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/lock.h> 52 #include <sys/ktr.h> 53 #include <sys/mutex.h> 54 #include <sys/proc.h> 55 #include <sys/resource.h> 56 #include <sys/resourcevar.h> 57 #include <sys/sched.h> 58 #include <sys/signalvar.h> 59 #include <sys/smp.h> 60 #include <vm/vm.h> 61 #include <vm/pmap.h> 62 #include <vm/vm_map.h> 63 #include <sys/sysctl.h> 64 #include <sys/bus.h> 65 #include <sys/interrupt.h> 66 #include <sys/limits.h> 67 #include <sys/timetc.h> 68 69 #ifdef GPROF 70 #include <sys/gmon.h> 71 #endif 72 73 #ifdef HWPMC_HOOKS 74 #include <sys/pmckern.h> 75 #endif 76 77 #ifdef DEVICE_POLLING 78 extern void hardclock_device_poll(void); 79 #endif /* DEVICE_POLLING */ 80 81 static void initclocks(void *dummy); 82 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL); 83 84 /* Spin-lock protecting profiling statistics. */ 85 static struct mtx time_lock; 86 87 static int 88 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS) 89 { 90 int error; 91 long cp_time[CPUSTATES]; 92 #ifdef SCTL_MASK32 93 int i; 94 unsigned int cp_time32[CPUSTATES]; 95 #endif 96 97 read_cpu_time(cp_time); 98 #ifdef SCTL_MASK32 99 if (req->flags & SCTL_MASK32) { 100 if (!req->oldptr) 101 return SYSCTL_OUT(req, 0, sizeof(cp_time32)); 102 for (i = 0; i < CPUSTATES; i++) 103 cp_time32[i] = (unsigned int)cp_time[i]; 104 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 105 } else 106 #endif 107 { 108 if (!req->oldptr) 109 return SYSCTL_OUT(req, 0, sizeof(cp_time)); 110 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time)); 111 } 112 return error; 113 } 114 115 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 116 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics"); 117 118 static long empty[CPUSTATES]; 119 120 static int 121 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS) 122 { 123 struct pcpu *pcpu; 124 int error; 125 int c; 126 long *cp_time; 127 #ifdef SCTL_MASK32 128 unsigned int cp_time32[CPUSTATES]; 129 int i; 130 #endif 131 132 if (!req->oldptr) { 133 #ifdef SCTL_MASK32 134 if (req->flags & SCTL_MASK32) 135 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1)); 136 else 137 #endif 138 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1)); 139 } 140 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) { 141 if (!CPU_ABSENT(c)) { 142 pcpu = pcpu_find(c); 143 cp_time = pcpu->pc_cp_time; 144 } else { 145 cp_time = empty; 146 } 147 #ifdef SCTL_MASK32 148 if (req->flags & SCTL_MASK32) { 149 for (i = 0; i < CPUSTATES; i++) 150 cp_time32[i] = (unsigned int)cp_time[i]; 151 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 152 } else 153 #endif 154 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES); 155 } 156 return error; 157 } 158 159 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 160 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics"); 161 162 void 163 read_cpu_time(long *cp_time) 164 { 165 struct pcpu *pc; 166 int i, j; 167 168 /* Sum up global cp_time[]. */ 169 bzero(cp_time, sizeof(long) * CPUSTATES); 170 for (i = 0; i <= mp_maxid; i++) { 171 if (CPU_ABSENT(i)) 172 continue; 173 pc = pcpu_find(i); 174 for (j = 0; j < CPUSTATES; j++) 175 cp_time[j] += pc->pc_cp_time[j]; 176 } 177 } 178 179 #ifdef SW_WATCHDOG 180 #include <sys/watchdog.h> 181 182 static int watchdog_ticks; 183 static int watchdog_enabled; 184 static void watchdog_fire(void); 185 static void watchdog_config(void *, u_int, int *); 186 #endif /* SW_WATCHDOG */ 187 188 /* 189 * Clock handling routines. 190 * 191 * This code is written to operate with two timers that run independently of 192 * each other. 193 * 194 * The main timer, running hz times per second, is used to trigger interval 195 * timers, timeouts and rescheduling as needed. 196 * 197 * The second timer handles kernel and user profiling, 198 * and does resource use estimation. If the second timer is programmable, 199 * it is randomized to avoid aliasing between the two clocks. For example, 200 * the randomization prevents an adversary from always giving up the cpu 201 * just before its quantum expires. Otherwise, it would never accumulate 202 * cpu ticks. The mean frequency of the second timer is stathz. 203 * 204 * If no second timer exists, stathz will be zero; in this case we drive 205 * profiling and statistics off the main clock. This WILL NOT be accurate; 206 * do not do it unless absolutely necessary. 207 * 208 * The statistics clock may (or may not) be run at a higher rate while 209 * profiling. This profile clock runs at profhz. We require that profhz 210 * be an integral multiple of stathz. 211 * 212 * If the statistics clock is running fast, it must be divided by the ratio 213 * profhz/stathz for statistics. (For profiling, every tick counts.) 214 * 215 * Time-of-day is maintained using a "timecounter", which may or may 216 * not be related to the hardware generating the above mentioned 217 * interrupts. 218 */ 219 220 int stathz; 221 int profhz; 222 int profprocs; 223 int ticks; 224 int psratio; 225 226 /* 227 * Initialize clock frequencies and start both clocks running. 228 */ 229 /* ARGSUSED*/ 230 static void 231 initclocks(dummy) 232 void *dummy; 233 { 234 register int i; 235 236 /* 237 * Set divisors to 1 (normal case) and let the machine-specific 238 * code do its bit. 239 */ 240 mtx_init(&time_lock, "time lock", NULL, MTX_SPIN); 241 cpu_initclocks(); 242 243 /* 244 * Compute profhz/stathz, and fix profhz if needed. 245 */ 246 i = stathz ? stathz : hz; 247 if (profhz == 0) 248 profhz = i; 249 psratio = profhz / i; 250 #ifdef SW_WATCHDOG 251 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 252 #endif 253 } 254 255 /* 256 * Each time the real-time timer fires, this function is called on all CPUs. 257 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only 258 * the other CPUs in the system need to call this function. 259 */ 260 void 261 hardclock_cpu(int usermode) 262 { 263 struct pstats *pstats; 264 struct thread *td = curthread; 265 struct proc *p = td->td_proc; 266 int flags; 267 268 /* 269 * Run current process's virtual and profile time, as needed. 270 */ 271 pstats = p->p_stats; 272 flags = 0; 273 if (usermode && 274 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { 275 PROC_SLOCK(p); 276 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 277 flags |= TDF_ALRMPEND | TDF_ASTPENDING; 278 PROC_SUNLOCK(p); 279 } 280 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { 281 PROC_SLOCK(p); 282 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 283 flags |= TDF_PROFPEND | TDF_ASTPENDING; 284 PROC_SUNLOCK(p); 285 } 286 thread_lock(td); 287 sched_tick(); 288 td->td_flags |= flags; 289 thread_unlock(td); 290 291 #ifdef HWPMC_HOOKS 292 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 293 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 294 #endif 295 callout_tick(); 296 } 297 298 /* 299 * The real-time timer, interrupting hz times per second. 300 */ 301 void 302 hardclock(int usermode, uintfptr_t pc) 303 { 304 305 atomic_add_int((volatile int *)&ticks, 1); 306 hardclock_cpu(usermode); 307 tc_ticktock(); 308 /* 309 * If no separate statistics clock is available, run it from here. 310 * 311 * XXX: this only works for UP 312 */ 313 if (stathz == 0) { 314 profclock(usermode, pc); 315 statclock(usermode); 316 } 317 #ifdef DEVICE_POLLING 318 hardclock_device_poll(); /* this is very short and quick */ 319 #endif /* DEVICE_POLLING */ 320 #ifdef SW_WATCHDOG 321 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 322 watchdog_fire(); 323 #endif /* SW_WATCHDOG */ 324 } 325 326 /* 327 * Compute number of ticks in the specified amount of time. 328 */ 329 int 330 tvtohz(tv) 331 struct timeval *tv; 332 { 333 register unsigned long ticks; 334 register long sec, usec; 335 336 /* 337 * If the number of usecs in the whole seconds part of the time 338 * difference fits in a long, then the total number of usecs will 339 * fit in an unsigned long. Compute the total and convert it to 340 * ticks, rounding up and adding 1 to allow for the current tick 341 * to expire. Rounding also depends on unsigned long arithmetic 342 * to avoid overflow. 343 * 344 * Otherwise, if the number of ticks in the whole seconds part of 345 * the time difference fits in a long, then convert the parts to 346 * ticks separately and add, using similar rounding methods and 347 * overflow avoidance. This method would work in the previous 348 * case but it is slightly slower and assumes that hz is integral. 349 * 350 * Otherwise, round the time difference down to the maximum 351 * representable value. 352 * 353 * If ints have 32 bits, then the maximum value for any timeout in 354 * 10ms ticks is 248 days. 355 */ 356 sec = tv->tv_sec; 357 usec = tv->tv_usec; 358 if (usec < 0) { 359 sec--; 360 usec += 1000000; 361 } 362 if (sec < 0) { 363 #ifdef DIAGNOSTIC 364 if (usec > 0) { 365 sec++; 366 usec -= 1000000; 367 } 368 printf("tvotohz: negative time difference %ld sec %ld usec\n", 369 sec, usec); 370 #endif 371 ticks = 1; 372 } else if (sec <= LONG_MAX / 1000000) 373 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 374 / tick + 1; 375 else if (sec <= LONG_MAX / hz) 376 ticks = sec * hz 377 + ((unsigned long)usec + (tick - 1)) / tick + 1; 378 else 379 ticks = LONG_MAX; 380 if (ticks > INT_MAX) 381 ticks = INT_MAX; 382 return ((int)ticks); 383 } 384 385 /* 386 * Start profiling on a process. 387 * 388 * Kernel profiling passes proc0 which never exits and hence 389 * keeps the profile clock running constantly. 390 */ 391 void 392 startprofclock(p) 393 register struct proc *p; 394 { 395 396 PROC_LOCK_ASSERT(p, MA_OWNED); 397 if (p->p_flag & P_STOPPROF) 398 return; 399 if ((p->p_flag & P_PROFIL) == 0) { 400 p->p_flag |= P_PROFIL; 401 mtx_lock_spin(&time_lock); 402 if (++profprocs == 1) 403 cpu_startprofclock(); 404 mtx_unlock_spin(&time_lock); 405 } 406 } 407 408 /* 409 * Stop profiling on a process. 410 */ 411 void 412 stopprofclock(p) 413 register struct proc *p; 414 { 415 416 PROC_LOCK_ASSERT(p, MA_OWNED); 417 if (p->p_flag & P_PROFIL) { 418 if (p->p_profthreads != 0) { 419 p->p_flag |= P_STOPPROF; 420 while (p->p_profthreads != 0) 421 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 422 "stopprof", 0); 423 p->p_flag &= ~P_STOPPROF; 424 } 425 if ((p->p_flag & P_PROFIL) == 0) 426 return; 427 p->p_flag &= ~P_PROFIL; 428 mtx_lock_spin(&time_lock); 429 if (--profprocs == 0) 430 cpu_stopprofclock(); 431 mtx_unlock_spin(&time_lock); 432 } 433 } 434 435 /* 436 * Statistics clock. Updates rusage information and calls the scheduler 437 * to adjust priorities of the active thread. 438 * 439 * This should be called by all active processors. 440 */ 441 void 442 statclock(int usermode) 443 { 444 struct rusage *ru; 445 struct vmspace *vm; 446 struct thread *td; 447 struct proc *p; 448 long rss; 449 long *cp_time; 450 451 td = curthread; 452 p = td->td_proc; 453 454 cp_time = (long *)PCPU_PTR(cp_time); 455 if (usermode) { 456 /* 457 * Charge the time as appropriate. 458 */ 459 td->td_uticks++; 460 if (p->p_nice > NZERO) 461 cp_time[CP_NICE]++; 462 else 463 cp_time[CP_USER]++; 464 } else { 465 /* 466 * Came from kernel mode, so we were: 467 * - handling an interrupt, 468 * - doing syscall or trap work on behalf of the current 469 * user process, or 470 * - spinning in the idle loop. 471 * Whichever it is, charge the time as appropriate. 472 * Note that we charge interrupts to the current process, 473 * regardless of whether they are ``for'' that process, 474 * so that we know how much of its real time was spent 475 * in ``non-process'' (i.e., interrupt) work. 476 */ 477 if ((td->td_pflags & TDP_ITHREAD) || 478 td->td_intr_nesting_level >= 2) { 479 td->td_iticks++; 480 cp_time[CP_INTR]++; 481 } else { 482 td->td_pticks++; 483 td->td_sticks++; 484 if (!TD_IS_IDLETHREAD(td)) 485 cp_time[CP_SYS]++; 486 else 487 cp_time[CP_IDLE]++; 488 } 489 } 490 491 /* Update resource usage integrals and maximums. */ 492 MPASS(p->p_vmspace != NULL); 493 vm = p->p_vmspace; 494 ru = &td->td_ru; 495 ru->ru_ixrss += pgtok(vm->vm_tsize); 496 ru->ru_idrss += pgtok(vm->vm_dsize); 497 ru->ru_isrss += pgtok(vm->vm_ssize); 498 rss = pgtok(vmspace_resident_count(vm)); 499 if (ru->ru_maxrss < rss) 500 ru->ru_maxrss = rss; 501 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock", 502 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz); 503 thread_lock_flags(td, MTX_QUIET); 504 sched_clock(td); 505 thread_unlock(td); 506 } 507 508 void 509 profclock(int usermode, uintfptr_t pc) 510 { 511 struct thread *td; 512 #ifdef GPROF 513 struct gmonparam *g; 514 uintfptr_t i; 515 #endif 516 517 td = curthread; 518 if (usermode) { 519 /* 520 * Came from user mode; CPU was in user state. 521 * If this process is being profiled, record the tick. 522 * if there is no related user location yet, don't 523 * bother trying to count it. 524 */ 525 if (td->td_proc->p_flag & P_PROFIL) 526 addupc_intr(td, pc, 1); 527 } 528 #ifdef GPROF 529 else { 530 /* 531 * Kernel statistics are just like addupc_intr, only easier. 532 */ 533 g = &_gmonparam; 534 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 535 i = PC_TO_I(g, pc); 536 if (i < g->textsize) { 537 KCOUNT(g, i)++; 538 } 539 } 540 } 541 #endif 542 } 543 544 /* 545 * Return information about system clocks. 546 */ 547 static int 548 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 549 { 550 struct clockinfo clkinfo; 551 /* 552 * Construct clockinfo structure. 553 */ 554 bzero(&clkinfo, sizeof(clkinfo)); 555 clkinfo.hz = hz; 556 clkinfo.tick = tick; 557 clkinfo.profhz = profhz; 558 clkinfo.stathz = stathz ? stathz : hz; 559 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 560 } 561 562 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, 563 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE, 564 0, 0, sysctl_kern_clockrate, "S,clockinfo", 565 "Rate and period of various kernel clocks"); 566 567 #ifdef SW_WATCHDOG 568 569 static void 570 watchdog_config(void *unused __unused, u_int cmd, int *error) 571 { 572 u_int u; 573 574 u = cmd & WD_INTERVAL; 575 if (u >= WD_TO_1SEC) { 576 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 577 watchdog_enabled = 1; 578 *error = 0; 579 } else { 580 watchdog_enabled = 0; 581 } 582 } 583 584 /* 585 * Handle a watchdog timeout by dumping interrupt information and 586 * then either dropping to DDB or panicking. 587 */ 588 static void 589 watchdog_fire(void) 590 { 591 int nintr; 592 u_int64_t inttotal; 593 u_long *curintr; 594 char *curname; 595 596 curintr = intrcnt; 597 curname = intrnames; 598 inttotal = 0; 599 nintr = eintrcnt - intrcnt; 600 601 printf("interrupt total\n"); 602 while (--nintr >= 0) { 603 if (*curintr) 604 printf("%-12s %20lu\n", curname, *curintr); 605 curname += strlen(curname) + 1; 606 inttotal += *curintr++; 607 } 608 printf("Total %20ju\n", (uintmax_t)inttotal); 609 610 #if defined(KDB) && !defined(KDB_UNATTENDED) 611 kdb_backtrace(); 612 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout"); 613 #else 614 panic("watchdog timeout"); 615 #endif 616 } 617 618 #endif /* SW_WATCHDOG */ 619