| 52 | * - BBCCs |
| 53 | * - call stack |
| 54 | * - call hash |
| 55 | * - event counters: last, current |
| 56 | * |
| 57 | * Even when ignoring MT, we need this functions to set up some |
| 58 | * datastructures for the process (= Thread 1). |
| 59 | */ |
| 60 | |
| 61 | /* current running thread */ |
| 62 | ThreadId CLG_(current_tid); |
| 63 | |
| 64 | static thread_info* thread[VG_N_THREADS]; |
| 65 | |
| 66 | thread_info** CLG_(get_threads)() |
| 67 | { |
| 68 | return thread; |
| 69 | } |
| 70 | |
| 71 | thread_info* CLG_(get_current_thread)() |
| 72 | { |
| 73 | return thread[CLG_(current_tid)]; |
| 74 | } |
| 75 | |
| 76 | void CLG_(init_threads)() |
| 77 | { |
| 78 | Int i; |
| 79 | for(i=0;i<VG_N_THREADS;i++) |
| 80 | thread[i] = 0; |
| 81 | CLG_(current_tid) = VG_INVALID_THREADID; |
| 82 | } |
| 83 | |
| 84 | /* switches through all threads and calls func */ |
| 85 | void CLG_(forall_threads)(void (*func)(thread_info*)) |
| 86 | { |
| 87 | Int t, orig_tid = CLG_(current_tid); |
| 88 | |
| 89 | for(t=1;t<VG_N_THREADS;t++) { |
| 90 | if (!thread[t]) continue; |
| 91 | CLG_(switch_thread)(t); |
| 92 | (*func)(thread[t]); |
| 93 | } |
| 94 | CLG_(switch_thread)(orig_tid); |
| 95 | } |
| 96 | |
| 97 | |
| 98 | static |
| 99 | thread_info* new_thread(void) |
| 100 | { |
| 101 | thread_info* t; |
| 102 | |
| 103 | t = (thread_info*) CLG_MALLOC(sizeof(thread_info)); |
| 104 | |
| 105 | /* init state */ |
| 106 | CLG_(init_exec_stack)( &(t->states) ); |
| 107 | CLG_(init_call_stack)( &(t->calls) ); |
| 108 | CLG_(init_fn_stack) ( &(t->fns) ); |
| 109 | /* t->states.entry[0]->cxt = CLG_(get_cxt)(t->fns.bottom); */ |
| 110 | |
| 111 | /* event counters */ |
| 112 | t->lastdump_cost = CLG_(get_eventset_cost)( CLG_(sets).full ); |
| 113 | t->sighandler_cost = CLG_(get_eventset_cost)( CLG_(sets).full ); |
| 114 | CLG_(init_cost)( CLG_(sets).full, t->lastdump_cost ); |
| 115 | CLG_(init_cost)( CLG_(sets).full, t->sighandler_cost ); |
| 116 | |
| 117 | /* init data containers */ |
| 118 | CLG_(init_fn_array)( &(t->fn_active) ); |
| 119 | CLG_(init_bbcc_hash)( &(t->bbccs) ); |
| 120 | CLG_(init_jcc_hash)( &(t->jccs) ); |
| 121 | |
| 122 | return t; |
| 123 | } |
| 124 | |
| 125 | |
| 126 | void CLG_(switch_thread)(ThreadId tid) |
| 127 | { |
| 128 | if (tid == CLG_(current_tid)) return; |
| 129 | |
| 130 | CLG_DEBUG(0, ">> thread %d (was %d)\n", tid, CLG_(current_tid)); |
| 131 | |
| 132 | if (CLG_(current_tid) != VG_INVALID_THREADID) { |
| 133 | /* save thread state */ |
| 134 | thread_info* t = thread[CLG_(current_tid)]; |
| 135 | |
| 136 | CLG_ASSERT(t != 0); |
| 137 | |
| 138 | /* current context (including signal handler contexts) */ |
| 139 | exec_state_save(); |
| 140 | CLG_(copy_current_exec_stack)( &(t->states) ); |
| 141 | CLG_(copy_current_call_stack)( &(t->calls) ); |
| 142 | CLG_(copy_current_fn_stack) ( &(t->fns) ); |
| 143 | |
| 144 | CLG_(copy_current_fn_array) ( &(t->fn_active) ); |
| 145 | /* If we cumulate costs of threads, use TID 1 for all jccs/bccs */ |
| 146 | if (!CLG_(clo).separate_threads) t = thread[1]; |
| 147 | CLG_(copy_current_bbcc_hash)( &(t->bbccs) ); |
| 148 | CLG_(copy_current_jcc_hash) ( &(t->jccs) ); |
| 149 | } |
| 150 | |
| 151 | CLG_(current_tid) = tid; |
| 152 | CLG_ASSERT(tid < VG_N_THREADS); |
| 153 | |
| 154 | if (tid != VG_INVALID_THREADID) { |
| 155 | thread_info* t; |
| 156 | |
| 157 | /* load thread state */ |
| 158 | |
| 159 | if (thread[tid] == 0) thread[tid] = new_thread(); |
| 160 | t = thread[tid]; |
| 161 | |
| 162 | /* current context (including signal handler contexts) */ |
| 163 | CLG_(set_current_exec_stack)( &(t->states) ); |
| 164 | exec_state_restore(); |
| 165 | CLG_(set_current_call_stack)( &(t->calls) ); |
| 166 | CLG_(set_current_fn_stack) ( &(t->fns) ); |
| 167 | |
| 168 | CLG_(set_current_fn_array) ( &(t->fn_active) ); |
| 169 | /* If we cumulate costs of threads, use TID 1 for all jccs/bccs */ |
| 170 | if (!CLG_(clo).separate_threads) t = thread[1]; |
| 171 | CLG_(set_current_bbcc_hash) ( &(t->bbccs) ); |
| 172 | CLG_(set_current_jcc_hash) ( &(t->jccs) ); |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | |
| 177 | void CLG_(run_thread)(ThreadId tid) |
| 178 | { |
| 179 | /* check for dumps needed */ |
| 180 | static ULong bbs_done = 0; |
| 181 | static Char buf[512]; |
| 182 | |
| 183 | if (CLG_(clo).dump_every_bb >0) { |
| 184 | if (CLG_(stat).bb_executions - bbs_done > CLG_(clo).dump_every_bb) { |
weidendo | 9e326b7 | 2006-03-31 13:16:15 +0000 | [diff] [blame] | 185 | VG_(sprintf)(buf, "--dump-every-bb=%llu", CLG_(clo).dump_every_bb); |
weidendo | a17f2a3 | 2006-03-20 10:27:30 +0000 | [diff] [blame] | 186 | CLG_(dump_profile)(buf, False); |
| 187 | bbs_done = CLG_(stat).bb_executions; |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | CLG_(check_command)(); |
| 192 | |
| 193 | /* now check for thread switch */ |
| 194 | CLG_(switch_thread)(tid); |
| 195 | } |
| 196 | |
| 197 | void CLG_(pre_signal)(ThreadId tid, Int sigNum, Bool alt_stack) |
| 198 | { |
| 199 | exec_state *es; |
| 200 | |
| 201 | CLG_DEBUG(0, ">> pre_signal(TID %d, sig %d, alt_st %s)\n", |
| 202 | tid, sigNum, alt_stack ? "yes":"no"); |
| 203 | |
| 204 | /* switch to the thread the handler runs in */ |
| 205 | CLG_(run_thread)(tid); |
| 206 | |
| 207 | /* save current execution state */ |
| 208 | exec_state_save(); |
| 209 | |
| 210 | /* setup current state for a spontaneous call */ |
| 211 | CLG_(init_exec_state)( &CLG_(current_state) ); |
| 212 | CLG_(push_cxt)(0); |
| 213 | |
| 214 | /* setup new cxtinfo struct for this signal handler */ |
| 215 | es = push_exec_state(sigNum); |
| 216 | CLG_(init_cost)( CLG_(sets).full, es->cost); |
| 217 | CLG_(current_state).cost = es->cost; |
| 218 | es->call_stack_bottom = CLG_(current_call_stack).sp; |
| 219 | |
| 220 | CLG_(current_state).sig = sigNum; |
| 221 | } |
| 222 | |
| 223 | /* Run post-signal if the stackpointer for call stack is at |
| 224 | * the bottom in current exec state (e.g. a signal handler) |
| 225 | * |
| 226 | * Called from CLG_(pop_call_stack) |
| 227 | */ |
| 228 | void CLG_(run_post_signal_on_call_stack_bottom)() |
| 229 | { |
| 230 | exec_state* es = top_exec_state(); |
| 231 | CLG_ASSERT(es != 0); |
| 232 | CLG_ASSERT(CLG_(current_state).sig >0); |
| 233 | |
| 234 | if (CLG_(current_call_stack).sp == es->call_stack_bottom) |
| 235 | CLG_(post_signal)( CLG_(current_tid), CLG_(current_state).sig ); |
| 236 | } |
| 237 | |
| 238 | void CLG_(post_signal)(ThreadId tid, Int sigNum) |
| 239 | { |
| 240 | exec_state* es; |
| 241 | UInt fn_number, *pactive; |
| 242 | |
| 243 | CLG_DEBUG(0, ">> post_signal(TID %d, sig %d)\n", |
| 244 | tid, sigNum); |
| 245 | |
| 246 | CLG_ASSERT(tid == CLG_(current_tid)); |
| 247 | CLG_ASSERT(sigNum == CLG_(current_state).sig); |
| 248 | |
| 249 | /* Unwind call stack of this signal handler. |
| 250 | * This should only be needed at finalisation time |
| 251 | */ |
| 252 | es = top_exec_state(); |
| 253 | CLG_ASSERT(es != 0); |
| 254 | while(CLG_(current_call_stack).sp > es->call_stack_bottom) |
| 255 | CLG_(pop_call_stack)(); |
| 256 | |
| 257 | if (CLG_(current_state).cxt) { |
| 258 | /* correct active counts */ |
| 259 | fn_number = CLG_(current_state).cxt->fn[0]->number; |
| 260 | pactive = CLG_(get_fn_entry)(fn_number); |
| 261 | (*pactive)--; |
| 262 | CLG_DEBUG(0, " set active count of %s back to %d\n", |
| 263 | CLG_(current_state).cxt->fn[0]->name, *pactive); |
| 264 | } |
| 265 | |
| 266 | if (CLG_(current_fn_stack).top > CLG_(current_fn_stack).bottom) { |
| 267 | /* set fn_stack_top back. |
| 268 | * top can point to 0 if nothing was executed in the signal handler; |
| 269 | * this is possible at end on unwinding handlers. |
| 270 | */ |
| 271 | if (*(CLG_(current_fn_stack).top) != 0) { |
| 272 | CLG_(current_fn_stack).top--; |
| 273 | CLG_ASSERT(*(CLG_(current_fn_stack).top) == 0); |
| 274 | } |
| 275 | if (CLG_(current_fn_stack).top > CLG_(current_fn_stack).bottom) |
| 276 | CLG_(current_fn_stack).top--; |
| 277 | } |
| 278 | |
| 279 | /* sum up costs */ |
| 280 | CLG_ASSERT(CLG_(current_state).cost == es->cost); |
| 281 | CLG_(add_and_zero_cost)( CLG_(sets).full, |
| 282 | thread[CLG_(current_tid)]->sighandler_cost, |
| 283 | CLG_(current_state).cost ); |
| 284 | |
| 285 | /* restore previous context */ |
| 286 | es->sig = -1; |
| 287 | current_states.sp--; |
| 288 | es = top_exec_state(); |
| 289 | CLG_(current_state).sig = es->sig; |
| 290 | exec_state_restore(); |
| 291 | |
| 292 | /* There is no way to reliable get the thread ID we are switching to |
| 293 | * after this handler returns. So we sync with actual TID at start of |
| 294 | * CLG_(setup_bb)(), which should be the next for callgrind. |
| 295 | */ |
| 296 | } |
| 297 | |
| 298 | |
| 299 | |
| 300 | /*------------------------------------------------------------*/ |
| 301 | /*--- Execution states in a thread & signal handlers ---*/ |
| 302 | /*------------------------------------------------------------*/ |
| 303 | |
| 304 | /* Each thread can be interrupted by a signal handler, and they |
| 305 | * themselves again. But as there's no scheduling among handlers |
| 306 | * of the same thread, we don't need additional stacks. |
| 307 | * So storing execution contexts and |
| 308 | * adding separators in the callstack(needed to not intermix normal/handler |
| 309 | * functions in contexts) should be enough. |
| 310 | */ |
| 311 | |
| 312 | /* not initialized: call_stack_bottom, sig */ |
| 313 | void CLG_(init_exec_state)(exec_state* es) |
| 314 | { |
| 315 | es->collect = CLG_(clo).collect_atstart; |
| 316 | es->cxt = 0; |
| 317 | es->jmps_passed = 0; |
| 318 | es->bbcc = 0; |
| 319 | es->nonskipped = 0; |
| 320 | } |
| 321 | |
| 322 | |
| 323 | static exec_state* new_exec_state(Int sigNum) |
| 324 | { |
| 325 | exec_state* es; |
| 326 | es = (exec_state*) CLG_MALLOC(sizeof(exec_state)); |
| 327 | |
| 328 | /* allocate real cost space: needed as incremented by |
| 329 | * simulation functions */ |
| 330 | es->cost = CLG_(get_eventset_cost)(CLG_(sets).full); |
| 331 | CLG_(init_cost)( CLG_(sets).full, es->cost ); |
| 332 | |
| 333 | CLG_(init_exec_state)(es); |
| 334 | es->sig = sigNum; |
| 335 | es->call_stack_bottom = 0; |
| 336 | |
| 337 | return es; |
| 338 | } |
| 339 | |
| 340 | void CLG_(init_exec_stack)(exec_stack* es) |
| 341 | { |
| 342 | Int i; |
| 343 | |
| 344 | /* The first element is for the main thread */ |
| 345 | es->entry[0] = new_exec_state(0); |
| 346 | for(i=1;i<MAX_SIGHANDLERS;i++) |
| 347 | es->entry[i] = 0; |
| 348 | es->sp = 0; |
| 349 | } |
| 350 | |
| 351 | void CLG_(copy_current_exec_stack)(exec_stack* dst) |
| 352 | { |
| 353 | Int i; |
| 354 | |
| 355 | dst->sp = current_states.sp; |
| 356 | for(i=0;i<MAX_SIGHANDLERS;i++) |
| 357 | dst->entry[i] = current_states.entry[i]; |
| 358 | } |
| 359 | |
| 360 | void CLG_(set_current_exec_stack)(exec_stack* dst) |
| 361 | { |
| 362 | Int i; |
| 363 | |
| 364 | current_states.sp = dst->sp; |
| 365 | for(i=0;i<MAX_SIGHANDLERS;i++) |
| 366 | current_states.entry[i] = dst->entry[i]; |
| 367 | } |
| 368 | |
| 369 | |
| 370 | /* Get top context info struct of current thread */ |
| 371 | static |
| 372 | exec_state* top_exec_state(void) |
| 373 | { |
| 374 | Int sp = current_states.sp; |
| 375 | exec_state* es; |
| 376 | |
| 377 | CLG_ASSERT((sp >= 0) && (sp < MAX_SIGHANDLERS)); |
| 378 | es = current_states.entry[sp]; |
| 379 | CLG_ASSERT(es != 0); |
| 380 | return es; |
| 381 | } |
| 382 | |
| 383 | /* Allocates a free context info structure for a new entered |
| 384 | * signal handler, putting it on the context stack. |
| 385 | * Returns a pointer to the structure. |
| 386 | */ |
| 387 | static exec_state* push_exec_state(int sigNum) |
| 388 | { |
| 389 | Int sp; |
| 390 | exec_state* es; |
| 391 | |
| 392 | current_states.sp++; |
| 393 | sp = current_states.sp; |
| 394 | |
| 395 | CLG_ASSERT((sigNum > 0) && (sigNum <= _VKI_NSIG)); |
| 396 | CLG_ASSERT((sp > 0) && (sp < MAX_SIGHANDLERS)); |
| 397 | es = current_states.entry[sp]; |
| 398 | if (!es) { |
| 399 | es = new_exec_state(sigNum); |
| 400 | current_states.entry[sp] = es; |
| 401 | } |
| 402 | else |
| 403 | es->sig = sigNum; |
| 404 | |
| 405 | return es; |
| 406 | } |
| 407 | |
| 408 | /* Save current context to top cxtinfo struct */ |
| 409 | static |
| 410 | exec_state* exec_state_save(void) |
| 411 | { |
| 412 | exec_state* es = top_exec_state(); |
| 413 | |
| 414 | es->cxt = CLG_(current_state).cxt; |
| 415 | es->collect = CLG_(current_state).collect; |
| 416 | es->jmps_passed = CLG_(current_state).jmps_passed; |
| 417 | es->bbcc = CLG_(current_state).bbcc; |
| 418 | es->nonskipped = CLG_(current_state).nonskipped; |
| 419 | |
| 420 | CLG_DEBUGIF(1) { |
| 421 | CLG_DEBUG(1, " cxtinfo_save(sig %d): collect %s, jmps_passed %d\n", |
| 422 | es->sig, es->collect ? "Yes": "No", es->jmps_passed); |
| 423 | CLG_(print_bbcc)(-9, es->bbcc, False); |
| 424 | CLG_(print_cost)(-9, CLG_(sets).full, es->cost); |
| 425 | } |
| 426 | |
| 427 | /* signal number does not need to be saved */ |
| 428 | CLG_ASSERT(CLG_(current_state).sig == es->sig); |
| 429 | |
| 430 | return es; |
| 431 | } |
| 432 | |
| 433 | static |
| 434 | exec_state* exec_state_restore(void) |
| 435 | { |
| 436 | exec_state* es = top_exec_state(); |
| 437 | |
| 438 | CLG_(current_state).cxt = es->cxt; |
| 439 | CLG_(current_state).collect = es->collect; |
| 440 | CLG_(current_state).jmps_passed = es->jmps_passed; |
| 441 | CLG_(current_state).bbcc = es->bbcc; |
| 442 | CLG_(current_state).nonskipped = es->nonskipped; |
| 443 | CLG_(current_state).cost = es->cost; |
| 444 | CLG_(current_state).sig = es->sig; |
| 445 | |
| 446 | CLG_DEBUGIF(1) { |
| 447 | CLG_DEBUG(1, " exec_state_restore(sig %d): collect %s, jmps_passed %d\n", |
| 448 | es->sig, es->collect ? "Yes": "No", es->jmps_passed); |
| 449 | CLG_(print_bbcc)(-9, es->bbcc, False); |
| 450 | CLG_(print_cxt)(-9, es->cxt, 0); |
| 451 | CLG_(print_cost)(-9, CLG_(sets).full, es->cost); |
| 452 | } |
| 453 | |
| 454 | return es; |
| 455 | } |
| 456 | |