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