1 /*- 2 * Copyright (c) 1982, 1986, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_kdb.h" 41 #include "opt_device_polling.h" 42 #include "opt_hwpmc_hooks.h" 43 #include "opt_ntp.h" 44 #include "opt_watchdog.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/callout.h> 49 #include <sys/kdb.h> 50 #include <sys/kernel.h> 51 #include <sys/kthread.h> 52 #include <sys/ktr.h> 53 #include <sys/lock.h> 54 #include <sys/mutex.h> 55 #include <sys/proc.h> 56 #include <sys/resource.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sched.h> 59 #include <sys/sdt.h> 60 #include <sys/signalvar.h> 61 #include <sys/sleepqueue.h> 62 #include <sys/smp.h> 63 #include <vm/vm.h> 64 #include <vm/pmap.h> 65 #include <vm/vm_map.h> 66 #include <sys/sysctl.h> 67 #include <sys/bus.h> 68 #include <sys/interrupt.h> 69 #include <sys/limits.h> 70 #include <sys/timetc.h> 71 72 #ifdef GPROF 73 #include <sys/gmon.h> 74 #endif 75 76 #ifdef HWPMC_HOOKS 77 #include <sys/pmckern.h> 78 PMC_SOFT_DEFINE( , , clock, hard); 79 PMC_SOFT_DEFINE( , , clock, stat); 80 PMC_SOFT_DEFINE_EX( , , clock, prof, \ 81 cpu_startprofclock, cpu_stopprofclock); 82 #endif 83 84 #ifdef DEVICE_POLLING 85 extern void hardclock_device_poll(void); 86 #endif /* DEVICE_POLLING */ 87 88 static void initclocks(void *dummy); 89 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL); 90 91 /* Spin-lock protecting profiling statistics. */ 92 static struct mtx time_lock; 93 94 SDT_PROVIDER_DECLARE(sched); 95 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *"); 96 97 static int 98 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS) 99 { 100 int error; 101 long cp_time[CPUSTATES]; 102 #ifdef SCTL_MASK32 103 int i; 104 unsigned int cp_time32[CPUSTATES]; 105 #endif 106 107 read_cpu_time(cp_time); 108 #ifdef SCTL_MASK32 109 if (req->flags & SCTL_MASK32) { 110 if (!req->oldptr) 111 return SYSCTL_OUT(req, 0, sizeof(cp_time32)); 112 for (i = 0; i < CPUSTATES; i++) 113 cp_time32[i] = (unsigned int)cp_time[i]; 114 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 115 } else 116 #endif 117 { 118 if (!req->oldptr) 119 return SYSCTL_OUT(req, 0, sizeof(cp_time)); 120 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time)); 121 } 122 return error; 123 } 124 125 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 126 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics"); 127 128 static long empty[CPUSTATES]; 129 130 static int 131 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS) 132 { 133 struct pcpu *pcpu; 134 int error; 135 int c; 136 long *cp_time; 137 #ifdef SCTL_MASK32 138 unsigned int cp_time32[CPUSTATES]; 139 int i; 140 #endif 141 142 if (!req->oldptr) { 143 #ifdef SCTL_MASK32 144 if (req->flags & SCTL_MASK32) 145 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1)); 146 else 147 #endif 148 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1)); 149 } 150 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) { 151 if (!CPU_ABSENT(c)) { 152 pcpu = pcpu_find(c); 153 cp_time = pcpu->pc_cp_time; 154 } else { 155 cp_time = empty; 156 } 157 #ifdef SCTL_MASK32 158 if (req->flags & SCTL_MASK32) { 159 for (i = 0; i < CPUSTATES; i++) 160 cp_time32[i] = (unsigned int)cp_time[i]; 161 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 162 } else 163 #endif 164 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES); 165 } 166 return error; 167 } 168 169 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 170 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics"); 171 172 #ifdef DEADLKRES 173 static const char *blessed[] = { 174 "getblk", 175 "so_snd_sx", 176 "so_rcv_sx", 177 NULL 178 }; 179 static int slptime_threshold = 1800; 180 static int blktime_threshold = 900; 181 static int sleepfreq = 3; 182 183 static void 184 deadlkres(void) 185 { 186 struct proc *p; 187 struct thread *td; 188 void *wchan; 189 int blkticks, i, slpticks, slptype, tryl, tticks; 190 191 tryl = 0; 192 for (;;) { 193 blkticks = blktime_threshold * hz; 194 slpticks = slptime_threshold * hz; 195 196 /* 197 * Avoid to sleep on the sx_lock in order to avoid a possible 198 * priority inversion problem leading to starvation. 199 * If the lock can't be held after 100 tries, panic. 200 */ 201 if (!sx_try_slock(&allproc_lock)) { 202 if (tryl > 100) 203 panic("%s: possible deadlock detected on allproc_lock\n", 204 __func__); 205 tryl++; 206 pause("allproc", sleepfreq * hz); 207 continue; 208 } 209 tryl = 0; 210 FOREACH_PROC_IN_SYSTEM(p) { 211 PROC_LOCK(p); 212 if (p->p_state == PRS_NEW) { 213 PROC_UNLOCK(p); 214 continue; 215 } 216 FOREACH_THREAD_IN_PROC(p, td) { 217 218 thread_lock(td); 219 if (TD_ON_LOCK(td)) { 220 221 /* 222 * The thread should be blocked on a 223 * turnstile, simply check if the 224 * turnstile channel is in good state. 225 */ 226 MPASS(td->td_blocked != NULL); 227 228 tticks = ticks - td->td_blktick; 229 thread_unlock(td); 230 if (tticks > blkticks) { 231 232 /* 233 * Accordingly with provided 234 * thresholds, this thread is 235 * stuck for too long on a 236 * turnstile. 237 */ 238 PROC_UNLOCK(p); 239 sx_sunlock(&allproc_lock); 240 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", 241 __func__, td, tticks); 242 } 243 } else if (TD_IS_SLEEPING(td) && 244 TD_ON_SLEEPQ(td)) { 245 246 /* 247 * Check if the thread is sleeping on a 248 * lock, otherwise skip the check. 249 * Drop the thread lock in order to 250 * avoid a LOR with the sleepqueue 251 * spinlock. 252 */ 253 wchan = td->td_wchan; 254 tticks = ticks - td->td_slptick; 255 thread_unlock(td); 256 slptype = sleepq_type(wchan); 257 if ((slptype == SLEEPQ_SX || 258 slptype == SLEEPQ_LK) && 259 tticks > slpticks) { 260 261 /* 262 * Accordingly with provided 263 * thresholds, this thread is 264 * stuck for too long on a 265 * sleepqueue. 266 * However, being on a 267 * sleepqueue, we might still 268 * check for the blessed 269 * list. 270 */ 271 tryl = 0; 272 for (i = 0; blessed[i] != NULL; 273 i++) { 274 if (!strcmp(blessed[i], 275 td->td_wmesg)) { 276 tryl = 1; 277 break; 278 } 279 } 280 if (tryl != 0) { 281 tryl = 0; 282 continue; 283 } 284 PROC_UNLOCK(p); 285 sx_sunlock(&allproc_lock); 286 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", 287 __func__, td, tticks); 288 } 289 } else 290 thread_unlock(td); 291 } 292 PROC_UNLOCK(p); 293 } 294 sx_sunlock(&allproc_lock); 295 296 /* Sleep for sleepfreq seconds. */ 297 pause("-", sleepfreq * hz); 298 } 299 } 300 301 static struct kthread_desc deadlkres_kd = { 302 "deadlkres", 303 deadlkres, 304 (struct thread **)NULL 305 }; 306 307 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd); 308 309 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, 310 "Deadlock resolver"); 311 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW, 312 &slptime_threshold, 0, 313 "Number of seconds within is valid to sleep on a sleepqueue"); 314 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW, 315 &blktime_threshold, 0, 316 "Number of seconds within is valid to block on a turnstile"); 317 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0, 318 "Number of seconds between any deadlock resolver thread run"); 319 #endif /* DEADLKRES */ 320 321 void 322 read_cpu_time(long *cp_time) 323 { 324 struct pcpu *pc; 325 int i, j; 326 327 /* Sum up global cp_time[]. */ 328 bzero(cp_time, sizeof(long) * CPUSTATES); 329 CPU_FOREACH(i) { 330 pc = pcpu_find(i); 331 for (j = 0; j < CPUSTATES; j++) 332 cp_time[j] += pc->pc_cp_time[j]; 333 } 334 } 335 336 #ifdef SW_WATCHDOG 337 #include <sys/watchdog.h> 338 339 static int watchdog_ticks; 340 static int watchdog_enabled; 341 static void watchdog_fire(void); 342 static void watchdog_config(void *, u_int, int *); 343 #endif /* SW_WATCHDOG */ 344 345 /* 346 * Clock handling routines. 347 * 348 * This code is written to operate with two timers that run independently of 349 * each other. 350 * 351 * The main timer, running hz times per second, is used to trigger interval 352 * timers, timeouts and rescheduling as needed. 353 * 354 * The second timer handles kernel and user profiling, 355 * and does resource use estimation. If the second timer is programmable, 356 * it is randomized to avoid aliasing between the two clocks. For example, 357 * the randomization prevents an adversary from always giving up the cpu 358 * just before its quantum expires. Otherwise, it would never accumulate 359 * cpu ticks. The mean frequency of the second timer is stathz. 360 * 361 * If no second timer exists, stathz will be zero; in this case we drive 362 * profiling and statistics off the main clock. This WILL NOT be accurate; 363 * do not do it unless absolutely necessary. 364 * 365 * The statistics clock may (or may not) be run at a higher rate while 366 * profiling. This profile clock runs at profhz. We require that profhz 367 * be an integral multiple of stathz. 368 * 369 * If the statistics clock is running fast, it must be divided by the ratio 370 * profhz/stathz for statistics. (For profiling, every tick counts.) 371 * 372 * Time-of-day is maintained using a "timecounter", which may or may 373 * not be related to the hardware generating the above mentioned 374 * interrupts. 375 */ 376 377 int stathz; 378 int profhz; 379 int profprocs; 380 volatile int ticks; 381 int psratio; 382 383 static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */ 384 static int global_hardclock_run = 0; 385 386 /* 387 * Initialize clock frequencies and start both clocks running. 388 */ 389 /* ARGSUSED*/ 390 static void 391 initclocks(dummy) 392 void *dummy; 393 { 394 register int i; 395 396 /* 397 * Set divisors to 1 (normal case) and let the machine-specific 398 * code do its bit. 399 */ 400 mtx_init(&time_lock, "time lock", NULL, MTX_DEF); 401 cpu_initclocks(); 402 403 /* 404 * Compute profhz/stathz, and fix profhz if needed. 405 */ 406 i = stathz ? stathz : hz; 407 if (profhz == 0) 408 profhz = i; 409 psratio = profhz / i; 410 #ifdef SW_WATCHDOG 411 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 412 #endif 413 /* 414 * Arrange for ticks to wrap 10 minutes after boot to help catch 415 * sign problems sooner. 416 */ 417 ticks = INT_MAX - (hz * 10 * 60); 418 } 419 420 /* 421 * Each time the real-time timer fires, this function is called on all CPUs. 422 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only 423 * the other CPUs in the system need to call this function. 424 */ 425 void 426 hardclock_cpu(int usermode) 427 { 428 struct pstats *pstats; 429 struct thread *td = curthread; 430 struct proc *p = td->td_proc; 431 int flags; 432 433 /* 434 * Run current process's virtual and profile time, as needed. 435 */ 436 pstats = p->p_stats; 437 flags = 0; 438 if (usermode && 439 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 440 PROC_ITIMLOCK(p); 441 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 442 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 443 PROC_ITIMUNLOCK(p); 444 } 445 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 446 PROC_ITIMLOCK(p); 447 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 448 flags |= TDF_PROFPEND | TDF_ASTPENDING; 449 PROC_ITIMUNLOCK(p); 450 } 451 thread_lock(td); 452 sched_tick(1); 453 td->td_flags |= flags; 454 thread_unlock(td); 455 456 #ifdef HWPMC_HOOKS 457 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 458 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 459 if (td->td_intr_frame != NULL) 460 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 461 #endif 462 callout_process(sbinuptime()); 463 } 464 465 /* 466 * The real-time timer, interrupting hz times per second. 467 */ 468 void 469 hardclock(int usermode, uintfptr_t pc) 470 { 471 472 atomic_add_int(&ticks, 1); 473 hardclock_cpu(usermode); 474 tc_ticktock(1); 475 cpu_tick_calibration(); 476 /* 477 * If no separate statistics clock is available, run it from here. 478 * 479 * XXX: this only works for UP 480 */ 481 if (stathz == 0) { 482 profclock(usermode, pc); 483 statclock(usermode); 484 } 485 #ifdef DEVICE_POLLING 486 hardclock_device_poll(); /* this is very short and quick */ 487 #endif /* DEVICE_POLLING */ 488 #ifdef SW_WATCHDOG 489 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 490 watchdog_fire(); 491 #endif /* SW_WATCHDOG */ 492 } 493 494 void 495 hardclock_cnt(int cnt, int usermode) 496 { 497 struct pstats *pstats; 498 struct thread *td = curthread; 499 struct proc *p = td->td_proc; 500 int *t = DPCPU_PTR(pcputicks); 501 int flags, global, newticks; 502 #ifdef SW_WATCHDOG 503 int i; 504 #endif /* SW_WATCHDOG */ 505 506 /* 507 * Update per-CPU and possibly global ticks values. 508 */ 509 *t += cnt; 510 do { 511 global = ticks; 512 newticks = *t - global; 513 if (newticks <= 0) { 514 if (newticks < -1) 515 *t = global - 1; 516 newticks = 0; 517 break; 518 } 519 } while (!atomic_cmpset_int(&ticks, global, *t)); 520 521 /* 522 * Run current process's virtual and profile time, as needed. 523 */ 524 pstats = p->p_stats; 525 flags = 0; 526 if (usermode && 527 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 528 PROC_ITIMLOCK(p); 529 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], 530 tick * cnt) == 0) 531 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 532 PROC_ITIMUNLOCK(p); 533 } 534 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 535 PROC_ITIMLOCK(p); 536 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], 537 tick * cnt) == 0) 538 flags |= TDF_PROFPEND | TDF_ASTPENDING; 539 PROC_ITIMUNLOCK(p); 540 } 541 thread_lock(td); 542 sched_tick(cnt); 543 td->td_flags |= flags; 544 thread_unlock(td); 545 546 #ifdef HWPMC_HOOKS 547 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 548 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 549 if (td->td_intr_frame != NULL) 550 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 551 #endif 552 /* We are in charge to handle this tick duty. */ 553 if (newticks > 0) { 554 /* Dangerous and no need to call these things concurrently. */ 555 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) { 556 tc_ticktock(newticks); 557 #ifdef DEVICE_POLLING 558 /* This is very short and quick. */ 559 hardclock_device_poll(); 560 #endif /* DEVICE_POLLING */ 561 atomic_store_rel_int(&global_hardclock_run, 0); 562 } 563 #ifdef SW_WATCHDOG 564 if (watchdog_enabled > 0) { 565 i = atomic_fetchadd_int(&watchdog_ticks, -newticks); 566 if (i > 0 && i <= newticks) 567 watchdog_fire(); 568 } 569 #endif /* SW_WATCHDOG */ 570 } 571 if (curcpu == CPU_FIRST()) 572 cpu_tick_calibration(); 573 } 574 575 void 576 hardclock_sync(int cpu) 577 { 578 int *t = DPCPU_ID_PTR(cpu, pcputicks); 579 580 *t = ticks; 581 } 582 583 /* 584 * Compute number of ticks in the specified amount of time. 585 */ 586 int 587 tvtohz(tv) 588 struct timeval *tv; 589 { 590 register unsigned long ticks; 591 register long sec, usec; 592 593 /* 594 * If the number of usecs in the whole seconds part of the time 595 * difference fits in a long, then the total number of usecs will 596 * fit in an unsigned long. Compute the total and convert it to 597 * ticks, rounding up and adding 1 to allow for the current tick 598 * to expire. Rounding also depends on unsigned long arithmetic 599 * to avoid overflow. 600 * 601 * Otherwise, if the number of ticks in the whole seconds part of 602 * the time difference fits in a long, then convert the parts to 603 * ticks separately and add, using similar rounding methods and 604 * overflow avoidance. This method would work in the previous 605 * case but it is slightly slower and assumes that hz is integral. 606 * 607 * Otherwise, round the time difference down to the maximum 608 * representable value. 609 * 610 * If ints have 32 bits, then the maximum value for any timeout in 611 * 10ms ticks is 248 days. 612 */ 613 sec = tv->tv_sec; 614 usec = tv->tv_usec; 615 if (usec < 0) { 616 sec--; 617 usec += 1000000; 618 } 619 if (sec < 0) { 620 #ifdef DIAGNOSTIC 621 if (usec > 0) { 622 sec++; 623 usec -= 1000000; 624 } 625 printf("tvotohz: negative time difference %ld sec %ld usec\n", 626 sec, usec); 627 #endif 628 ticks = 1; 629 } else if (sec <= LONG_MAX / 1000000) 630 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 631 / tick + 1; 632 else if (sec <= LONG_MAX / hz) 633 ticks = sec * hz 634 + ((unsigned long)usec + (tick - 1)) / tick + 1; 635 else 636 ticks = LONG_MAX; 637 if (ticks > INT_MAX) 638 ticks = INT_MAX; 639 return ((int)ticks); 640 } 641 642 /* 643 * Start profiling on a process. 644 * 645 * Kernel profiling passes proc0 which never exits and hence 646 * keeps the profile clock running constantly. 647 */ 648 void 649 startprofclock(p) 650 register struct proc *p; 651 { 652 653 PROC_LOCK_ASSERT(p, MA_OWNED); 654 if (p->p_flag & P_STOPPROF) 655 return; 656 if ((p->p_flag & P_PROFIL) == 0) { 657 p->p_flag |= P_PROFIL; 658 mtx_lock(&time_lock); 659 if (++profprocs == 1) 660 cpu_startprofclock(); 661 mtx_unlock(&time_lock); 662 } 663 } 664 665 /* 666 * Stop profiling on a process. 667 */ 668 void 669 stopprofclock(p) 670 register struct proc *p; 671 { 672 673 PROC_LOCK_ASSERT(p, MA_OWNED); 674 if (p->p_flag & P_PROFIL) { 675 if (p->p_profthreads != 0) { 676 while (p->p_profthreads != 0) { 677 p->p_flag |= P_STOPPROF; 678 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 679 "stopprof", 0); 680 } 681 } 682 if ((p->p_flag & P_PROFIL) == 0) 683 return; 684 p->p_flag &= ~P_PROFIL; 685 mtx_lock(&time_lock); 686 if (--profprocs == 0) 687 cpu_stopprofclock(); 688 mtx_unlock(&time_lock); 689 } 690 } 691 692 /* 693 * Statistics clock. Updates rusage information and calls the scheduler 694 * to adjust priorities of the active thread. 695 * 696 * This should be called by all active processors. 697 */ 698 void 699 statclock(int usermode) 700 { 701 702 statclock_cnt(1, usermode); 703 } 704 705 void 706 statclock_cnt(int cnt, int usermode) 707 { 708 struct rusage *ru; 709 struct vmspace *vm; 710 struct thread *td; 711 struct proc *p; 712 long rss; 713 long *cp_time; 714 715 td = curthread; 716 p = td->td_proc; 717 718 cp_time = (long *)PCPU_PTR(cp_time); 719 if (usermode) { 720 /* 721 * Charge the time as appropriate. 722 */ 723 td->td_uticks += cnt; 724 if (p->p_nice > NZERO) 725 cp_time[CP_NICE] += cnt; 726 else 727 cp_time[CP_USER] += cnt; 728 } else { 729 /* 730 * Came from kernel mode, so we were: 731 * - handling an interrupt, 732 * - doing syscall or trap work on behalf of the current 733 * user process, or 734 * - spinning in the idle loop. 735 * Whichever it is, charge the time as appropriate. 736 * Note that we charge interrupts to the current process, 737 * regardless of whether they are ``for'' that process, 738 * so that we know how much of its real time was spent 739 * in ``non-process'' (i.e., interrupt) work. 740 */ 741 if ((td->td_pflags & TDP_ITHREAD) || 742 td->td_intr_nesting_level >= 2) { 743 td->td_iticks += cnt; 744 cp_time[CP_INTR] += cnt; 745 } else { 746 td->td_pticks += cnt; 747 td->td_sticks += cnt; 748 if (!TD_IS_IDLETHREAD(td)) 749 cp_time[CP_SYS] += cnt; 750 else 751 cp_time[CP_IDLE] += cnt; 752 } 753 } 754 755 /* Update resource usage integrals and maximums. */ 756 MPASS(p->p_vmspace != NULL); 757 vm = p->p_vmspace; 758 ru = &td->td_ru; 759 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt; 760 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt; 761 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt; 762 rss = pgtok(vmspace_resident_count(vm)); 763 if (ru->ru_maxrss < rss) 764 ru->ru_maxrss = rss; 765 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock", 766 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz); 767 SDT_PROBE2(sched, , , tick, td, td->td_proc); 768 thread_lock_flags(td, MTX_QUIET); 769 for ( ; cnt > 0; cnt--) 770 sched_clock(td); 771 thread_unlock(td); 772 #ifdef HWPMC_HOOKS 773 if (td->td_intr_frame != NULL) 774 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame); 775 #endif 776 } 777 778 void 779 profclock(int usermode, uintfptr_t pc) 780 { 781 782 profclock_cnt(1, usermode, pc); 783 } 784 785 void 786 profclock_cnt(int cnt, int usermode, uintfptr_t pc) 787 { 788 struct thread *td; 789 #ifdef GPROF 790 struct gmonparam *g; 791 uintfptr_t i; 792 #endif 793 794 td = curthread; 795 if (usermode) { 796 /* 797 * Came from user mode; CPU was in user state. 798 * If this process is being profiled, record the tick. 799 * if there is no related user location yet, don't 800 * bother trying to count it. 801 */ 802 if (td->td_proc->p_flag & P_PROFIL) 803 addupc_intr(td, pc, cnt); 804 } 805 #ifdef GPROF 806 else { 807 /* 808 * Kernel statistics are just like addupc_intr, only easier. 809 */ 810 g = &_gmonparam; 811 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 812 i = PC_TO_I(g, pc); 813 if (i < g->textsize) { 814 KCOUNT(g, i) += cnt; 815 } 816 } 817 } 818 #endif 819 #ifdef HWPMC_HOOKS 820 if (td->td_intr_frame != NULL) 821 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame); 822 #endif 823 } 824 825 /* 826 * Return information about system clocks. 827 */ 828 static int 829 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 830 { 831 struct clockinfo clkinfo; 832 /* 833 * Construct clockinfo structure. 834 */ 835 bzero(&clkinfo, sizeof(clkinfo)); 836 clkinfo.hz = hz; 837 clkinfo.tick = tick; 838 clkinfo.profhz = profhz; 839 clkinfo.stathz = stathz ? stathz : hz; 840 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 841 } 842 843 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, 844 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE, 845 0, 0, sysctl_kern_clockrate, "S,clockinfo", 846 "Rate and period of various kernel clocks"); 847 848 #ifdef SW_WATCHDOG 849 850 static void 851 watchdog_config(void *unused __unused, u_int cmd, int *error) 852 { 853 u_int u; 854 855 u = cmd & WD_INTERVAL; 856 if (u >= WD_TO_1SEC) { 857 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 858 watchdog_enabled = 1; 859 *error = 0; 860 } else { 861 watchdog_enabled = 0; 862 } 863 } 864 865 /* 866 * Handle a watchdog timeout by dumping interrupt information and 867 * then either dropping to DDB or panicking. 868 */ 869 static void 870 watchdog_fire(void) 871 { 872 int nintr; 873 uint64_t inttotal; 874 u_long *curintr; 875 char *curname; 876 877 curintr = intrcnt; 878 curname = intrnames; 879 inttotal = 0; 880 nintr = sintrcnt / sizeof(u_long); 881 882 printf("interrupt total\n"); 883 while (--nintr >= 0) { 884 if (*curintr) 885 printf("%-12s %20lu\n", curname, *curintr); 886 curname += strlen(curname) + 1; 887 inttotal += *curintr++; 888 } 889 printf("Total %20ju\n", (uintmax_t)inttotal); 890 891 #if defined(KDB) && !defined(KDB_UNATTENDED) 892 kdb_backtrace(); 893 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout"); 894 #else 895 panic("watchdog timeout"); 896 #endif 897 } 898 899 #endif /* SW_WATCHDOG */ 900