Mathias Agopian | 7c0c379 | 2011-09-05 23:54:55 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2011 The Android Open Source Project |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * Redistribution and use in source and binary forms, with or without |
| 6 | * modification, are permitted provided that the following conditions |
| 7 | * are met: |
| 8 | * * Redistributions of source code must retain the above copyright |
| 9 | * notice, this list of conditions and the following disclaimer. |
| 10 | * * Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in |
| 12 | * the documentation and/or other materials provided with the |
| 13 | * distribution. |
| 14 | * |
| 15 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 16 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 17 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 18 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 19 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 20 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 21 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS |
| 22 | * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 23 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 24 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 25 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 26 | * SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | #include <sys/types.h> |
| 30 | #include <sys/atomics.h> |
| 31 | #include <sys/system_properties.h> |
| 32 | #include <sys/mman.h> |
| 33 | |
| 34 | #if HAVE_DLADDR |
| 35 | #include <dlfcn.h> |
| 36 | #endif |
| 37 | #include <stdint.h> |
| 38 | #include <stdio.h> |
| 39 | #include <stdlib.h> |
| 40 | #include <errno.h> |
| 41 | #include <pthread.h> |
| 42 | #include <unwind.h> |
| 43 | #include <unistd.h> |
| 44 | |
| 45 | #include "logd.h" |
| 46 | #include "bionic_tls.h" |
| 47 | |
| 48 | /* |
| 49 | * =========================================================================== |
| 50 | * Deadlock prediction |
| 51 | * =========================================================================== |
| 52 | */ |
| 53 | /* |
| 54 | The idea is to predict the possibility of deadlock by recording the order |
| 55 | in which locks are acquired. If we see an attempt to acquire a lock |
| 56 | out of order, we can identify the locks and offending code. |
| 57 | |
| 58 | To make this work, we need to keep track of the locks held by each thread, |
| 59 | and create history trees for each lock. When a thread tries to acquire |
| 60 | a new lock, we walk through the "history children" of the lock, looking |
| 61 | for a match with locks the thread already holds. If we find a match, |
| 62 | it means the thread has made a request that could result in a deadlock. |
| 63 | |
| 64 | To support recursive locks, we always allow re-locking a currently-held |
| 65 | lock, and maintain a recursion depth count. |
| 66 | |
| 67 | An ASCII-art example, where letters represent locks: |
| 68 | |
| 69 | A |
| 70 | /|\ |
| 71 | / | \ |
| 72 | B | D |
| 73 | \ | |
| 74 | \| |
| 75 | C |
| 76 | |
| 77 | The above is the tree we'd have after handling lock synchronization |
| 78 | sequences "ABC", "AC", "AD". A has three children, {B, C, D}. C is also |
| 79 | a child of B. (The lines represent pointers between parent and child. |
| 80 | Every node can have multiple parents and multiple children.) |
| 81 | |
| 82 | If we hold AC, and want to lock B, we recursively search through B's |
| 83 | children to see if A or C appears. It does, so we reject the attempt. |
| 84 | (A straightforward way to implement it: add a link from C to B, then |
| 85 | determine whether the graph starting at B contains a cycle.) |
| 86 | |
| 87 | If we hold AC and want to lock D, we would succeed, creating a new link |
| 88 | from C to D. |
| 89 | |
| 90 | Updates to MutexInfo structs are only allowed for the thread that holds |
| 91 | the lock, so we actually do most of our deadlock prediction work after |
| 92 | the lock has been acquired. |
| 93 | */ |
| 94 | |
| 95 | // ============================================================================= |
| 96 | // log functions |
| 97 | // ============================================================================= |
| 98 | |
| 99 | #define LOGD(format, ...) \ |
| 100 | __libc_android_log_print(ANDROID_LOG_DEBUG, \ |
| 101 | "pthread_debug", (format), ##__VA_ARGS__ ) |
| 102 | |
| 103 | #define LOGW(format, ...) \ |
| 104 | __libc_android_log_print(ANDROID_LOG_WARN, \ |
| 105 | "pthread_debug", (format), ##__VA_ARGS__ ) |
| 106 | |
| 107 | #define LOGE(format, ...) \ |
| 108 | __libc_android_log_print(ANDROID_LOG_ERROR, \ |
| 109 | "pthread_debug", (format), ##__VA_ARGS__ ) |
| 110 | |
| 111 | #define LOGI(format, ...) \ |
| 112 | __libc_android_log_print(ANDROID_LOG_INFO, \ |
| 113 | "pthread_debug", (format), ##__VA_ARGS__ ) |
| 114 | |
| 115 | static const char* const kStartBanner = |
| 116 | "==============================================================="; |
| 117 | |
| 118 | static const char* const kEndBanner = |
| 119 | "==============================================================="; |
| 120 | |
| 121 | extern char* __progname; |
| 122 | |
| 123 | // ============================================================================= |
| 124 | // map info functions |
| 125 | // ============================================================================= |
| 126 | |
| 127 | typedef struct mapinfo { |
| 128 | struct mapinfo *next; |
| 129 | unsigned start; |
| 130 | unsigned end; |
| 131 | char name[]; |
| 132 | } mapinfo; |
| 133 | |
| 134 | static mapinfo* sMapInfo = NULL; |
| 135 | |
| 136 | static mapinfo *parse_maps_line(char *line) |
| 137 | { |
| 138 | mapinfo *mi; |
| 139 | int len = strlen(line); |
| 140 | |
| 141 | if(len < 1) return 0; |
| 142 | line[--len] = 0; |
| 143 | |
| 144 | if(len < 50) return 0; |
| 145 | if(line[20] != 'x') return 0; |
| 146 | |
| 147 | mi = malloc(sizeof(mapinfo) + (len - 47)); |
| 148 | if(mi == 0) return 0; |
| 149 | |
| 150 | mi->start = strtoul(line, 0, 16); |
| 151 | mi->end = strtoul(line + 9, 0, 16); |
| 152 | /* To be filled in parse_elf_info if the mapped section starts with |
| 153 | * elf_header |
| 154 | */ |
| 155 | mi->next = 0; |
| 156 | strcpy(mi->name, line + 49); |
| 157 | |
| 158 | return mi; |
| 159 | } |
| 160 | |
| 161 | static mapinfo *init_mapinfo(int pid) |
| 162 | { |
| 163 | struct mapinfo *milist = NULL; |
| 164 | char data[1024]; |
| 165 | sprintf(data, "/proc/%d/maps", pid); |
| 166 | FILE *fp = fopen(data, "r"); |
| 167 | if(fp) { |
| 168 | while(fgets(data, sizeof(data), fp)) { |
| 169 | mapinfo *mi = parse_maps_line(data); |
| 170 | if(mi) { |
| 171 | mi->next = milist; |
| 172 | milist = mi; |
| 173 | } |
| 174 | } |
| 175 | fclose(fp); |
| 176 | } |
| 177 | |
| 178 | return milist; |
| 179 | } |
| 180 | |
| 181 | static void deinit_mapinfo(mapinfo *mi) |
| 182 | { |
| 183 | mapinfo *del; |
| 184 | while(mi) { |
| 185 | del = mi; |
| 186 | mi = mi->next; |
| 187 | free(del); |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | /* Find the containing map info for the pc */ |
| 192 | static const mapinfo *pc_to_mapinfo(mapinfo *mi, unsigned pc, unsigned *rel_pc) |
| 193 | { |
| 194 | *rel_pc = pc; |
| 195 | while(mi) { |
| 196 | if((pc >= mi->start) && (pc < mi->end)){ |
| 197 | // Only calculate the relative offset for shared libraries |
| 198 | if (strstr(mi->name, ".so")) { |
| 199 | *rel_pc -= mi->start; |
| 200 | } |
| 201 | return mi; |
| 202 | } |
| 203 | mi = mi->next; |
| 204 | } |
| 205 | return NULL; |
| 206 | } |
| 207 | |
| 208 | // ============================================================================= |
| 209 | // stack trace functions |
| 210 | // ============================================================================= |
| 211 | |
| 212 | #define STACK_TRACE_DEPTH 16 |
| 213 | |
| 214 | typedef struct |
| 215 | { |
| 216 | size_t count; |
| 217 | intptr_t* addrs; |
| 218 | } stack_crawl_state_t; |
| 219 | |
| 220 | /* depends how the system includes define this */ |
| 221 | #ifdef HAVE_UNWIND_CONTEXT_STRUCT |
| 222 | typedef struct _Unwind_Context __unwind_context; |
| 223 | #else |
| 224 | typedef _Unwind_Context __unwind_context; |
| 225 | #endif |
| 226 | |
| 227 | static _Unwind_Reason_Code trace_function(__unwind_context *context, void *arg) |
| 228 | { |
| 229 | stack_crawl_state_t* state = (stack_crawl_state_t*)arg; |
| 230 | if (state->count) { |
| 231 | intptr_t ip = (intptr_t)_Unwind_GetIP(context); |
| 232 | if (ip) { |
| 233 | state->addrs[0] = ip; |
| 234 | state->addrs++; |
| 235 | state->count--; |
| 236 | return _URC_NO_REASON; |
| 237 | } |
| 238 | } |
| 239 | /* |
| 240 | * If we run out of space to record the address or 0 has been seen, stop |
| 241 | * unwinding the stack. |
| 242 | */ |
| 243 | return _URC_END_OF_STACK; |
| 244 | } |
| 245 | |
| 246 | static inline |
| 247 | int get_backtrace(intptr_t* addrs, size_t max_entries) |
| 248 | { |
| 249 | stack_crawl_state_t state; |
| 250 | state.count = max_entries; |
| 251 | state.addrs = (intptr_t*)addrs; |
| 252 | _Unwind_Backtrace(trace_function, (void*)&state); |
| 253 | return max_entries - state.count; |
| 254 | } |
| 255 | |
| 256 | static void log_backtrace(intptr_t* addrs, size_t c) |
| 257 | { |
| 258 | int index = 0; |
| 259 | size_t i; |
| 260 | for (i=0 ; i<c; i++) { |
| 261 | unsigned int relpc; |
| 262 | void* offset = 0; |
| 263 | const char* symbol = NULL; |
| 264 | |
| 265 | #if HAVE_DLADDR |
| 266 | Dl_info info; |
| 267 | if (dladdr((void*)addrs[i], &info)) { |
| 268 | offset = info.dli_saddr; |
| 269 | symbol = info.dli_sname; |
| 270 | } |
| 271 | #endif |
| 272 | |
| 273 | if (symbol || index>0 || !HAVE_DLADDR) { |
| 274 | /* |
| 275 | * this test is a bit sketchy, but it allows us to skip the |
| 276 | * stack trace entries due to this debugging code. it works |
| 277 | * because those don't have a symbol (they're not exported) |
| 278 | */ |
| 279 | mapinfo const* mi = pc_to_mapinfo(sMapInfo, addrs[i], &relpc); |
| 280 | char const* soname = mi ? mi->name : NULL; |
| 281 | #if HAVE_DLADDR |
| 282 | if (!soname) |
| 283 | soname = info.dli_fname; |
| 284 | #endif |
| 285 | if (!soname) |
| 286 | soname = "unknown"; |
| 287 | |
| 288 | if (symbol) { |
| 289 | LOGW(" " |
| 290 | "#%02d pc %08lx %s (%s+0x%x)", |
| 291 | index, relpc, soname, symbol, |
| 292 | addrs[i] - (intptr_t)offset); |
| 293 | } else { |
| 294 | LOGW(" " |
| 295 | "#%02d pc %08lx %s", |
| 296 | index, relpc, soname); |
| 297 | } |
| 298 | index++; |
| 299 | } |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | /****************************************************************************/ |
| 304 | |
| 305 | /* |
| 306 | * level <= 0 : deadlock prediction disabled |
| 307 | * level 1 : deadlock prediction enabled, w/o call stacks |
| 308 | * level 2 : deadlock prediction enabled w/ call stacks |
| 309 | */ |
| 310 | #define CAPTURE_CALLSTACK 2 |
| 311 | static int sPthreadDebugLevel = 0; |
| 312 | static pid_t sPthreadDebugDisabledThread = -1; |
| 313 | static pthread_mutex_t sDbgLock = PTHREAD_MUTEX_INITIALIZER; |
| 314 | |
| 315 | /****************************************************************************/ |
| 316 | |
| 317 | /* some simple/lame malloc replacement |
| 318 | * NOT thread-safe and leaks everything |
| 319 | */ |
| 320 | |
| 321 | #define DBG_ALLOC_BLOCK_SIZE PAGESIZE |
| 322 | static size_t sDbgAllocOffset = DBG_ALLOC_BLOCK_SIZE; |
| 323 | static char* sDbgAllocPtr = NULL; |
| 324 | |
| 325 | static void* DbgAllocLocked(size_t size) { |
| 326 | if ((sDbgAllocOffset + size) > DBG_ALLOC_BLOCK_SIZE) { |
| 327 | sDbgAllocOffset = 0; |
| 328 | sDbgAllocPtr = mmap(NULL, DBG_ALLOC_BLOCK_SIZE, PROT_READ|PROT_WRITE, |
| 329 | MAP_ANON | MAP_PRIVATE, 0, 0); |
| 330 | if (sDbgAllocPtr == MAP_FAILED) { |
| 331 | return NULL; |
| 332 | } |
| 333 | } |
| 334 | void* addr = sDbgAllocPtr + sDbgAllocOffset; |
| 335 | sDbgAllocOffset += size; |
| 336 | return addr; |
| 337 | } |
| 338 | |
| 339 | static void* debug_realloc(void *ptr, size_t size, size_t old_size) { |
| 340 | void* addr = mmap(NULL, size, PROT_READ|PROT_WRITE, |
| 341 | MAP_ANON | MAP_PRIVATE, 0, 0); |
| 342 | if (addr != MAP_FAILED) { |
| 343 | if (ptr) { |
| 344 | memcpy(addr, ptr, old_size); |
| 345 | munmap(ptr, old_size); |
| 346 | } |
| 347 | } else { |
| 348 | addr = NULL; |
| 349 | } |
| 350 | return addr; |
| 351 | } |
| 352 | |
| 353 | /*****************************************************************************/ |
| 354 | |
| 355 | struct MutexInfo; |
| 356 | |
| 357 | typedef struct CallStack { |
| 358 | intptr_t depth; |
| 359 | intptr_t* addrs; |
| 360 | } CallStack; |
| 361 | |
| 362 | typedef struct MutexInfo* MutexInfoListEntry; |
| 363 | typedef struct CallStack CallStackListEntry; |
| 364 | |
| 365 | typedef struct GrowingList { |
| 366 | int alloc; |
| 367 | int count; |
| 368 | union { |
| 369 | void* data; |
| 370 | MutexInfoListEntry* list; |
| 371 | CallStackListEntry* stack; |
| 372 | }; |
| 373 | } GrowingList; |
| 374 | |
| 375 | typedef GrowingList MutexInfoList; |
| 376 | typedef GrowingList CallStackList; |
| 377 | |
| 378 | typedef struct MutexInfo { |
| 379 | // thread currently holding the lock or 0 |
| 380 | pid_t owner; |
| 381 | |
| 382 | // most-recently-locked doubly-linked list |
| 383 | struct MutexInfo* prev; |
| 384 | struct MutexInfo* next; |
| 385 | |
| 386 | // for reentrant locks |
| 387 | int lockCount; |
| 388 | // when looking for loops in the graph, marks visited nodes |
| 389 | int historyMark; |
| 390 | // the actual mutex |
| 391 | pthread_mutex_t* mutex; |
| 392 | // list of locks directly acquired AFTER this one in the same thread |
| 393 | MutexInfoList children; |
| 394 | // list of locks directly acquired BEFORE this one in the same thread |
| 395 | MutexInfoList parents; |
| 396 | // list of call stacks when a new link is established to this lock form its parent |
| 397 | CallStackList stacks; |
| 398 | // call stack when this lock was acquired last |
| 399 | int stackDepth; |
| 400 | intptr_t stackTrace[STACK_TRACE_DEPTH]; |
| 401 | } MutexInfo; |
| 402 | |
| 403 | static void growingListInit(GrowingList* list) { |
| 404 | list->alloc = 0; |
| 405 | list->count = 0; |
| 406 | list->data = NULL; |
| 407 | } |
| 408 | |
| 409 | static void growingListAdd(GrowingList* pList, size_t objSize) { |
| 410 | if (pList->count == pList->alloc) { |
| 411 | size_t oldsize = pList->alloc * objSize; |
| 412 | pList->alloc += PAGESIZE / objSize; |
| 413 | size_t size = pList->alloc * objSize; |
| 414 | pList->data = debug_realloc(pList->data, size, oldsize); |
| 415 | } |
| 416 | pList->count++; |
| 417 | } |
| 418 | |
| 419 | static void initMutexInfo(MutexInfo* object, pthread_mutex_t* mutex) { |
| 420 | object->owner = 0; |
| 421 | object->prev = 0; |
| 422 | object->next = 0; |
| 423 | object->lockCount = 0; |
| 424 | object->historyMark = 0; |
| 425 | object->mutex = mutex; |
| 426 | growingListInit(&object->children); |
| 427 | growingListInit(&object->parents); |
| 428 | growingListInit(&object->stacks); |
| 429 | object->stackDepth = 0; |
| 430 | } |
| 431 | |
| 432 | typedef struct ThreadInfo { |
| 433 | pid_t pid; |
| 434 | MutexInfo* mrl; |
| 435 | } ThreadInfo; |
| 436 | |
| 437 | static void initThreadInfo(ThreadInfo* object, pid_t pid) { |
| 438 | object->pid = pid; |
| 439 | object->mrl = NULL; |
| 440 | } |
| 441 | |
| 442 | /****************************************************************************/ |
| 443 | |
| 444 | static MutexInfo* get_mutex_info(pthread_mutex_t *mutex); |
| 445 | static void mutex_lock_checked(MutexInfo* mrl, MutexInfo* object); |
| 446 | static void mutex_unlock_checked(MutexInfo* object); |
| 447 | |
| 448 | /****************************************************************************/ |
| 449 | |
| 450 | extern int pthread_mutex_lock_impl(pthread_mutex_t *mutex); |
| 451 | extern int pthread_mutex_unlock_impl(pthread_mutex_t *mutex); |
| 452 | |
| 453 | static int pthread_mutex_lock_unchecked(pthread_mutex_t *mutex) { |
| 454 | return pthread_mutex_lock_impl(mutex); |
| 455 | } |
| 456 | |
| 457 | static int pthread_mutex_unlock_unchecked(pthread_mutex_t *mutex) { |
| 458 | return pthread_mutex_unlock_impl(mutex); |
| 459 | } |
| 460 | |
| 461 | /****************************************************************************/ |
| 462 | |
| 463 | static void dup_backtrace(CallStack* stack, int count, intptr_t const* addrs) { |
| 464 | stack->depth = count; |
| 465 | stack->addrs = DbgAllocLocked(count * sizeof(intptr_t)); |
| 466 | memcpy(stack->addrs, addrs, count * sizeof(intptr_t)); |
| 467 | } |
| 468 | |
| 469 | /****************************************************************************/ |
| 470 | |
| 471 | static int historyListHas( |
| 472 | const MutexInfoList* list, MutexInfo const * obj) { |
| 473 | int i; |
| 474 | for (i=0; i<list->count; i++) { |
| 475 | if (list->list[i] == obj) { |
| 476 | return i; |
| 477 | } |
| 478 | } |
| 479 | return -1; |
| 480 | } |
| 481 | |
| 482 | static void historyListAdd(MutexInfoList* pList, MutexInfo* obj) { |
| 483 | growingListAdd(pList, sizeof(MutexInfoListEntry)); |
| 484 | pList->list[pList->count - 1] = obj; |
| 485 | } |
| 486 | |
| 487 | static int historyListRemove(MutexInfoList* pList, MutexInfo* obj) { |
| 488 | int i; |
| 489 | for (i = pList->count-1; i >= 0; i--) { |
| 490 | if (pList->list[i] == obj) { |
| 491 | break; |
| 492 | } |
| 493 | } |
| 494 | if (i < 0) { |
| 495 | // not found! |
| 496 | return 0; |
| 497 | } |
| 498 | |
| 499 | if (i != pList->count-1) { |
| 500 | // copy the last entry to the new free slot |
| 501 | pList->list[i] = pList->list[pList->count-1]; |
| 502 | } |
| 503 | pList->count--; |
| 504 | memset(&pList->list[pList->count], 0, sizeof(MutexInfoListEntry)); |
| 505 | return 1; |
| 506 | } |
| 507 | |
| 508 | static void linkParentToChild(MutexInfo* parent, MutexInfo* child) { |
| 509 | historyListAdd(&parent->children, child); |
| 510 | historyListAdd(&child->parents, parent); |
| 511 | } |
| 512 | |
| 513 | static void unlinkParentFromChild(MutexInfo* parent, MutexInfo* child) { |
| 514 | historyListRemove(&parent->children, child); |
| 515 | historyListRemove(&child->parents, parent); |
| 516 | } |
| 517 | |
| 518 | /****************************************************************************/ |
| 519 | |
| 520 | static void callstackListAdd(CallStackList* pList, |
| 521 | int count, intptr_t const* addrs) { |
| 522 | growingListAdd(pList, sizeof(CallStackListEntry)); |
| 523 | dup_backtrace(&pList->stack[pList->count - 1], count, addrs); |
| 524 | } |
| 525 | |
| 526 | /****************************************************************************/ |
| 527 | |
| 528 | /* |
| 529 | * Recursively traverse the object hierarchy starting at "obj". We mark |
| 530 | * ourselves on entry and clear the mark on exit. If we ever encounter |
| 531 | * a marked object, we have a cycle. |
| 532 | * |
| 533 | * Returns "true" if all is well, "false" if we found a cycle. |
| 534 | */ |
| 535 | |
| 536 | static int traverseTree(MutexInfo* obj, MutexInfo const* objParent) |
| 537 | { |
| 538 | /* |
| 539 | * Have we been here before? |
| 540 | */ |
| 541 | if (obj->historyMark) { |
| 542 | int stackDepth; |
| 543 | intptr_t addrs[STACK_TRACE_DEPTH]; |
| 544 | |
| 545 | /* Turn off prediction temporarily in this thread while logging */ |
| 546 | sPthreadDebugDisabledThread = gettid(); |
| 547 | |
| 548 | if (sMapInfo == NULL) { |
| 549 | // note: we're protected by sDbgLock |
| 550 | sMapInfo = init_mapinfo(getpid()); |
| 551 | } |
| 552 | |
| 553 | LOGW("%s\n", kStartBanner); |
| 554 | LOGW("pid: %d, tid: %d >>> %s <<<", getpid(), gettid(), __progname); |
| 555 | LOGW("Illegal lock attempt:\n"); |
| 556 | LOGW("--- pthread_mutex_t at %p\n", obj->mutex); |
| 557 | stackDepth = get_backtrace(addrs, STACK_TRACE_DEPTH); |
| 558 | log_backtrace(addrs, stackDepth); |
| 559 | |
| 560 | LOGW("+++ Currently held locks in this thread (in reverse order):"); |
| 561 | MutexInfo* cur = obj; |
| 562 | pid_t ourtid = gettid(); |
| 563 | int i; |
| 564 | for (i=0 ; i<cur->parents.count ; i++) { |
| 565 | MutexInfo* parent = cur->parents.list[i]; |
| 566 | if (parent->owner == ourtid) { |
| 567 | LOGW("--- pthread_mutex_t at %p\n", parent->mutex); |
| 568 | if (sPthreadDebugLevel >= CAPTURE_CALLSTACK) { |
| 569 | log_backtrace(parent->stackTrace, parent->stackDepth); |
| 570 | } |
| 571 | cur = parent; |
| 572 | break; |
| 573 | } |
| 574 | } |
| 575 | |
| 576 | LOGW("+++ Earlier, the following lock order (from last to first) was established\n"); |
| 577 | return 0; |
| 578 | } |
| 579 | |
| 580 | obj->historyMark = 1; |
| 581 | |
| 582 | MutexInfoList* pList = &obj->children; |
| 583 | int result = 1; |
| 584 | int i; |
| 585 | for (i = pList->count-1; i >= 0; i--) { |
| 586 | MutexInfo* child = pList->list[i]; |
| 587 | if (!traverseTree(child, obj)) { |
| 588 | LOGW("--- pthread_mutex_t at %p\n", obj->mutex); |
| 589 | if (sPthreadDebugLevel >= CAPTURE_CALLSTACK) { |
| 590 | int index = historyListHas(&obj->parents, objParent); |
| 591 | if ((size_t)index < (size_t)obj->stacks.count) { |
| 592 | log_backtrace( |
| 593 | obj->stacks.stack[index].addrs, |
| 594 | obj->stacks.stack[index].depth); |
| 595 | } else { |
| 596 | log_backtrace( |
| 597 | obj->stackTrace, |
| 598 | obj->stackDepth); |
| 599 | } |
| 600 | } |
| 601 | result = 0; |
| 602 | break; |
| 603 | } |
| 604 | } |
| 605 | |
| 606 | obj->historyMark = 0; |
| 607 | return result; |
| 608 | } |
| 609 | |
| 610 | /****************************************************************************/ |
| 611 | |
| 612 | static void mutex_lock_checked(MutexInfo* mrl, MutexInfo* object) |
| 613 | { |
| 614 | pid_t tid = gettid(); |
| 615 | if (object->owner == tid) { |
| 616 | object->lockCount++; |
| 617 | return; |
| 618 | } |
| 619 | |
| 620 | object->owner = tid; |
| 621 | object->lockCount = 0; |
| 622 | |
| 623 | if (sPthreadDebugLevel >= CAPTURE_CALLSTACK) { |
| 624 | // always record the call stack when acquiring a lock. |
| 625 | // it's not efficient, but is useful during diagnostics |
| 626 | object->stackDepth = get_backtrace(object->stackTrace, STACK_TRACE_DEPTH); |
| 627 | } |
| 628 | |
| 629 | // no other locks held in this thread -- no deadlock possible! |
| 630 | if (mrl == NULL) |
| 631 | return; |
| 632 | |
| 633 | // check if the lock we're trying to acquire is a direct descendant of |
| 634 | // the most recently locked mutex in this thread, in which case we're |
| 635 | // in a good situation -- no deadlock possible |
| 636 | if (historyListHas(&mrl->children, object) >= 0) |
| 637 | return; |
| 638 | |
| 639 | pthread_mutex_lock_unchecked(&sDbgLock); |
| 640 | |
| 641 | linkParentToChild(mrl, object); |
| 642 | if (!traverseTree(object, mrl)) { |
| 643 | deinit_mapinfo(sMapInfo); |
| 644 | sMapInfo = NULL; |
| 645 | LOGW("%s\n", kEndBanner); |
| 646 | unlinkParentFromChild(mrl, object); |
| 647 | // reenable pthread debugging for this thread |
| 648 | sPthreadDebugDisabledThread = -1; |
| 649 | } else { |
| 650 | // record the call stack for this link |
| 651 | // NOTE: the call stack is added at the same index |
| 652 | // as mrl in object->parents[] |
| 653 | // ie: object->parents.count == object->stacks.count, which is |
| 654 | // also the index. |
| 655 | if (sPthreadDebugLevel >= CAPTURE_CALLSTACK) { |
| 656 | callstackListAdd(&object->stacks, |
| 657 | object->stackDepth, object->stackTrace); |
| 658 | } |
| 659 | } |
| 660 | |
| 661 | pthread_mutex_unlock_unchecked(&sDbgLock); |
| 662 | } |
| 663 | |
| 664 | static void mutex_unlock_checked(MutexInfo* object) |
| 665 | { |
| 666 | pid_t tid = gettid(); |
| 667 | if (object->owner == tid) { |
| 668 | if (object->lockCount == 0) { |
| 669 | object->owner = 0; |
| 670 | } else { |
| 671 | object->lockCount--; |
| 672 | } |
| 673 | } |
| 674 | } |
| 675 | |
| 676 | |
| 677 | // ============================================================================= |
| 678 | // Hash Table functions |
| 679 | // ============================================================================= |
| 680 | |
| 681 | /****************************************************************************/ |
| 682 | |
| 683 | #define HASHTABLE_SIZE 256 |
| 684 | |
| 685 | typedef struct HashEntry HashEntry; |
| 686 | struct HashEntry { |
| 687 | size_t slot; |
| 688 | HashEntry* prev; |
| 689 | HashEntry* next; |
| 690 | void* data; |
| 691 | }; |
| 692 | |
| 693 | typedef struct HashTable HashTable; |
| 694 | struct HashTable { |
| 695 | HashEntry* slots[HASHTABLE_SIZE]; |
| 696 | }; |
| 697 | |
| 698 | static HashTable sMutexMap; |
| 699 | static HashTable sThreadMap; |
| 700 | |
| 701 | /****************************************************************************/ |
| 702 | |
| 703 | static uint32_t get_hashcode(void const * key, size_t keySize) |
| 704 | { |
| 705 | uint32_t h = keySize; |
| 706 | char const* data = (char const*)key; |
| 707 | size_t i; |
| 708 | for (i = 0; i < keySize; i++) { |
| 709 | h = h * 31 + *data; |
| 710 | data++; |
| 711 | } |
| 712 | return (uint32_t)h; |
| 713 | } |
| 714 | |
| 715 | static size_t get_index(uint32_t h) |
| 716 | { |
| 717 | // We apply this secondary hashing discovered by Doug Lea to defend |
| 718 | // against bad hashes. |
| 719 | h += ~(h << 9); |
| 720 | h ^= (((unsigned int) h) >> 14); |
| 721 | h += (h << 4); |
| 722 | h ^= (((unsigned int) h) >> 10); |
| 723 | return (size_t)h & (HASHTABLE_SIZE - 1); |
| 724 | } |
| 725 | |
| 726 | /****************************************************************************/ |
| 727 | |
| 728 | static void hashmap_init(HashTable* table) { |
| 729 | memset(table, 0, sizeof(HashTable)); |
| 730 | } |
| 731 | |
| 732 | static void hashmap_removeEntry(HashTable* table, HashEntry* entry) |
| 733 | { |
| 734 | HashEntry* prev = entry->prev; |
| 735 | HashEntry* next = entry->next; |
| 736 | if (prev != NULL) entry->prev->next = next; |
| 737 | if (next != NULL) entry->next->prev = prev; |
| 738 | if (prev == NULL) { |
| 739 | // we are the head of the list. set the head to be next |
| 740 | table->slots[entry->slot] = entry->next; |
| 741 | } |
| 742 | } |
| 743 | |
| 744 | static HashEntry* hashmap_lookup(HashTable* table, |
| 745 | void const* key, size_t ksize, |
| 746 | int (*equals)(void const* data, void const* key)) |
| 747 | { |
| 748 | const uint32_t hash = get_hashcode(key, ksize); |
| 749 | const size_t slot = get_index(hash); |
| 750 | |
| 751 | HashEntry* entry = table->slots[slot]; |
| 752 | while (entry) { |
| 753 | if (equals(entry->data, key)) { |
| 754 | break; |
| 755 | } |
| 756 | entry = entry->next; |
| 757 | } |
| 758 | |
| 759 | if (entry == NULL) { |
| 760 | // create a new entry |
| 761 | entry = (HashEntry*)DbgAllocLocked(sizeof(HashEntry)); |
| 762 | entry->data = NULL; |
| 763 | entry->slot = slot; |
| 764 | entry->prev = NULL; |
| 765 | entry->next = table->slots[slot]; |
| 766 | if (entry->next != NULL) { |
| 767 | entry->next->prev = entry; |
| 768 | } |
| 769 | table->slots[slot] = entry; |
| 770 | } |
| 771 | return entry; |
| 772 | } |
| 773 | |
| 774 | /****************************************************************************/ |
| 775 | |
| 776 | static int MutexInfo_equals(void const* data, void const* key) { |
| 777 | return ((MutexInfo const *)data)->mutex == *(pthread_mutex_t **)key; |
| 778 | } |
| 779 | |
| 780 | static MutexInfo* get_mutex_info(pthread_mutex_t *mutex) |
| 781 | { |
| 782 | pthread_mutex_lock_unchecked(&sDbgLock); |
| 783 | |
| 784 | HashEntry* entry = hashmap_lookup(&sMutexMap, |
| 785 | &mutex, sizeof(mutex), |
| 786 | &MutexInfo_equals); |
| 787 | if (entry->data == NULL) { |
| 788 | entry->data = (MutexInfo*)DbgAllocLocked(sizeof(MutexInfo)); |
| 789 | initMutexInfo(entry->data, mutex); |
| 790 | } |
| 791 | |
| 792 | pthread_mutex_unlock_unchecked(&sDbgLock); |
| 793 | |
| 794 | return (MutexInfo *)entry->data; |
| 795 | } |
| 796 | |
| 797 | /****************************************************************************/ |
| 798 | |
| 799 | static int ThreadInfo_equals(void const* data, void const* key) { |
| 800 | return ((ThreadInfo const *)data)->pid == *(pid_t *)key; |
| 801 | } |
| 802 | |
| 803 | static ThreadInfo* get_thread_info(pid_t pid) |
| 804 | { |
| 805 | pthread_mutex_lock_unchecked(&sDbgLock); |
| 806 | |
| 807 | HashEntry* entry = hashmap_lookup(&sThreadMap, |
| 808 | &pid, sizeof(pid), |
| 809 | &ThreadInfo_equals); |
| 810 | if (entry->data == NULL) { |
| 811 | entry->data = (ThreadInfo*)DbgAllocLocked(sizeof(ThreadInfo)); |
| 812 | initThreadInfo(entry->data, pid); |
| 813 | } |
| 814 | |
| 815 | pthread_mutex_unlock_unchecked(&sDbgLock); |
| 816 | |
| 817 | return (ThreadInfo *)entry->data; |
| 818 | } |
| 819 | |
| 820 | static void push_most_recently_locked(MutexInfo* mrl) { |
| 821 | ThreadInfo* tinfo = get_thread_info(gettid()); |
| 822 | mrl->next = NULL; |
| 823 | mrl->prev = tinfo->mrl; |
| 824 | tinfo->mrl = mrl; |
| 825 | } |
| 826 | |
| 827 | static void remove_most_recently_locked(MutexInfo* mrl) { |
| 828 | ThreadInfo* tinfo = get_thread_info(gettid()); |
| 829 | if (mrl->next) { |
| 830 | (mrl->next)->prev = mrl->prev; |
| 831 | } |
| 832 | if (mrl->prev) { |
| 833 | (mrl->prev)->next = mrl->next; |
| 834 | } |
| 835 | if (tinfo->mrl == mrl) { |
| 836 | tinfo->mrl = mrl->next; |
| 837 | } |
| 838 | } |
| 839 | |
| 840 | static MutexInfo* get_most_recently_locked() { |
| 841 | ThreadInfo* tinfo = get_thread_info(gettid()); |
| 842 | return tinfo->mrl; |
| 843 | } |
| 844 | |
| 845 | /****************************************************************************/ |
| 846 | |
| 847 | /* pthread_debug_init() is called from libc_init_dynamic() just |
| 848 | * after system properties have been initialized |
| 849 | */ |
| 850 | |
| 851 | __LIBC_HIDDEN__ |
| 852 | void pthread_debug_init(void) { |
| 853 | char env[PROP_VALUE_MAX]; |
| 854 | if (__system_property_get("debug.libc.pthread", env)) { |
| 855 | int level = atoi(env); |
| 856 | if (level) { |
| 857 | LOGI("pthread deadlock detection level %d enabled for pid %d (%s)", |
| 858 | level, getpid(), __progname); |
| 859 | hashmap_init(&sMutexMap); |
| 860 | sPthreadDebugLevel = level; |
| 861 | } |
| 862 | } |
| 863 | } |
| 864 | |
| 865 | /* |
| 866 | * See if we were allowed to grab the lock at this time. We do it |
| 867 | * *after* acquiring the lock, rather than before, so that we can |
| 868 | * freely update the MutexInfo struct. This seems counter-intuitive, |
| 869 | * but our goal is deadlock *prediction* not deadlock *prevention*. |
| 870 | * (If we actually deadlock, the situation is easy to diagnose from |
| 871 | * a thread dump, so there's no point making a special effort to do |
| 872 | * the checks before the lock is held.) |
| 873 | */ |
| 874 | |
| 875 | __LIBC_HIDDEN__ |
| 876 | void pthread_debug_mutex_lock_check(pthread_mutex_t *mutex) |
| 877 | { |
| 878 | if (sPthreadDebugLevel == 0) return; |
| 879 | // prediction disabled for this thread |
| 880 | if (sPthreadDebugDisabledThread == gettid()) |
| 881 | return; |
| 882 | MutexInfo* object = get_mutex_info(mutex); |
| 883 | MutexInfo* mrl = get_most_recently_locked(); |
| 884 | mutex_lock_checked(mrl, object); |
| 885 | push_most_recently_locked(object); |
| 886 | } |
| 887 | |
| 888 | /* |
| 889 | * pthread_debug_mutex_unlock_check() must be called with the mutex |
| 890 | * still held (ie: before calling the real unlock) |
| 891 | */ |
| 892 | |
| 893 | __LIBC_HIDDEN__ |
| 894 | void pthread_debug_mutex_unlock_check(pthread_mutex_t *mutex) |
| 895 | { |
| 896 | if (sPthreadDebugLevel == 0) return; |
| 897 | // prediction disabled for this thread |
| 898 | if (sPthreadDebugDisabledThread == gettid()) |
| 899 | return; |
| 900 | MutexInfo* object = get_mutex_info(mutex); |
| 901 | remove_most_recently_locked(object); |
| 902 | mutex_unlock_checked(object); |
| 903 | } |