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 #ifdef EARLY_AP_STARTUP 395 struct proc *p; 396 struct thread *td; 397 #endif 398 register int i; 399 400 /* 401 * Set divisors to 1 (normal case) and let the machine-specific 402 * code do its bit. 403 */ 404 mtx_init(&time_lock, "time lock", NULL, MTX_DEF); 405 cpu_initclocks(); 406 407 /* 408 * Compute profhz/stathz, and fix profhz if needed. 409 */ 410 i = stathz ? stathz : hz; 411 if (profhz == 0) 412 profhz = i; 413 psratio = profhz / i; 414 #ifdef SW_WATCHDOG 415 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 416 #endif 417 /* 418 * Arrange for ticks to wrap 10 minutes after boot to help catch 419 * sign problems sooner. 420 */ 421 ticks = INT_MAX - (hz * 10 * 60); 422 423 #ifdef EARLY_AP_STARTUP 424 /* 425 * Fixup the tick counts in any blocked or sleeping threads to 426 * account for the jump above. 427 */ 428 sx_slock(&allproc_lock); 429 FOREACH_PROC_IN_SYSTEM(p) { 430 PROC_LOCK(p); 431 if (p->p_state == PRS_NEW) { 432 PROC_UNLOCK(p); 433 continue; 434 } 435 FOREACH_THREAD_IN_PROC(p, td) { 436 thread_lock(td); 437 if (TD_ON_LOCK(td)) { 438 MPASS(td->td_blktick == 0); 439 td->td_blktick = ticks; 440 } 441 if (TD_ON_SLEEPQ(td)) { 442 MPASS(td->td_slptick == 0); 443 td->td_slptick = ticks; 444 } 445 thread_unlock(td); 446 } 447 PROC_UNLOCK(p); 448 } 449 sx_sunlock(&allproc_lock); 450 #endif 451 } 452 453 /* 454 * Each time the real-time timer fires, this function is called on all CPUs. 455 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only 456 * the other CPUs in the system need to call this function. 457 */ 458 void 459 hardclock_cpu(int usermode) 460 { 461 struct pstats *pstats; 462 struct thread *td = curthread; 463 struct proc *p = td->td_proc; 464 int flags; 465 466 /* 467 * Run current process's virtual and profile time, as needed. 468 */ 469 pstats = p->p_stats; 470 flags = 0; 471 if (usermode && 472 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 473 PROC_ITIMLOCK(p); 474 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 475 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 476 PROC_ITIMUNLOCK(p); 477 } 478 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 479 PROC_ITIMLOCK(p); 480 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 481 flags |= TDF_PROFPEND | TDF_ASTPENDING; 482 PROC_ITIMUNLOCK(p); 483 } 484 thread_lock(td); 485 td->td_flags |= flags; 486 thread_unlock(td); 487 488 #ifdef HWPMC_HOOKS 489 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 490 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 491 if (td->td_intr_frame != NULL) 492 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 493 #endif 494 callout_process(sbinuptime()); 495 } 496 497 /* 498 * The real-time timer, interrupting hz times per second. 499 */ 500 void 501 hardclock(int usermode, uintfptr_t pc) 502 { 503 504 atomic_add_int(&ticks, 1); 505 hardclock_cpu(usermode); 506 tc_ticktock(1); 507 cpu_tick_calibration(); 508 /* 509 * If no separate statistics clock is available, run it from here. 510 * 511 * XXX: this only works for UP 512 */ 513 if (stathz == 0) { 514 profclock(usermode, pc); 515 statclock(usermode); 516 } 517 #ifdef DEVICE_POLLING 518 hardclock_device_poll(); /* this is very short and quick */ 519 #endif /* DEVICE_POLLING */ 520 #ifdef SW_WATCHDOG 521 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 522 watchdog_fire(); 523 #endif /* SW_WATCHDOG */ 524 } 525 526 void 527 hardclock_cnt(int cnt, int usermode) 528 { 529 struct pstats *pstats; 530 struct thread *td = curthread; 531 struct proc *p = td->td_proc; 532 int *t = DPCPU_PTR(pcputicks); 533 int flags, global, newticks; 534 #ifdef SW_WATCHDOG 535 int i; 536 #endif /* SW_WATCHDOG */ 537 538 /* 539 * Update per-CPU and possibly global ticks values. 540 */ 541 *t += cnt; 542 do { 543 global = ticks; 544 newticks = *t - global; 545 if (newticks <= 0) { 546 if (newticks < -1) 547 *t = global - 1; 548 newticks = 0; 549 break; 550 } 551 } while (!atomic_cmpset_int(&ticks, global, *t)); 552 553 /* 554 * Run current process's virtual and profile time, as needed. 555 */ 556 pstats = p->p_stats; 557 flags = 0; 558 if (usermode && 559 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 560 PROC_ITIMLOCK(p); 561 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], 562 tick * cnt) == 0) 563 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 564 PROC_ITIMUNLOCK(p); 565 } 566 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 567 PROC_ITIMLOCK(p); 568 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], 569 tick * cnt) == 0) 570 flags |= TDF_PROFPEND | TDF_ASTPENDING; 571 PROC_ITIMUNLOCK(p); 572 } 573 if (flags != 0) { 574 thread_lock(td); 575 td->td_flags |= flags; 576 thread_unlock(td); 577 } 578 579 #ifdef HWPMC_HOOKS 580 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 581 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 582 if (td->td_intr_frame != NULL) 583 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 584 #endif 585 /* We are in charge to handle this tick duty. */ 586 if (newticks > 0) { 587 /* Dangerous and no need to call these things concurrently. */ 588 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) { 589 tc_ticktock(newticks); 590 #ifdef DEVICE_POLLING 591 /* This is very short and quick. */ 592 hardclock_device_poll(); 593 #endif /* DEVICE_POLLING */ 594 atomic_store_rel_int(&global_hardclock_run, 0); 595 } 596 #ifdef SW_WATCHDOG 597 if (watchdog_enabled > 0) { 598 i = atomic_fetchadd_int(&watchdog_ticks, -newticks); 599 if (i > 0 && i <= newticks) 600 watchdog_fire(); 601 } 602 #endif /* SW_WATCHDOG */ 603 } 604 if (curcpu == CPU_FIRST()) 605 cpu_tick_calibration(); 606 } 607 608 void 609 hardclock_sync(int cpu) 610 { 611 int *t = DPCPU_ID_PTR(cpu, pcputicks); 612 613 *t = ticks; 614 } 615 616 /* 617 * Compute number of ticks in the specified amount of time. 618 */ 619 int 620 tvtohz(tv) 621 struct timeval *tv; 622 { 623 register unsigned long ticks; 624 register long sec, usec; 625 626 /* 627 * If the number of usecs in the whole seconds part of the time 628 * difference fits in a long, then the total number of usecs will 629 * fit in an unsigned long. Compute the total and convert it to 630 * ticks, rounding up and adding 1 to allow for the current tick 631 * to expire. Rounding also depends on unsigned long arithmetic 632 * to avoid overflow. 633 * 634 * Otherwise, if the number of ticks in the whole seconds part of 635 * the time difference fits in a long, then convert the parts to 636 * ticks separately and add, using similar rounding methods and 637 * overflow avoidance. This method would work in the previous 638 * case but it is slightly slower and assumes that hz is integral. 639 * 640 * Otherwise, round the time difference down to the maximum 641 * representable value. 642 * 643 * If ints have 32 bits, then the maximum value for any timeout in 644 * 10ms ticks is 248 days. 645 */ 646 sec = tv->tv_sec; 647 usec = tv->tv_usec; 648 if (usec < 0) { 649 sec--; 650 usec += 1000000; 651 } 652 if (sec < 0) { 653 #ifdef DIAGNOSTIC 654 if (usec > 0) { 655 sec++; 656 usec -= 1000000; 657 } 658 printf("tvotohz: negative time difference %ld sec %ld usec\n", 659 sec, usec); 660 #endif 661 ticks = 1; 662 } else if (sec <= LONG_MAX / 1000000) 663 ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1; 664 else if (sec <= LONG_MAX / hz) 665 ticks = sec * hz 666 + howmany((unsigned long)usec, tick) + 1; 667 else 668 ticks = LONG_MAX; 669 if (ticks > INT_MAX) 670 ticks = INT_MAX; 671 return ((int)ticks); 672 } 673 674 /* 675 * Start profiling on a process. 676 * 677 * Kernel profiling passes proc0 which never exits and hence 678 * keeps the profile clock running constantly. 679 */ 680 void 681 startprofclock(p) 682 register struct proc *p; 683 { 684 685 PROC_LOCK_ASSERT(p, MA_OWNED); 686 if (p->p_flag & P_STOPPROF) 687 return; 688 if ((p->p_flag & P_PROFIL) == 0) { 689 p->p_flag |= P_PROFIL; 690 mtx_lock(&time_lock); 691 if (++profprocs == 1) 692 cpu_startprofclock(); 693 mtx_unlock(&time_lock); 694 } 695 } 696 697 /* 698 * Stop profiling on a process. 699 */ 700 void 701 stopprofclock(p) 702 register struct proc *p; 703 { 704 705 PROC_LOCK_ASSERT(p, MA_OWNED); 706 if (p->p_flag & P_PROFIL) { 707 if (p->p_profthreads != 0) { 708 while (p->p_profthreads != 0) { 709 p->p_flag |= P_STOPPROF; 710 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 711 "stopprof", 0); 712 } 713 } 714 if ((p->p_flag & P_PROFIL) == 0) 715 return; 716 p->p_flag &= ~P_PROFIL; 717 mtx_lock(&time_lock); 718 if (--profprocs == 0) 719 cpu_stopprofclock(); 720 mtx_unlock(&time_lock); 721 } 722 } 723 724 /* 725 * Statistics clock. Updates rusage information and calls the scheduler 726 * to adjust priorities of the active thread. 727 * 728 * This should be called by all active processors. 729 */ 730 void 731 statclock(int usermode) 732 { 733 734 statclock_cnt(1, usermode); 735 } 736 737 void 738 statclock_cnt(int cnt, int usermode) 739 { 740 struct rusage *ru; 741 struct vmspace *vm; 742 struct thread *td; 743 struct proc *p; 744 long rss; 745 long *cp_time; 746 747 td = curthread; 748 p = td->td_proc; 749 750 cp_time = (long *)PCPU_PTR(cp_time); 751 if (usermode) { 752 /* 753 * Charge the time as appropriate. 754 */ 755 td->td_uticks += cnt; 756 if (p->p_nice > NZERO) 757 cp_time[CP_NICE] += cnt; 758 else 759 cp_time[CP_USER] += cnt; 760 } else { 761 /* 762 * Came from kernel mode, so we were: 763 * - handling an interrupt, 764 * - doing syscall or trap work on behalf of the current 765 * user process, or 766 * - spinning in the idle loop. 767 * Whichever it is, charge the time as appropriate. 768 * Note that we charge interrupts to the current process, 769 * regardless of whether they are ``for'' that process, 770 * so that we know how much of its real time was spent 771 * in ``non-process'' (i.e., interrupt) work. 772 */ 773 if ((td->td_pflags & TDP_ITHREAD) || 774 td->td_intr_nesting_level >= 2) { 775 td->td_iticks += cnt; 776 cp_time[CP_INTR] += cnt; 777 } else { 778 td->td_pticks += cnt; 779 td->td_sticks += cnt; 780 if (!TD_IS_IDLETHREAD(td)) 781 cp_time[CP_SYS] += cnt; 782 else 783 cp_time[CP_IDLE] += cnt; 784 } 785 } 786 787 /* Update resource usage integrals and maximums. */ 788 MPASS(p->p_vmspace != NULL); 789 vm = p->p_vmspace; 790 ru = &td->td_ru; 791 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt; 792 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt; 793 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt; 794 rss = pgtok(vmspace_resident_count(vm)); 795 if (ru->ru_maxrss < rss) 796 ru->ru_maxrss = rss; 797 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock", 798 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz); 799 SDT_PROBE2(sched, , , tick, td, td->td_proc); 800 thread_lock_flags(td, MTX_QUIET); 801 for ( ; cnt > 0; cnt--) 802 sched_clock(td); 803 thread_unlock(td); 804 #ifdef HWPMC_HOOKS 805 if (td->td_intr_frame != NULL) 806 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame); 807 #endif 808 } 809 810 void 811 profclock(int usermode, uintfptr_t pc) 812 { 813 814 profclock_cnt(1, usermode, pc); 815 } 816 817 void 818 profclock_cnt(int cnt, int usermode, uintfptr_t pc) 819 { 820 struct thread *td; 821 #ifdef GPROF 822 struct gmonparam *g; 823 uintfptr_t i; 824 #endif 825 826 td = curthread; 827 if (usermode) { 828 /* 829 * Came from user mode; CPU was in user state. 830 * If this process is being profiled, record the tick. 831 * if there is no related user location yet, don't 832 * bother trying to count it. 833 */ 834 if (td->td_proc->p_flag & P_PROFIL) 835 addupc_intr(td, pc, cnt); 836 } 837 #ifdef GPROF 838 else { 839 /* 840 * Kernel statistics are just like addupc_intr, only easier. 841 */ 842 g = &_gmonparam; 843 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 844 i = PC_TO_I(g, pc); 845 if (i < g->textsize) { 846 KCOUNT(g, i) += cnt; 847 } 848 } 849 } 850 #endif 851 #ifdef HWPMC_HOOKS 852 if (td->td_intr_frame != NULL) 853 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame); 854 #endif 855 } 856 857 /* 858 * Return information about system clocks. 859 */ 860 static int 861 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 862 { 863 struct clockinfo clkinfo; 864 /* 865 * Construct clockinfo structure. 866 */ 867 bzero(&clkinfo, sizeof(clkinfo)); 868 clkinfo.hz = hz; 869 clkinfo.tick = tick; 870 clkinfo.profhz = profhz; 871 clkinfo.stathz = stathz ? stathz : hz; 872 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 873 } 874 875 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, 876 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE, 877 0, 0, sysctl_kern_clockrate, "S,clockinfo", 878 "Rate and period of various kernel clocks"); 879 880 #ifdef SW_WATCHDOG 881 882 static void 883 watchdog_config(void *unused __unused, u_int cmd, int *error) 884 { 885 u_int u; 886 887 u = cmd & WD_INTERVAL; 888 if (u >= WD_TO_1SEC) { 889 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 890 watchdog_enabled = 1; 891 *error = 0; 892 } else { 893 watchdog_enabled = 0; 894 } 895 } 896 897 /* 898 * Handle a watchdog timeout by dumping interrupt information and 899 * then either dropping to DDB or panicking. 900 */ 901 static void 902 watchdog_fire(void) 903 { 904 int nintr; 905 uint64_t inttotal; 906 u_long *curintr; 907 char *curname; 908 909 curintr = intrcnt; 910 curname = intrnames; 911 inttotal = 0; 912 nintr = sintrcnt / sizeof(u_long); 913 914 printf("interrupt total\n"); 915 while (--nintr >= 0) { 916 if (*curintr) 917 printf("%-12s %20lu\n", curname, *curintr); 918 curname += strlen(curname) + 1; 919 inttotal += *curintr++; 920 } 921 printf("Total %20ju\n", (uintmax_t)inttotal); 922 923 #if defined(KDB) && !defined(KDB_UNATTENDED) 924 kdb_backtrace(); 925 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout"); 926 #else 927 panic("watchdog timeout"); 928 #endif 929 } 930 931 #endif /* SW_WATCHDOG */ 932