1 /* 2 * Copyright (c) 2015, 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 12 * in the documentation and/or other materials provided with the 13 * distribution. 14 * * Neither the name of Google, Inc. nor the names of its contributors 15 * may be used to endorse or promote products derived from this 16 * software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 22 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS 25 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 26 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 27 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 28 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <binder/IBinder.h> 33 #include <binder/IServiceManager.h> 34 #include <binder/Parcel.h> 35 36 #include <ctime> 37 #include <cutils/properties.h> 38 #include <signal.h> 39 #include <stdbool.h> 40 #include <stdio.h> 41 #include <stdlib.h> 42 #include <string.h> 43 44 #include <sys/resource.h> 45 #include <sys/stat.h> 46 #include <sys/time.h> 47 #include <sys/types.h> 48 #include <unistd.h> 49 50 #include <utils/Log.h> 51 #include <utils/String8.h> 52 #include <utils/Trace.h> 53 #include <zlib.h> 54 55 using namespace android; 56 57 #ifdef LOG_TAG 58 #undef LOG_TAG 59 #endif 60 61 #define LOG_TAG "anrdaemon" 62 63 static const int check_period = 1; // in sec 64 static const int tracing_check_period = 500000; // in micro sec 65 static const int cpu_stat_entries = 7; // number of cpu stat entries 66 static const int min_buffer_size = 16; 67 static const int max_buffer_size = 2048; 68 static const char *min_buffer_size_str = "16"; 69 static const char *max_buffer_size_str = "2048"; 70 static const int time_buf_size = 20; 71 static const int path_buf_size = 60; 72 73 typedef struct cpu_stat { 74 unsigned long utime, ntime, stime, itime; 75 unsigned long iowtime, irqtime, sirqtime, steal; 76 unsigned long total; 77 } cpu_stat_t; 78 79 /* 80 * Logging on/off threshold. 81 * Uint: 0.01%; default to 99.90% cpu. 82 */ 83 static int idle_threshold = 10; 84 85 static bool quit = false; 86 static bool suspend= false; 87 static bool err = false; 88 static char err_msg[100]; 89 static bool tracing = false; 90 91 static const char *buf_size_kb = "2048"; 92 static const char *apps = ""; 93 static uint64_t tag = 0; 94 95 static cpu_stat_t new_cpu; 96 static cpu_stat_t old_cpu; 97 98 /* Log certain kernel activity when enabled */ 99 static bool log_sched = false; 100 static bool log_stack = false; 101 static bool log_irq = false; 102 static bool log_sync = false; 103 static bool log_workq = false; 104 105 /* Paths for debugfs controls*/ 106 static const char* dfs_trace_output_path = 107 "/d/tracing/trace"; 108 static const char* dfs_irq_path = 109 "/d/tracing/events/irq/enable"; 110 static const char* dfs_sync_path = 111 "/d/tracing/events/sync/enable"; 112 static const char* dfs_workq_path = 113 "/d/tracing/events/workqueue/enable"; 114 static const char* dfs_stack_path = 115 "/d/tracing/options/stacktrace"; 116 static const char* dfs_sched_switch_path = 117 "/d/tracing/events/sched/sched_switch/enable"; 118 static const char* dfs_sched_wakeup_path = 119 "/d/tracing/events/sched/sched_wakeup/enable"; 120 static const char* dfs_control_path = 121 "/d/tracing/tracing_on"; 122 static const char* dfs_buffer_size_path = 123 "/d/tracing/buffer_size_kb"; 124 static const char* dfs_tags_property = "debug.atrace.tags.enableflags"; 125 static const char* dfs_apps_property = "debug.atrace.app_cmdlines"; 126 127 /* 128 * Read accumulated cpu data from /proc/stat 129 */ 130 static void get_cpu_stat(cpu_stat_t *cpu) { 131 FILE *fp = NULL; 132 const char *params = "cpu %lu %lu %lu %lu %lu %lu %lu %*d %*d %*d\n"; 133 134 if ((fp = fopen("/proc/stat", "r")) == NULL) { 135 err = true; 136 sprintf(err_msg, "can't read from /proc/stat with errno %d", errno); 137 } else { 138 if (fscanf(fp, params, &cpu->utime, &cpu->ntime, 139 &cpu->stime, &cpu->itime, &cpu->iowtime, &cpu->irqtime, 140 &cpu->sirqtime) != cpu_stat_entries) { 141 /* 142 * If failed in getting status, new_cpu won't be updated and 143 * is_heavy_loaded() will return false. 144 */ 145 ALOGE("Error in getting cpu status. Skipping this check."); 146 return; 147 } 148 149 cpu->total = cpu->utime + cpu->ntime + cpu->stime + cpu->itime 150 + cpu->iowtime + cpu->irqtime + cpu->sirqtime; 151 152 fclose(fp); 153 } 154 } 155 156 /* 157 * Calculate cpu usage in the past interval. 158 * If tracing is on, increase the idle threshold by 1.00% so that we do not 159 * turn on and off tracing frequently whe the cpu load is right close to 160 * threshold. 161 */ 162 static bool is_heavy_load(void) { 163 unsigned long diff_idle, diff_total; 164 int threshold = idle_threshold + (tracing?100:0); 165 get_cpu_stat(&new_cpu); 166 diff_idle = new_cpu.itime - old_cpu.itime; 167 diff_total = new_cpu.total - old_cpu.total; 168 old_cpu = new_cpu; 169 return (diff_idle * 10000 < diff_total * threshold); 170 } 171 172 /* 173 * Force the userland processes to refresh their property for logging. 174 */ 175 static void dfs_poke_binder(void) { 176 sp<IServiceManager> sm = defaultServiceManager(); 177 Vector<String16> services = sm->listServices(); 178 for (size_t i = 0; i < services.size(); i++) { 179 sp<IBinder> obj = sm->checkService(services[i]); 180 if (obj != NULL) { 181 Parcel data; 182 obj->transact(IBinder::SYSPROPS_TRANSACTION, data, NULL, 0); 183 } 184 } 185 } 186 187 /* 188 * Enable/disable a debugfs property by writing 0/1 to its path. 189 */ 190 static int dfs_enable(bool enable, const char* path) { 191 int fd = open(path, O_WRONLY); 192 if (fd == -1) { 193 err = true; 194 sprintf(err_msg, "Can't open %s. Error: %d", path, errno); 195 return -1; 196 } 197 const char* control = (enable?"1":"0"); 198 ssize_t len = strlen(control); 199 int max_try = 10; // Fail if write was interrupted for 10 times 200 while (write(fd, control, len) != len) { 201 if (errno == EINTR && max_try-- > 0) { 202 usleep(100); 203 continue; 204 } 205 206 err = true; 207 sprintf(err_msg, "Error %d in writing to %s.", errno, path); 208 } 209 close(fd); 210 return (err?-1:0); 211 } 212 213 /* 214 * Set the userland tracing properties. 215 */ 216 static void dfs_set_property(uint64_t mtag, const char* mapp, bool enable) { 217 char buf[64]; 218 snprintf(buf, 64, "%#" PRIx64, mtag); 219 if (property_set(dfs_tags_property, buf) < 0) { 220 err = true; 221 sprintf(err_msg, "Failed to set debug tags system properties."); 222 } 223 224 if (strlen(mapp) > 0 225 && property_set(dfs_apps_property, mapp) < 0) { 226 err = true; 227 sprintf(err_msg, "Failed to set debug applications."); 228 } 229 230 if (log_sched) { 231 dfs_enable(enable, dfs_sched_switch_path); 232 dfs_enable(enable, dfs_sched_wakeup_path); 233 } 234 if (log_stack) { 235 dfs_enable(enable, dfs_stack_path); 236 } 237 if (log_irq) { 238 dfs_enable(enable, dfs_irq_path); 239 } 240 if (log_sync) { 241 dfs_enable(enable, dfs_sync_path); 242 } 243 if (log_workq) { 244 dfs_enable(enable, dfs_workq_path); 245 } 246 } 247 248 /* 249 * Start logging when cpu usage is high. Meanwhile, moniter the cpu usage and 250 * stop logging when it drops down. 251 */ 252 static void start_tracing(void) { 253 ALOGD("High cpu usage, start logging."); 254 255 if (dfs_enable(true, dfs_control_path) != 0) { 256 ALOGE("Failed to start tracing."); 257 return; 258 } 259 tracing = true; 260 261 /* Stop logging when cpu usage drops or the daemon is suspended.*/ 262 do { 263 usleep(tracing_check_period); 264 } while (!suspend && is_heavy_load()); 265 266 if (dfs_enable(false, dfs_control_path) != 0) { 267 ALOGE("Failed to stop tracing."); 268 } 269 270 ALOGD("Usage back to low, stop logging."); 271 tracing = false; 272 } 273 274 /* 275 * Set the tracing log buffer size. 276 * Note the actual buffer size will be buf_size_kb * number of cores. 277 * E.g. for dory, the total buffer size is buf_size_kb * 4. 278 */ 279 static int set_tracing_buffer_size(void) { 280 int fd = open(dfs_buffer_size_path, O_WRONLY); 281 if (fd == -1) { 282 err = true; 283 sprintf(err_msg, "Can't open atrace buffer size file under /d/tracing."); 284 return -1; 285 } 286 ssize_t len = strlen(buf_size_kb); 287 if (write(fd, buf_size_kb, len) != len) { 288 err = true; 289 sprintf(err_msg, "Error in writing to atrace buffer size file."); 290 } 291 close(fd); 292 return (err?-1:0); 293 294 } 295 296 /* 297 * Main loop to moniter the cpu usage and decided whether to start logging. 298 */ 299 static void start(void) { 300 if ((set_tracing_buffer_size()) != 0) 301 return; 302 303 dfs_set_property(tag, apps, true); 304 dfs_poke_binder(); 305 306 get_cpu_stat(&old_cpu); 307 sleep(check_period); 308 309 while (!quit && !err) { 310 if (!suspend && is_heavy_load()) { 311 /* 312 * Increase process priority to make sure we can stop logging when 313 * necessary and do not overwrite the buffer 314 */ 315 setpriority(PRIO_PROCESS, 0, -20); 316 start_tracing(); 317 setpriority(PRIO_PROCESS, 0, 0); 318 } 319 sleep(check_period); 320 } 321 return; 322 } 323 324 /* 325 * Dump the log in a compressed format for systrace to visualize. 326 */ 327 static void dump_trace() 328 { 329 int remain_attempts = 5; 330 suspend = true; 331 while (tracing) { 332 ALOGI("Waiting logging to stop."); 333 usleep(tracing_check_period); 334 remain_attempts--; 335 if (remain_attempts == 0) { 336 ALOGE("Can't stop logging after 5 attempts. Dump aborted."); 337 return; 338 } 339 } 340 341 /* 342 * Create a dump file "dump_of_anrdaemon.<current_time>" under /data/misc/anrd 343 */ 344 time_t now = time(0); 345 struct tm tstruct; 346 char time_buf[time_buf_size]; 347 char path_buf[path_buf_size]; 348 const char* header = " done\nTRACE:\n"; 349 ssize_t header_len = strlen(header); 350 tstruct = *localtime(&now); 351 strftime(time_buf, time_buf_size, "%Y-%m-%d.%X", &tstruct); 352 snprintf(path_buf, path_buf_size, "/data/misc/anrd/dump_of_anrdaemon.%s", time_buf); 353 int output_fd = creat(path_buf, S_IRWXU); 354 if (output_fd == -1) { 355 ALOGE("Failed to create %s. Dump aborted.", path_buf); 356 return; 357 } 358 359 if (write(output_fd, header, strlen(header)) != header_len) { 360 ALOGE("Failed to write the header."); 361 close(output_fd); 362 return; 363 } 364 365 int trace_fd = open(dfs_trace_output_path, O_RDWR); 366 if (trace_fd == -1) { 367 ALOGE("Failed to open %s. Dump aborted.", dfs_trace_output_path); 368 close(output_fd); 369 return; 370 } 371 372 z_stream zs; 373 uint8_t *in, *out; 374 int result, flush; 375 376 memset(&zs, 0, sizeof(zs)); 377 result = deflateInit(&zs, Z_DEFAULT_COMPRESSION); 378 if (result != Z_OK) { 379 ALOGE("error initializing zlib: %d\n", result); 380 close(trace_fd); 381 close(output_fd); 382 return; 383 } 384 385 const size_t bufSize = 64*1024; 386 in = (uint8_t*)malloc(bufSize); 387 out = (uint8_t*)malloc(bufSize); 388 flush = Z_NO_FLUSH; 389 390 zs.next_out = out; 391 zs.avail_out = bufSize; 392 393 do { 394 if (zs.avail_in == 0) { 395 result = read(trace_fd, in, bufSize); 396 if (result < 0) { 397 ALOGE("error reading trace: %s", strerror(errno)); 398 result = Z_STREAM_END; 399 break; 400 } else if (result == 0) { 401 flush = Z_FINISH; 402 } else { 403 zs.next_in = in; 404 zs.avail_in = result; 405 } 406 } 407 408 if (zs.avail_out == 0) { 409 result = write(output_fd, out, bufSize); 410 if ((size_t)result < bufSize) { 411 ALOGE("error writing deflated trace: %s", strerror(errno)); 412 result = Z_STREAM_END; 413 zs.avail_out = bufSize; 414 break; 415 } 416 zs.next_out = out; 417 zs.avail_out = bufSize; 418 } 419 420 } while ((result = deflate(&zs, flush)) == Z_OK); 421 422 if (result != Z_STREAM_END) { 423 ALOGE("error deflating trace: %s\n", zs.msg); 424 } 425 426 if (zs.avail_out < bufSize) { 427 size_t bytes = bufSize - zs.avail_out; 428 result = write(output_fd, out, bytes); 429 if ((size_t)result < bytes) { 430 ALOGE("error writing deflated trace: %s", strerror(errno)); 431 } 432 } 433 434 result = deflateEnd(&zs); 435 if (result != Z_OK) { 436 ALOGE("error cleaning up zlib: %d\n", result); 437 } 438 439 free(in); 440 free(out); 441 442 close(trace_fd); 443 close(output_fd); 444 445 suspend = false; 446 ALOGI("Finished dump. Output file stored at: %s", path_buf); 447 } 448 449 static void handle_signal(int signo) 450 { 451 switch (signo) { 452 case SIGQUIT: 453 suspend = true; 454 quit = true; 455 break; 456 case SIGSTOP: 457 suspend = true; 458 break; 459 case SIGCONT: 460 suspend = false; 461 break; 462 case SIGUSR1: 463 dump_trace(); 464 } 465 } 466 467 /* 468 * Set the signal handler: 469 * SIGQUIT: Reset debugfs and tracing property and terminate the daemon. 470 * SIGSTOP: Stop logging and suspend the daemon. 471 * SIGCONT: Resume the daemon as normal. 472 * SIGUSR1: Dump the logging to a compressed format for systrace to visualize. 473 */ 474 static void register_sighandler(void) 475 { 476 struct sigaction sa; 477 sigset_t block_mask; 478 479 sigemptyset(&block_mask); 480 sigaddset (&block_mask, SIGQUIT); 481 sigaddset (&block_mask, SIGSTOP); 482 sigaddset (&block_mask, SIGCONT); 483 sigaddset (&block_mask, SIGUSR1); 484 485 sa.sa_flags = 0; 486 sa.sa_mask = block_mask; 487 sa.sa_handler = handle_signal; 488 sigaction(SIGQUIT, &sa, NULL); 489 sigaction(SIGSTOP, &sa, NULL); 490 sigaction(SIGCONT, &sa, NULL); 491 sigaction(SIGUSR1, &sa, NULL); 492 } 493 494 static void show_help(void) { 495 496 fprintf(stderr, "usage: ANRdaemon [options] [categoris...]\n"); 497 fprintf(stdout, "Options includes:\n" 498 " -a appname enable app-level tracing for a comma " 499 "separated list of cmdlines\n" 500 " -t N cpu threshold for logging to start " 501 "(uint = 0.01%%, min = 5000, max = 9999, default = 9990)\n" 502 " -s N use a trace buffer size of N KB " 503 "default to 2048KB\n" 504 " -h show helps\n"); 505 fprintf(stdout, "Categoris includes:\n" 506 " am - activity manager\n" 507 " sm - sync manager\n" 508 " input - input\n" 509 " dalvik - dalvik VM\n" 510 " audio - Audio\n" 511 " gfx - Graphics\n" 512 " rs - RenderScript\n" 513 " hal - Hardware Modules\n" 514 " irq - kernel irq events\n" 515 " sched - kernel scheduler activity\n" 516 " stack - kernel stack\n" 517 " sync - kernel sync activity\n" 518 " workq - kernel work queues\n"); 519 fprintf(stdout, "Control includes:\n" 520 " SIGQUIT: terminate the process\n" 521 " SIGSTOP: suspend all function of the daemon\n" 522 " SIGCONT: resume the normal function\n" 523 " SIGUSR1: dump the current logging in a compressed form\n"); 524 exit(0); 525 } 526 527 static int get_options(int argc, char *argv[]) { 528 int opt = 0; 529 int threshold; 530 while ((opt = getopt(argc, argv, "a:s:t:h")) >= 0) { 531 switch(opt) { 532 case 'a': 533 apps = optarg; 534 break; 535 case 's': 536 if (atoi(optarg) > max_buffer_size) 537 buf_size_kb = max_buffer_size_str; 538 else if (atoi(optarg) < min_buffer_size) 539 buf_size_kb = min_buffer_size_str; 540 else 541 buf_size_kb = optarg; 542 break; 543 case 't': 544 threshold = atoi(optarg); 545 if (threshold > 9999 || threshold < 5000) { 546 fprintf(stderr, "logging threshold should be 5000-9999\n"); 547 return 1; 548 } 549 idle_threshold = 10000 - threshold; 550 break; 551 case 'h': 552 show_help(); 553 break; 554 default: 555 fprintf(stderr, "Error in getting options.\n" 556 "run \"%s -h\" for usage.\n", argv[0]); 557 return 1; 558 } 559 } 560 561 for (int i = optind; i < argc; i++) { 562 if (strcmp(argv[i], "am") == 0) { 563 tag |= ATRACE_TAG_ACTIVITY_MANAGER; 564 } else if (strcmp(argv[i], "input") == 0) { 565 tag |= ATRACE_TAG_INPUT; 566 } else if (strcmp(argv[i], "sm") == 0) { 567 tag |= ATRACE_TAG_SYNC_MANAGER; 568 } else if (strcmp(argv[i], "dalvik") == 0) { 569 tag |= ATRACE_TAG_DALVIK; 570 } else if (strcmp(argv[i], "gfx") == 0) { 571 tag |= ATRACE_TAG_GRAPHICS; 572 } else if (strcmp(argv[i], "audio") == 0) { 573 tag |= ATRACE_TAG_AUDIO; 574 } else if (strcmp(argv[i], "hal") == 0) { 575 tag |= ATRACE_TAG_HAL; 576 } else if (strcmp(argv[i], "rs") == 0) { 577 tag |= ATRACE_TAG_RS; 578 } else if (strcmp(argv[i], "sched") == 0) { 579 log_sched = true; 580 } else if (strcmp(argv[i], "stack") == 0) { 581 log_stack = true; 582 } else if (strcmp(argv[i], "workq") == 0) { 583 log_workq = true; 584 } else if (strcmp(argv[i], "irq") == 0) { 585 log_irq = true; 586 } else if (strcmp(argv[i], "sync") == 0) { 587 log_sync = true; 588 } else { 589 fprintf(stderr, "invalid category: %s\n" 590 "run \"%s -h\" for usage.\n", argv[i], argv[0]); 591 return 1; 592 } 593 } 594 595 /* If nothing is enabled, don't run */ 596 if (!tag && !log_sched && !log_stack && !log_workq && !log_irq && !log_sync) { 597 ALOGE("Specify at least one category to trace."); 598 return 1; 599 } 600 601 return 0; 602 } 603 604 int main(int argc, char *argv[]) 605 { 606 if(get_options(argc, argv) != 0) 607 return 1; 608 609 if (daemon(0, 0) != 0) 610 return 1; 611 612 register_sighandler(); 613 614 /* Clear any the trace log file by overwrite it with a new file */ 615 int fd = creat(dfs_trace_output_path, 0); 616 if (fd == -1) { 617 ALOGE("Faield to open and cleaup previous log"); 618 return 1; 619 } 620 close(fd); 621 622 ALOGI("ANRdaemon starting"); 623 start(); 624 625 if (err) 626 ALOGE("ANRdaemon stopped due to Error: %s", err_msg); 627 628 ALOGI("ANRdaemon terminated."); 629 630 return (err?1:0); 631 } 632