kern_clock.c revision 260817
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: stable/10/sys/kern/kern_clock.c 260817 2014-01-17 10:58:59Z avg $"); 39 40#include "opt_kdb.h" 41#include "opt_device_polling.h" 42#include "opt_hwpmc_hooks.h" 43#include "opt_kdtrace.h" 44#include "opt_ntp.h" 45#include "opt_watchdog.h" 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/callout.h> 50#include <sys/kdb.h> 51#include <sys/kernel.h> 52#include <sys/kthread.h> 53#include <sys/ktr.h> 54#include <sys/lock.h> 55#include <sys/mutex.h> 56#include <sys/proc.h> 57#include <sys/resource.h> 58#include <sys/resourcevar.h> 59#include <sys/sched.h> 60#include <sys/sdt.h> 61#include <sys/signalvar.h> 62#include <sys/sleepqueue.h> 63#include <sys/smp.h> 64#include <vm/vm.h> 65#include <vm/pmap.h> 66#include <vm/vm_map.h> 67#include <sys/sysctl.h> 68#include <sys/bus.h> 69#include <sys/interrupt.h> 70#include <sys/limits.h> 71#include <sys/timetc.h> 72 73#ifdef GPROF 74#include <sys/gmon.h> 75#endif 76 77#ifdef HWPMC_HOOKS 78#include <sys/pmckern.h> 79PMC_SOFT_DEFINE( , , clock, hard); 80PMC_SOFT_DEFINE( , , clock, stat); 81PMC_SOFT_DEFINE_EX( , , clock, prof, \ 82 cpu_startprofclock, cpu_stopprofclock); 83#endif 84 85#ifdef DEVICE_POLLING 86extern void hardclock_device_poll(void); 87#endif /* DEVICE_POLLING */ 88 89static void initclocks(void *dummy); 90SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL); 91 92/* Spin-lock protecting profiling statistics. */ 93static struct mtx time_lock; 94 95SDT_PROVIDER_DECLARE(sched); 96SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *"); 97 98static int 99sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS) 100{ 101 int error; 102 long cp_time[CPUSTATES]; 103#ifdef SCTL_MASK32 104 int i; 105 unsigned int cp_time32[CPUSTATES]; 106#endif 107 108 read_cpu_time(cp_time); 109#ifdef SCTL_MASK32 110 if (req->flags & SCTL_MASK32) { 111 if (!req->oldptr) 112 return SYSCTL_OUT(req, 0, sizeof(cp_time32)); 113 for (i = 0; i < CPUSTATES; i++) 114 cp_time32[i] = (unsigned int)cp_time[i]; 115 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 116 } else 117#endif 118 { 119 if (!req->oldptr) 120 return SYSCTL_OUT(req, 0, sizeof(cp_time)); 121 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time)); 122 } 123 return error; 124} 125 126SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 127 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics"); 128 129static long empty[CPUSTATES]; 130 131static int 132sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS) 133{ 134 struct pcpu *pcpu; 135 int error; 136 int c; 137 long *cp_time; 138#ifdef SCTL_MASK32 139 unsigned int cp_time32[CPUSTATES]; 140 int i; 141#endif 142 143 if (!req->oldptr) { 144#ifdef SCTL_MASK32 145 if (req->flags & SCTL_MASK32) 146 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1)); 147 else 148#endif 149 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1)); 150 } 151 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) { 152 if (!CPU_ABSENT(c)) { 153 pcpu = pcpu_find(c); 154 cp_time = pcpu->pc_cp_time; 155 } else { 156 cp_time = empty; 157 } 158#ifdef SCTL_MASK32 159 if (req->flags & SCTL_MASK32) { 160 for (i = 0; i < CPUSTATES; i++) 161 cp_time32[i] = (unsigned int)cp_time[i]; 162 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 163 } else 164#endif 165 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES); 166 } 167 return error; 168} 169 170SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 171 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics"); 172 173#ifdef DEADLKRES 174static const char *blessed[] = { 175 "getblk", 176 "so_snd_sx", 177 "so_rcv_sx", 178 NULL 179}; 180static int slptime_threshold = 1800; 181static int blktime_threshold = 900; 182static int sleepfreq = 3; 183 184static void 185deadlkres(void) 186{ 187 struct proc *p; 188 struct thread *td; 189 void *wchan; 190 int blkticks, i, slpticks, slptype, tryl, tticks; 191 192 tryl = 0; 193 for (;;) { 194 blkticks = blktime_threshold * hz; 195 slpticks = slptime_threshold * hz; 196 197 /* 198 * Avoid to sleep on the sx_lock in order to avoid a possible 199 * priority inversion problem leading to starvation. 200 * If the lock can't be held after 100 tries, panic. 201 */ 202 if (!sx_try_slock(&allproc_lock)) { 203 if (tryl > 100) 204 panic("%s: possible deadlock detected on allproc_lock\n", 205 __func__); 206 tryl++; 207 pause("allproc", sleepfreq * hz); 208 continue; 209 } 210 tryl = 0; 211 FOREACH_PROC_IN_SYSTEM(p) { 212 PROC_LOCK(p); 213 if (p->p_state == PRS_NEW) { 214 PROC_UNLOCK(p); 215 continue; 216 } 217 FOREACH_THREAD_IN_PROC(p, td) { 218 219 thread_lock(td); 220 if (TD_ON_LOCK(td)) { 221 222 /* 223 * The thread should be blocked on a 224 * turnstile, simply check if the 225 * turnstile channel is in good state. 226 */ 227 MPASS(td->td_blocked != NULL); 228 229 tticks = ticks - td->td_blktick; 230 thread_unlock(td); 231 if (tticks > blkticks) { 232 233 /* 234 * Accordingly with provided 235 * thresholds, this thread is 236 * stuck for too long on a 237 * turnstile. 238 */ 239 PROC_UNLOCK(p); 240 sx_sunlock(&allproc_lock); 241 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", 242 __func__, td, tticks); 243 } 244 } else if (TD_IS_SLEEPING(td) && 245 TD_ON_SLEEPQ(td)) { 246 247 /* 248 * Check if the thread is sleeping on a 249 * lock, otherwise skip the check. 250 * Drop the thread lock in order to 251 * avoid a LOR with the sleepqueue 252 * spinlock. 253 */ 254 wchan = td->td_wchan; 255 tticks = ticks - td->td_slptick; 256 thread_unlock(td); 257 slptype = sleepq_type(wchan); 258 if ((slptype == SLEEPQ_SX || 259 slptype == SLEEPQ_LK) && 260 tticks > slpticks) { 261 262 /* 263 * Accordingly with provided 264 * thresholds, this thread is 265 * stuck for too long on a 266 * sleepqueue. 267 * However, being on a 268 * sleepqueue, we might still 269 * check for the blessed 270 * list. 271 */ 272 tryl = 0; 273 for (i = 0; blessed[i] != NULL; 274 i++) { 275 if (!strcmp(blessed[i], 276 td->td_wmesg)) { 277 tryl = 1; 278 break; 279 } 280 } 281 if (tryl != 0) { 282 tryl = 0; 283 continue; 284 } 285 PROC_UNLOCK(p); 286 sx_sunlock(&allproc_lock); 287 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", 288 __func__, td, tticks); 289 } 290 } else 291 thread_unlock(td); 292 } 293 PROC_UNLOCK(p); 294 } 295 sx_sunlock(&allproc_lock); 296 297 /* Sleep for sleepfreq seconds. */ 298 pause("-", sleepfreq * hz); 299 } 300} 301 302static struct kthread_desc deadlkres_kd = { 303 "deadlkres", 304 deadlkres, 305 (struct thread **)NULL 306}; 307 308SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd); 309 310static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, 311 "Deadlock resolver"); 312SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW, 313 &slptime_threshold, 0, 314 "Number of seconds within is valid to sleep on a sleepqueue"); 315SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW, 316 &blktime_threshold, 0, 317 "Number of seconds within is valid to block on a turnstile"); 318SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0, 319 "Number of seconds between any deadlock resolver thread run"); 320#endif /* DEADLKRES */ 321 322void 323read_cpu_time(long *cp_time) 324{ 325 struct pcpu *pc; 326 int i, j; 327 328 /* Sum up global cp_time[]. */ 329 bzero(cp_time, sizeof(long) * CPUSTATES); 330 CPU_FOREACH(i) { 331 pc = pcpu_find(i); 332 for (j = 0; j < CPUSTATES; j++) 333 cp_time[j] += pc->pc_cp_time[j]; 334 } 335} 336 337#ifdef SW_WATCHDOG 338#include <sys/watchdog.h> 339 340static int watchdog_ticks; 341static int watchdog_enabled; 342static void watchdog_fire(void); 343static void watchdog_config(void *, u_int, int *); 344#endif /* SW_WATCHDOG */ 345 346/* 347 * Clock handling routines. 348 * 349 * This code is written to operate with two timers that run independently of 350 * each other. 351 * 352 * The main timer, running hz times per second, is used to trigger interval 353 * timers, timeouts and rescheduling as needed. 354 * 355 * The second timer handles kernel and user profiling, 356 * and does resource use estimation. If the second timer is programmable, 357 * it is randomized to avoid aliasing between the two clocks. For example, 358 * the randomization prevents an adversary from always giving up the cpu 359 * just before its quantum expires. Otherwise, it would never accumulate 360 * cpu ticks. The mean frequency of the second timer is stathz. 361 * 362 * If no second timer exists, stathz will be zero; in this case we drive 363 * profiling and statistics off the main clock. This WILL NOT be accurate; 364 * do not do it unless absolutely necessary. 365 * 366 * The statistics clock may (or may not) be run at a higher rate while 367 * profiling. This profile clock runs at profhz. We require that profhz 368 * be an integral multiple of stathz. 369 * 370 * If the statistics clock is running fast, it must be divided by the ratio 371 * profhz/stathz for statistics. (For profiling, every tick counts.) 372 * 373 * Time-of-day is maintained using a "timecounter", which may or may 374 * not be related to the hardware generating the above mentioned 375 * interrupts. 376 */ 377 378int stathz; 379int profhz; 380int profprocs; 381volatile int ticks; 382int psratio; 383 384static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */ 385static int global_hardclock_run = 0; 386 387/* 388 * Initialize clock frequencies and start both clocks running. 389 */ 390/* ARGSUSED*/ 391static void 392initclocks(dummy) 393 void *dummy; 394{ 395 register int i; 396 397 /* 398 * Set divisors to 1 (normal case) and let the machine-specific 399 * code do its bit. 400 */ 401 mtx_init(&time_lock, "time lock", NULL, MTX_DEF); 402 cpu_initclocks(); 403 404 /* 405 * Compute profhz/stathz, and fix profhz if needed. 406 */ 407 i = stathz ? stathz : hz; 408 if (profhz == 0) 409 profhz = i; 410 psratio = profhz / i; 411#ifdef SW_WATCHDOG 412 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 413#endif 414} 415 416/* 417 * Each time the real-time timer fires, this function is called on all CPUs. 418 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only 419 * the other CPUs in the system need to call this function. 420 */ 421void 422hardclock_cpu(int usermode) 423{ 424 struct pstats *pstats; 425 struct thread *td = curthread; 426 struct proc *p = td->td_proc; 427 int flags; 428 429 /* 430 * Run current process's virtual and profile time, as needed. 431 */ 432 pstats = p->p_stats; 433 flags = 0; 434 if (usermode && 435 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 436 PROC_SLOCK(p); 437 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 438 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 439 PROC_SUNLOCK(p); 440 } 441 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 442 PROC_SLOCK(p); 443 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 444 flags |= TDF_PROFPEND | TDF_ASTPENDING; 445 PROC_SUNLOCK(p); 446 } 447 thread_lock(td); 448 sched_tick(1); 449 td->td_flags |= flags; 450 thread_unlock(td); 451 452#ifdef HWPMC_HOOKS 453 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 454 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 455 if (td->td_intr_frame != NULL) 456 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 457#endif 458 callout_process(sbinuptime()); 459} 460 461/* 462 * The real-time timer, interrupting hz times per second. 463 */ 464void 465hardclock(int usermode, uintfptr_t pc) 466{ 467 468 atomic_add_int(&ticks, 1); 469 hardclock_cpu(usermode); 470 tc_ticktock(1); 471 cpu_tick_calibration(); 472 /* 473 * If no separate statistics clock is available, run it from here. 474 * 475 * XXX: this only works for UP 476 */ 477 if (stathz == 0) { 478 profclock(usermode, pc); 479 statclock(usermode); 480 } 481#ifdef DEVICE_POLLING 482 hardclock_device_poll(); /* this is very short and quick */ 483#endif /* DEVICE_POLLING */ 484#ifdef SW_WATCHDOG 485 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 486 watchdog_fire(); 487#endif /* SW_WATCHDOG */ 488} 489 490void 491hardclock_cnt(int cnt, int usermode) 492{ 493 struct pstats *pstats; 494 struct thread *td = curthread; 495 struct proc *p = td->td_proc; 496 int *t = DPCPU_PTR(pcputicks); 497 int flags, global, newticks; 498#ifdef SW_WATCHDOG 499 int i; 500#endif /* SW_WATCHDOG */ 501 502 /* 503 * Update per-CPU and possibly global ticks values. 504 */ 505 *t += cnt; 506 do { 507 global = ticks; 508 newticks = *t - global; 509 if (newticks <= 0) { 510 if (newticks < -1) 511 *t = global - 1; 512 newticks = 0; 513 break; 514 } 515 } while (!atomic_cmpset_int(&ticks, global, *t)); 516 517 /* 518 * Run current process's virtual and profile time, as needed. 519 */ 520 pstats = p->p_stats; 521 flags = 0; 522 if (usermode && 523 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 524 PROC_SLOCK(p); 525 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], 526 tick * cnt) == 0) 527 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 528 PROC_SUNLOCK(p); 529 } 530 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 531 PROC_SLOCK(p); 532 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], 533 tick * cnt) == 0) 534 flags |= TDF_PROFPEND | TDF_ASTPENDING; 535 PROC_SUNLOCK(p); 536 } 537 thread_lock(td); 538 sched_tick(cnt); 539 td->td_flags |= flags; 540 thread_unlock(td); 541 542#ifdef HWPMC_HOOKS 543 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 544 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 545 if (td->td_intr_frame != NULL) 546 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 547#endif 548 /* We are in charge to handle this tick duty. */ 549 if (newticks > 0) { 550 /* Dangerous and no need to call these things concurrently. */ 551 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) { 552 tc_ticktock(newticks); 553#ifdef DEVICE_POLLING 554 /* This is very short and quick. */ 555 hardclock_device_poll(); 556#endif /* DEVICE_POLLING */ 557 atomic_store_rel_int(&global_hardclock_run, 0); 558 } 559#ifdef SW_WATCHDOG 560 if (watchdog_enabled > 0) { 561 i = atomic_fetchadd_int(&watchdog_ticks, -newticks); 562 if (i > 0 && i <= newticks) 563 watchdog_fire(); 564 } 565#endif /* SW_WATCHDOG */ 566 } 567 if (curcpu == CPU_FIRST()) 568 cpu_tick_calibration(); 569} 570 571void 572hardclock_sync(int cpu) 573{ 574 int *t = DPCPU_ID_PTR(cpu, pcputicks); 575 576 *t = ticks; 577} 578 579/* 580 * Compute number of ticks in the specified amount of time. 581 */ 582int 583tvtohz(tv) 584 struct timeval *tv; 585{ 586 register unsigned long ticks; 587 register long sec, usec; 588 589 /* 590 * If the number of usecs in the whole seconds part of the time 591 * difference fits in a long, then the total number of usecs will 592 * fit in an unsigned long. Compute the total and convert it to 593 * ticks, rounding up and adding 1 to allow for the current tick 594 * to expire. Rounding also depends on unsigned long arithmetic 595 * to avoid overflow. 596 * 597 * Otherwise, if the number of ticks in the whole seconds part of 598 * the time difference fits in a long, then convert the parts to 599 * ticks separately and add, using similar rounding methods and 600 * overflow avoidance. This method would work in the previous 601 * case but it is slightly slower and assumes that hz is integral. 602 * 603 * Otherwise, round the time difference down to the maximum 604 * representable value. 605 * 606 * If ints have 32 bits, then the maximum value for any timeout in 607 * 10ms ticks is 248 days. 608 */ 609 sec = tv->tv_sec; 610 usec = tv->tv_usec; 611 if (usec < 0) { 612 sec--; 613 usec += 1000000; 614 } 615 if (sec < 0) { 616#ifdef DIAGNOSTIC 617 if (usec > 0) { 618 sec++; 619 usec -= 1000000; 620 } 621 printf("tvotohz: negative time difference %ld sec %ld usec\n", 622 sec, usec); 623#endif 624 ticks = 1; 625 } else if (sec <= LONG_MAX / 1000000) 626 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 627 / tick + 1; 628 else if (sec <= LONG_MAX / hz) 629 ticks = sec * hz 630 + ((unsigned long)usec + (tick - 1)) / tick + 1; 631 else 632 ticks = LONG_MAX; 633 if (ticks > INT_MAX) 634 ticks = INT_MAX; 635 return ((int)ticks); 636} 637 638/* 639 * Start profiling on a process. 640 * 641 * Kernel profiling passes proc0 which never exits and hence 642 * keeps the profile clock running constantly. 643 */ 644void 645startprofclock(p) 646 register struct proc *p; 647{ 648 649 PROC_LOCK_ASSERT(p, MA_OWNED); 650 if (p->p_flag & P_STOPPROF) 651 return; 652 if ((p->p_flag & P_PROFIL) == 0) { 653 p->p_flag |= P_PROFIL; 654 mtx_lock(&time_lock); 655 if (++profprocs == 1) 656 cpu_startprofclock(); 657 mtx_unlock(&time_lock); 658 } 659} 660 661/* 662 * Stop profiling on a process. 663 */ 664void 665stopprofclock(p) 666 register 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 p->p_flag |= P_STOPPROF; 673 while (p->p_profthreads != 0) 674 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 675 "stopprof", 0); 676 p->p_flag &= ~P_STOPPROF; 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 */ 694void 695statclock(int usermode) 696{ 697 698 statclock_cnt(1, usermode); 699} 700 701void 702statclock_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 774void 775profclock(int usermode, uintfptr_t pc) 776{ 777 778 profclock_cnt(1, usermode, pc); 779} 780 781void 782profclock_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 */ 824static int 825sysctl_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 839SYSCTL_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 846static void 847watchdog_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 */ 865static void 866watchdog_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