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/epoch.h> 52 #include <sys/eventhandler.h> 53 #include <sys/gtaskqueue.h> 54 #include <sys/kdb.h> 55 #include <sys/kernel.h> 56 #include <sys/kthread.h> 57 #include <sys/ktr.h> 58 #include <sys/lock.h> 59 #include <sys/mutex.h> 60 #include <sys/proc.h> 61 #include <sys/resource.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sched.h> 64 #include <sys/sdt.h> 65 #include <sys/signalvar.h> 66 #include <sys/sleepqueue.h> 67 #include <sys/smp.h> 68 #include <vm/vm.h> 69 #include <vm/pmap.h> 70 #include <vm/vm_map.h> 71 #include <sys/sysctl.h> 72 #include <sys/bus.h> 73 #include <sys/interrupt.h> 74 #include <sys/limits.h> 75 #include <sys/timetc.h> 76 77 #ifdef GPROF 78 #include <sys/gmon.h> 79 #endif 80 81 #ifdef HWPMC_HOOKS 82 #include <sys/pmckern.h> 83 PMC_SOFT_DEFINE( , , clock, hard); 84 PMC_SOFT_DEFINE( , , clock, stat); 85 PMC_SOFT_DEFINE_EX( , , clock, prof, \ 86 cpu_startprofclock, cpu_stopprofclock); 87 #endif 88 89 #ifdef DEVICE_POLLING 90 extern void hardclock_device_poll(void); 91 #endif /* DEVICE_POLLING */ 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 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks) 187 { 188 int tticks; 189 190 sx_assert(&allproc_lock, SX_LOCKED); 191 PROC_LOCK_ASSERT(p, MA_OWNED); 192 THREAD_LOCK_ASSERT(td, MA_OWNED); 193 /* 194 * The thread should be blocked on a turnstile, simply check 195 * if the turnstile channel is in good state. 196 */ 197 MPASS(td->td_blocked != NULL); 198 199 tticks = ticks - td->td_blktick; 200 if (tticks > blkticks) 201 /* 202 * Accordingly with provided thresholds, this thread is stuck 203 * for too long on a turnstile. 204 */ 205 panic("%s: possible deadlock detected for %p (%s), " 206 "blocked for %d ticks\n", __func__, 207 td, sched_tdname(td), tticks); 208 } 209 210 static void 211 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks) 212 { 213 const void *wchan; 214 int i, slptype, tticks; 215 216 sx_assert(&allproc_lock, SX_LOCKED); 217 PROC_LOCK_ASSERT(p, MA_OWNED); 218 THREAD_LOCK_ASSERT(td, MA_OWNED); 219 /* 220 * Check if the thread is sleeping on a lock, otherwise skip the check. 221 * Drop the thread lock in order to avoid a LOR with the sleepqueue 222 * spinlock. 223 */ 224 wchan = td->td_wchan; 225 tticks = ticks - td->td_slptick; 226 slptype = sleepq_type(wchan); 227 if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) && 228 tticks > slpticks) { 229 /* 230 * Accordingly with provided thresholds, this thread is stuck 231 * for too long on a sleepqueue. 232 * However, being on a sleepqueue, we might still check for the 233 * blessed list. 234 */ 235 for (i = 0; blessed[i] != NULL; i++) 236 if (!strcmp(blessed[i], td->td_wmesg)) 237 return; 238 239 panic("%s: possible deadlock detected for %p (%s), " 240 "blocked for %d ticks\n", __func__, 241 td, sched_tdname(td), tticks); 242 } 243 } 244 245 static void 246 deadlkres(void) 247 { 248 struct proc *p; 249 struct thread *td; 250 int blkticks, slpticks, tryl; 251 252 tryl = 0; 253 for (;;) { 254 blkticks = blktime_threshold * hz; 255 slpticks = slptime_threshold * hz; 256 257 /* 258 * Avoid to sleep on the sx_lock in order to avoid a 259 * possible priority inversion problem leading to 260 * starvation. 261 * If the lock can't be held after 100 tries, panic. 262 */ 263 if (!sx_try_slock(&allproc_lock)) { 264 if (tryl > 100) 265 panic("%s: possible deadlock detected " 266 "on allproc_lock\n", __func__); 267 tryl++; 268 pause("allproc", sleepfreq * hz); 269 continue; 270 } 271 tryl = 0; 272 FOREACH_PROC_IN_SYSTEM(p) { 273 PROC_LOCK(p); 274 if (p->p_state == PRS_NEW) { 275 PROC_UNLOCK(p); 276 continue; 277 } 278 FOREACH_THREAD_IN_PROC(p, td) { 279 thread_lock(td); 280 if (TD_ON_LOCK(td)) 281 deadlres_td_on_lock(p, td, 282 blkticks); 283 else if (TD_IS_SLEEPING(td)) 284 deadlres_td_sleep_q(p, td, 285 slpticks); 286 thread_unlock(td); 287 } 288 PROC_UNLOCK(p); 289 } 290 sx_sunlock(&allproc_lock); 291 292 /* Sleep for sleepfreq seconds. */ 293 pause("-", sleepfreq * hz); 294 } 295 } 296 297 static struct kthread_desc deadlkres_kd = { 298 "deadlkres", 299 deadlkres, 300 (struct thread **)NULL 301 }; 302 303 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd); 304 305 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 306 "Deadlock resolver"); 307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW, 308 &slptime_threshold, 0, 309 "Number of seconds within is valid to sleep on a sleepqueue"); 310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW, 311 &blktime_threshold, 0, 312 "Number of seconds within is valid to block on a turnstile"); 313 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0, 314 "Number of seconds between any deadlock resolver thread run"); 315 #endif /* DEADLKRES */ 316 317 void 318 read_cpu_time(long *cp_time) 319 { 320 struct pcpu *pc; 321 int i, j; 322 323 /* Sum up global cp_time[]. */ 324 bzero(cp_time, sizeof(long) * CPUSTATES); 325 CPU_FOREACH(i) { 326 pc = pcpu_find(i); 327 for (j = 0; j < CPUSTATES; j++) 328 cp_time[j] += pc->pc_cp_time[j]; 329 } 330 } 331 332 #include <sys/watchdog.h> 333 334 static int watchdog_ticks; 335 static int watchdog_enabled; 336 static void watchdog_fire(void); 337 static void watchdog_config(void *, u_int, int *); 338 339 static void 340 watchdog_attach(void) 341 { 342 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 343 } 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 DPCPU_DEFINE_STATIC(int, pcputicks); /* Per-CPU version of ticks. */ 384 #ifdef DEVICE_POLLING 385 static int devpoll_run = 0; 386 #endif 387 388 static void 389 ast_oweupc(struct thread *td, int tda __unused) 390 { 391 if ((td->td_proc->p_flag & P_PROFIL) == 0) 392 return; 393 addupc_task(td, td->td_profil_addr, td->td_profil_ticks); 394 td->td_profil_ticks = 0; 395 td->td_pflags &= ~TDP_OWEUPC; 396 } 397 398 static void 399 ast_alrm(struct thread *td, int tda __unused) 400 { 401 struct proc *p; 402 403 p = td->td_proc; 404 PROC_LOCK(p); 405 kern_psignal(p, SIGVTALRM); 406 PROC_UNLOCK(p); 407 } 408 409 static void 410 ast_prof(struct thread *td, int tda __unused) 411 { 412 struct proc *p; 413 414 p = td->td_proc; 415 PROC_LOCK(p); 416 kern_psignal(p, SIGPROF); 417 PROC_UNLOCK(p); 418 } 419 420 /* 421 * Initialize clock frequencies and start both clocks running. 422 */ 423 static void 424 initclocks(void *dummy __unused) 425 { 426 int i; 427 428 /* 429 * Set divisors to 1 (normal case) and let the machine-specific 430 * code do its bit. 431 */ 432 mtx_init(&time_lock, "time lock", NULL, MTX_DEF); 433 cpu_initclocks(); 434 435 /* 436 * Compute profhz/stathz, and fix profhz if needed. 437 */ 438 i = stathz ? stathz : hz; 439 if (profhz == 0) 440 profhz = i; 441 psratio = profhz / i; 442 443 ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc); 444 ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm); 445 ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof); 446 447 #ifdef SW_WATCHDOG 448 /* Enable hardclock watchdog now, even if a hardware watchdog exists. */ 449 watchdog_attach(); 450 #else 451 /* Volunteer to run a software watchdog. */ 452 if (wdog_software_attach == NULL) 453 wdog_software_attach = watchdog_attach; 454 #endif 455 } 456 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL); 457 458 static __noinline void 459 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode) 460 { 461 struct proc *p; 462 int ast; 463 464 ast = 0; 465 p = td->td_proc; 466 if (usermode && 467 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 468 PROC_ITIMLOCK(p); 469 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], 470 tick * cnt) == 0) 471 ast |= TDAI(TDA_ALRM); 472 PROC_ITIMUNLOCK(p); 473 } 474 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 475 PROC_ITIMLOCK(p); 476 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], 477 tick * cnt) == 0) 478 ast |= TDAI(TDA_PROF); 479 PROC_ITIMUNLOCK(p); 480 } 481 if (ast != 0) 482 ast_sched_mask(td, ast); 483 } 484 485 void 486 hardclock(int cnt, int usermode) 487 { 488 struct pstats *pstats; 489 struct thread *td = curthread; 490 struct proc *p = td->td_proc; 491 int *t = DPCPU_PTR(pcputicks); 492 int global, i, newticks; 493 494 /* 495 * Update per-CPU and possibly global ticks values. 496 */ 497 *t += cnt; 498 global = ticks; 499 do { 500 newticks = *t - global; 501 if (newticks <= 0) { 502 if (newticks < -1) 503 *t = global - 1; 504 newticks = 0; 505 break; 506 } 507 } while (!atomic_fcmpset_int(&ticks, &global, *t)); 508 509 /* 510 * Run current process's virtual and profile time, as needed. 511 */ 512 pstats = p->p_stats; 513 if (__predict_false( 514 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) || 515 timevalisset(&pstats->p_timer[ITIMER_PROF].it_value))) 516 hardclock_itimer(td, pstats, cnt, usermode); 517 518 #ifdef HWPMC_HOOKS 519 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 520 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 521 if (td->td_intr_frame != NULL) 522 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); 523 #endif 524 /* We are in charge to handle this tick duty. */ 525 if (newticks > 0) { 526 tc_ticktock(newticks); 527 #ifdef DEVICE_POLLING 528 /* Dangerous and no need to call these things concurrently. */ 529 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) { 530 /* This is very short and quick. */ 531 hardclock_device_poll(); 532 atomic_store_rel_int(&devpoll_run, 0); 533 } 534 #endif /* DEVICE_POLLING */ 535 if (watchdog_enabled > 0) { 536 i = atomic_fetchadd_int(&watchdog_ticks, -newticks); 537 if (i > 0 && i <= newticks) 538 watchdog_fire(); 539 } 540 intr_event_handle(clk_intr_event, NULL); 541 } 542 if (curcpu == CPU_FIRST()) 543 cpu_tick_calibration(); 544 if (__predict_false(DPCPU_GET(epoch_cb_count))) 545 GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task)); 546 } 547 548 void 549 hardclock_sync(int cpu) 550 { 551 int *t; 552 KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu)); 553 t = DPCPU_ID_PTR(cpu, pcputicks); 554 555 *t = ticks; 556 } 557 558 /* 559 * Regular integer scaling formula without losing precision: 560 */ 561 #define TIME_INT_SCALE(value, mul, div) \ 562 (((value) / (div)) * (mul) + (((value) % (div)) * (mul)) / (div)) 563 564 /* 565 * Macro for converting seconds and microseconds into actual ticks, 566 * based on the given hz value: 567 */ 568 #define TIME_TO_TICKS(sec, usec, hz) \ 569 ((sec) * (hz) + TIME_INT_SCALE(usec, hz, 1 << 6) / (1000000 >> 6)) 570 571 #define TIME_ASSERT_VALID_HZ(hz) \ 572 _Static_assert(TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) >= 0 && \ 573 TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) < INT_MAX, \ 574 "tvtohz() can overflow the regular integer type") 575 576 /* 577 * Compile time assert the maximum and minimum values to fit into a 578 * regular integer when computing TIME_TO_TICKS(): 579 */ 580 TIME_ASSERT_VALID_HZ(HZ_MAXIMUM); 581 TIME_ASSERT_VALID_HZ(HZ_MINIMUM); 582 583 /* 584 * The formula is mostly linear, but test some more common values just 585 * in case: 586 */ 587 TIME_ASSERT_VALID_HZ(1024); 588 TIME_ASSERT_VALID_HZ(1000); 589 TIME_ASSERT_VALID_HZ(128); 590 TIME_ASSERT_VALID_HZ(100); 591 592 /* 593 * Compute number of ticks representing the specified amount of time. 594 * If the specified time is negative, a value of 1 is returned. This 595 * function returns a value from 1 up to and including INT_MAX. 596 */ 597 int 598 tvtohz(struct timeval *tv) 599 { 600 int retval; 601 602 /* 603 * The values passed here may come from user-space and these 604 * checks ensure "tv_usec" is within its allowed range: 605 */ 606 607 /* check for tv_usec underflow */ 608 if (__predict_false(tv->tv_usec < 0)) { 609 tv->tv_sec += tv->tv_usec / 1000000; 610 tv->tv_usec = tv->tv_usec % 1000000; 611 /* convert tv_usec to a positive value */ 612 if (__predict_true(tv->tv_usec < 0)) { 613 tv->tv_usec += 1000000; 614 tv->tv_sec -= 1; 615 } 616 /* check for tv_usec overflow */ 617 } else if (__predict_false(tv->tv_usec >= 1000000)) { 618 tv->tv_sec += tv->tv_usec / 1000000; 619 tv->tv_usec = tv->tv_usec % 1000000; 620 } 621 622 /* check for tv_sec underflow */ 623 if (__predict_false(tv->tv_sec < 0)) 624 return (1); 625 /* check for tv_sec overflow (including room for the tv_usec part) */ 626 else if (__predict_false(tv->tv_sec >= tick_seconds_max)) 627 return (INT_MAX); 628 629 /* cast to "int" to avoid platform differences */ 630 retval = TIME_TO_TICKS((int)tv->tv_sec, (int)tv->tv_usec, hz); 631 632 /* add one additional tick */ 633 return (retval + 1); 634 } 635 636 /* 637 * Start profiling on a process. 638 * 639 * Kernel profiling passes proc0 which never exits and hence 640 * keeps the profile clock running constantly. 641 */ 642 void 643 startprofclock(struct proc *p) 644 { 645 646 PROC_LOCK_ASSERT(p, MA_OWNED); 647 if (p->p_flag & P_STOPPROF) 648 return; 649 if ((p->p_flag & P_PROFIL) == 0) { 650 p->p_flag |= P_PROFIL; 651 mtx_lock(&time_lock); 652 if (++profprocs == 1) 653 cpu_startprofclock(); 654 mtx_unlock(&time_lock); 655 } 656 } 657 658 /* 659 * Stop profiling on a process. 660 */ 661 void 662 stopprofclock(struct proc *p) 663 { 664 665 PROC_LOCK_ASSERT(p, MA_OWNED); 666 if (p->p_flag & P_PROFIL) { 667 if (p->p_profthreads != 0) { 668 while (p->p_profthreads != 0) { 669 p->p_flag |= P_STOPPROF; 670 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 671 "stopprof", 0); 672 } 673 } 674 if ((p->p_flag & P_PROFIL) == 0) 675 return; 676 p->p_flag &= ~P_PROFIL; 677 mtx_lock(&time_lock); 678 if (--profprocs == 0) 679 cpu_stopprofclock(); 680 mtx_unlock(&time_lock); 681 } 682 } 683 684 /* 685 * Statistics clock. Updates rusage information and calls the scheduler 686 * to adjust priorities of the active thread. 687 * 688 * This should be called by all active processors. 689 */ 690 void 691 statclock(int cnt, int usermode) 692 { 693 struct rusage *ru; 694 struct vmspace *vm; 695 struct thread *td; 696 struct proc *p; 697 long rss; 698 long *cp_time; 699 uint64_t runtime, new_switchtime; 700 701 td = curthread; 702 p = td->td_proc; 703 704 cp_time = (long *)PCPU_PTR(cp_time); 705 if (usermode) { 706 /* 707 * Charge the time as appropriate. 708 */ 709 td->td_uticks += cnt; 710 if (p->p_nice > NZERO) 711 cp_time[CP_NICE] += cnt; 712 else 713 cp_time[CP_USER] += cnt; 714 } else { 715 /* 716 * Came from kernel mode, so we were: 717 * - handling an interrupt, 718 * - doing syscall or trap work on behalf of the current 719 * user process, or 720 * - spinning in the idle loop. 721 * Whichever it is, charge the time as appropriate. 722 * Note that we charge interrupts to the current process, 723 * regardless of whether they are ``for'' that process, 724 * so that we know how much of its real time was spent 725 * in ``non-process'' (i.e., interrupt) work. 726 */ 727 if ((td->td_pflags & TDP_ITHREAD) || 728 td->td_intr_nesting_level >= 2) { 729 td->td_iticks += cnt; 730 cp_time[CP_INTR] += cnt; 731 } else { 732 td->td_pticks += cnt; 733 td->td_sticks += cnt; 734 if (!TD_IS_IDLETHREAD(td)) 735 cp_time[CP_SYS] += cnt; 736 else 737 cp_time[CP_IDLE] += cnt; 738 } 739 } 740 741 /* Update resource usage integrals and maximums. */ 742 MPASS(p->p_vmspace != NULL); 743 vm = p->p_vmspace; 744 ru = &td->td_ru; 745 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt; 746 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt; 747 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt; 748 rss = pgtok(vmspace_resident_count(vm)); 749 if (ru->ru_maxrss < rss) 750 ru->ru_maxrss = rss; 751 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock", 752 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz); 753 SDT_PROBE2(sched, , , tick, td, td->td_proc); 754 thread_lock_flags(td, MTX_QUIET); 755 756 /* 757 * Compute the amount of time during which the current 758 * thread was running, and add that to its total so far. 759 */ 760 new_switchtime = cpu_ticks(); 761 runtime = new_switchtime - PCPU_GET(switchtime); 762 td->td_runtime += runtime; 763 td->td_incruntime += runtime; 764 PCPU_SET(switchtime, new_switchtime); 765 766 sched_clock(td, cnt); 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 cnt, int usermode, uintfptr_t pc) 776 { 777 struct thread *td; 778 #ifdef GPROF 779 struct gmonparam *g; 780 uintfptr_t i; 781 #endif 782 783 td = curthread; 784 if (usermode) { 785 /* 786 * Came from user mode; CPU was in user state. 787 * If this process is being profiled, record the tick. 788 * if there is no related user location yet, don't 789 * bother trying to count it. 790 */ 791 if (td->td_proc->p_flag & P_PROFIL) 792 addupc_intr(td, pc, cnt); 793 } 794 #ifdef GPROF 795 else { 796 /* 797 * Kernel statistics are just like addupc_intr, only easier. 798 */ 799 g = &_gmonparam; 800 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 801 i = PC_TO_I(g, pc); 802 if (i < g->textsize) { 803 KCOUNT(g, i) += cnt; 804 } 805 } 806 } 807 #endif 808 #ifdef HWPMC_HOOKS 809 if (td->td_intr_frame != NULL) 810 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame); 811 #endif 812 } 813 814 /* 815 * Return information about system clocks. 816 */ 817 static int 818 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 819 { 820 struct clockinfo clkinfo; 821 /* 822 * Construct clockinfo structure. 823 */ 824 bzero(&clkinfo, sizeof(clkinfo)); 825 clkinfo.hz = hz; 826 clkinfo.tick = tick; 827 clkinfo.profhz = profhz; 828 clkinfo.stathz = stathz ? stathz : hz; 829 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 830 } 831 832 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, 833 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE, 834 0, 0, sysctl_kern_clockrate, "S,clockinfo", 835 "Rate and period of various kernel clocks"); 836 837 static void 838 watchdog_config(void *unused __unused, u_int cmd, int *error) 839 { 840 u_int u; 841 842 u = cmd & WD_INTERVAL; 843 if (u >= WD_TO_1SEC) { 844 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 845 watchdog_enabled = 1; 846 *error = 0; 847 } else { 848 watchdog_enabled = 0; 849 } 850 } 851 852 /* 853 * Handle a watchdog timeout by dumping interrupt information and 854 * then either dropping to DDB or panicking. 855 */ 856 static void 857 watchdog_fire(void) 858 { 859 int nintr; 860 uint64_t inttotal; 861 u_long *curintr; 862 char *curname; 863 864 curintr = intrcnt; 865 curname = intrnames; 866 inttotal = 0; 867 nintr = sintrcnt / sizeof(u_long); 868 869 printf("interrupt total\n"); 870 while (--nintr >= 0) { 871 if (*curintr) 872 printf("%-12s %20lu\n", curname, *curintr); 873 curname += strlen(curname) + 1; 874 inttotal += *curintr++; 875 } 876 printf("Total %20ju\n", (uintmax_t)inttotal); 877 878 #if defined(KDB) && !defined(KDB_UNATTENDED) 879 kdb_backtrace(); 880 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout"); 881 #else 882 panic("watchdog timeout"); 883 #endif 884 } 885