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_device_polling.h" 41 #include "opt_hwpmc_hooks.h" 42 #include "opt_ntp.h" 43 #include "opt_watchdog.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/callout.h> 48 #include <sys/kdb.h> 49 #include <sys/kernel.h> 50 #include <sys/lock.h> 51 #include <sys/ktr.h> 52 #include <sys/mutex.h> 53 #include <sys/proc.h> 54 #include <sys/resource.h> 55 #include <sys/resourcevar.h> 56 #include <sys/sched.h> 57 #include <sys/signalvar.h> 58 #include <sys/smp.h> 59 #include <vm/vm.h> 60 #include <vm/pmap.h> 61 #include <vm/vm_map.h> 62 #include <sys/sysctl.h> 63 #include <sys/bus.h> 64 #include <sys/interrupt.h> 65 #include <sys/limits.h> 66 #include <sys/timetc.h> 67 68 #ifdef GPROF 69 #include <sys/gmon.h> 70 #endif 71 72 #ifdef HWPMC_HOOKS 73 #include <sys/pmckern.h> 74 #endif 75 76 #ifdef DEVICE_POLLING 77 extern void hardclock_device_poll(void); 78 #endif /* DEVICE_POLLING */ 79 80 static void initclocks(void *dummy); 81 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) 82 83 /* Some of these don't belong here, but it's easiest to concentrate them. */ 84 long cp_time[CPUSTATES]; 85 86 static int 87 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS) 88 { 89 int error; 90 #ifdef SCTL_MASK32 91 int i; 92 unsigned int cp_time32[CPUSTATES]; 93 94 if (req->flags & SCTL_MASK32) { 95 if (!req->oldptr) 96 return SYSCTL_OUT(req, 0, sizeof(cp_time32)); 97 for (i = 0; i < CPUSTATES; i++) 98 cp_time32[i] = (unsigned int)cp_time[i]; 99 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); 100 } else 101 #endif 102 { 103 if (!req->oldptr) 104 return SYSCTL_OUT(req, 0, sizeof(cp_time)); 105 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time)); 106 } 107 return error; 108 } 109 110 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD, 111 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics"); 112 113 #ifdef SW_WATCHDOG 114 #include <sys/watchdog.h> 115 116 static int watchdog_ticks; 117 static int watchdog_enabled; 118 static void watchdog_fire(void); 119 static void watchdog_config(void *, u_int, int *); 120 #endif /* SW_WATCHDOG */ 121 122 /* 123 * Clock handling routines. 124 * 125 * This code is written to operate with two timers that run independently of 126 * each other. 127 * 128 * The main timer, running hz times per second, is used to trigger interval 129 * timers, timeouts and rescheduling as needed. 130 * 131 * The second timer handles kernel and user profiling, 132 * and does resource use estimation. If the second timer is programmable, 133 * it is randomized to avoid aliasing between the two clocks. For example, 134 * the randomization prevents an adversary from always giving up the cpu 135 * just before its quantum expires. Otherwise, it would never accumulate 136 * cpu ticks. The mean frequency of the second timer is stathz. 137 * 138 * If no second timer exists, stathz will be zero; in this case we drive 139 * profiling and statistics off the main clock. This WILL NOT be accurate; 140 * do not do it unless absolutely necessary. 141 * 142 * The statistics clock may (or may not) be run at a higher rate while 143 * profiling. This profile clock runs at profhz. We require that profhz 144 * be an integral multiple of stathz. 145 * 146 * If the statistics clock is running fast, it must be divided by the ratio 147 * profhz/stathz for statistics. (For profiling, every tick counts.) 148 * 149 * Time-of-day is maintained using a "timecounter", which may or may 150 * not be related to the hardware generating the above mentioned 151 * interrupts. 152 */ 153 154 int stathz; 155 int profhz; 156 int profprocs; 157 int ticks; 158 int psratio; 159 160 /* 161 * Initialize clock frequencies and start both clocks running. 162 */ 163 /* ARGSUSED*/ 164 static void 165 initclocks(dummy) 166 void *dummy; 167 { 168 register int i; 169 170 /* 171 * Set divisors to 1 (normal case) and let the machine-specific 172 * code do its bit. 173 */ 174 cpu_initclocks(); 175 176 /* 177 * Compute profhz/stathz, and fix profhz if needed. 178 */ 179 i = stathz ? stathz : hz; 180 if (profhz == 0) 181 profhz = i; 182 psratio = profhz / i; 183 #ifdef SW_WATCHDOG 184 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 185 #endif 186 } 187 188 /* 189 * Each time the real-time timer fires, this function is called on all CPUs. 190 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only 191 * the other CPUs in the system need to call this function. 192 */ 193 void 194 hardclock_cpu(int usermode) 195 { 196 struct pstats *pstats; 197 struct thread *td = curthread; 198 struct proc *p = td->td_proc; 199 200 /* 201 * Run current process's virtual and profile time, as needed. 202 */ 203 mtx_lock_spin_flags(&sched_lock, MTX_QUIET); 204 sched_tick(); 205 if (p->p_flag & P_SA) { 206 /* XXXKSE What to do? */ 207 } else { 208 pstats = p->p_stats; 209 if (usermode && 210 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 211 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 212 p->p_sflag |= PS_ALRMPEND; 213 td->td_flags |= TDF_ASTPENDING; 214 } 215 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 216 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 217 p->p_sflag |= PS_PROFPEND; 218 td->td_flags |= TDF_ASTPENDING; 219 } 220 } 221 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 222 223 #ifdef HWPMC_HOOKS 224 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 225 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 226 #endif 227 } 228 229 /* 230 * The real-time timer, interrupting hz times per second. 231 */ 232 void 233 hardclock(int usermode, uintfptr_t pc) 234 { 235 int need_softclock = 0; 236 237 hardclock_cpu(usermode); 238 239 tc_ticktock(); 240 /* 241 * If no separate statistics clock is available, run it from here. 242 * 243 * XXX: this only works for UP 244 */ 245 if (stathz == 0) { 246 profclock(usermode, pc); 247 statclock(usermode); 248 } 249 250 #ifdef DEVICE_POLLING 251 hardclock_device_poll(); /* this is very short and quick */ 252 #endif /* DEVICE_POLLING */ 253 254 /* 255 * Process callouts at a very low cpu priority, so we don't keep the 256 * relatively high clock interrupt priority any longer than necessary. 257 */ 258 mtx_lock_spin_flags(&callout_lock, MTX_QUIET); 259 ticks++; 260 if (!TAILQ_EMPTY(&callwheel[ticks & callwheelmask])) { 261 need_softclock = 1; 262 } else if (softticks + 1 == ticks) 263 ++softticks; 264 mtx_unlock_spin_flags(&callout_lock, MTX_QUIET); 265 266 /* 267 * swi_sched acquires sched_lock, so we don't want to call it with 268 * callout_lock held; incorrect locking order. 269 */ 270 if (need_softclock) 271 swi_sched(softclock_ih, 0); 272 273 #ifdef SW_WATCHDOG 274 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 275 watchdog_fire(); 276 #endif /* SW_WATCHDOG */ 277 } 278 279 /* 280 * Compute number of ticks in the specified amount of time. 281 */ 282 int 283 tvtohz(tv) 284 struct timeval *tv; 285 { 286 register unsigned long ticks; 287 register long sec, usec; 288 289 /* 290 * If the number of usecs in the whole seconds part of the time 291 * difference fits in a long, then the total number of usecs will 292 * fit in an unsigned long. Compute the total and convert it to 293 * ticks, rounding up and adding 1 to allow for the current tick 294 * to expire. Rounding also depends on unsigned long arithmetic 295 * to avoid overflow. 296 * 297 * Otherwise, if the number of ticks in the whole seconds part of 298 * the time difference fits in a long, then convert the parts to 299 * ticks separately and add, using similar rounding methods and 300 * overflow avoidance. This method would work in the previous 301 * case but it is slightly slower and assumes that hz is integral. 302 * 303 * Otherwise, round the time difference down to the maximum 304 * representable value. 305 * 306 * If ints have 32 bits, then the maximum value for any timeout in 307 * 10ms ticks is 248 days. 308 */ 309 sec = tv->tv_sec; 310 usec = tv->tv_usec; 311 if (usec < 0) { 312 sec--; 313 usec += 1000000; 314 } 315 if (sec < 0) { 316 #ifdef DIAGNOSTIC 317 if (usec > 0) { 318 sec++; 319 usec -= 1000000; 320 } 321 printf("tvotohz: negative time difference %ld sec %ld usec\n", 322 sec, usec); 323 #endif 324 ticks = 1; 325 } else if (sec <= LONG_MAX / 1000000) 326 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 327 / tick + 1; 328 else if (sec <= LONG_MAX / hz) 329 ticks = sec * hz 330 + ((unsigned long)usec + (tick - 1)) / tick + 1; 331 else 332 ticks = LONG_MAX; 333 if (ticks > INT_MAX) 334 ticks = INT_MAX; 335 return ((int)ticks); 336 } 337 338 /* 339 * Start profiling on a process. 340 * 341 * Kernel profiling passes proc0 which never exits and hence 342 * keeps the profile clock running constantly. 343 */ 344 void 345 startprofclock(p) 346 register struct proc *p; 347 { 348 349 /* 350 * XXX; Right now sched_lock protects statclock(), but perhaps 351 * it should be protected later on by a time_lock, which would 352 * cover psdiv, etc. as well. 353 */ 354 PROC_LOCK_ASSERT(p, MA_OWNED); 355 if (p->p_flag & P_STOPPROF) 356 return; 357 if ((p->p_flag & P_PROFIL) == 0) { 358 mtx_lock_spin(&sched_lock); 359 p->p_flag |= P_PROFIL; 360 if (++profprocs == 1) 361 cpu_startprofclock(); 362 mtx_unlock_spin(&sched_lock); 363 } 364 } 365 366 /* 367 * Stop profiling on a process. 368 */ 369 void 370 stopprofclock(p) 371 register struct proc *p; 372 { 373 374 PROC_LOCK_ASSERT(p, MA_OWNED); 375 if (p->p_flag & P_PROFIL) { 376 if (p->p_profthreads != 0) { 377 p->p_flag |= P_STOPPROF; 378 while (p->p_profthreads != 0) 379 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 380 "stopprof", 0); 381 p->p_flag &= ~P_STOPPROF; 382 } 383 if ((p->p_flag & P_PROFIL) == 0) 384 return; 385 mtx_lock_spin(&sched_lock); 386 p->p_flag &= ~P_PROFIL; 387 if (--profprocs == 0) 388 cpu_stopprofclock(); 389 mtx_unlock_spin(&sched_lock); 390 } 391 } 392 393 /* 394 * Statistics clock. Grab profile sample, and if divider reaches 0, 395 * do process and kernel statistics. Most of the statistics are only 396 * used by user-level statistics programs. The main exceptions are 397 * ke->ke_uticks, p->p_rux.rux_sticks, p->p_rux.rux_iticks, and p->p_estcpu. 398 * This should be called by all active processors. 399 */ 400 void 401 statclock(int usermode) 402 { 403 struct rusage *ru; 404 struct vmspace *vm; 405 struct thread *td; 406 struct proc *p; 407 long rss; 408 409 td = curthread; 410 p = td->td_proc; 411 412 mtx_lock_spin_flags(&sched_lock, MTX_QUIET); 413 if (usermode) { 414 /* 415 * Charge the time as appropriate. 416 */ 417 if (p->p_flag & P_SA) 418 thread_statclock(1); 419 td->td_uticks++; 420 if (p->p_nice > NZERO) 421 cp_time[CP_NICE]++; 422 else 423 cp_time[CP_USER]++; 424 } else { 425 /* 426 * Came from kernel mode, so we were: 427 * - handling an interrupt, 428 * - doing syscall or trap work on behalf of the current 429 * user process, or 430 * - spinning in the idle loop. 431 * Whichever it is, charge the time as appropriate. 432 * Note that we charge interrupts to the current process, 433 * regardless of whether they are ``for'' that process, 434 * so that we know how much of its real time was spent 435 * in ``non-process'' (i.e., interrupt) work. 436 */ 437 if ((td->td_pflags & TDP_ITHREAD) || 438 td->td_intr_nesting_level >= 2) { 439 td->td_iticks++; 440 cp_time[CP_INTR]++; 441 } else { 442 if (p->p_flag & P_SA) 443 thread_statclock(0); 444 td->td_pticks++; 445 td->td_sticks++; 446 if (td != PCPU_GET(idlethread)) 447 cp_time[CP_SYS]++; 448 else 449 cp_time[CP_IDLE]++; 450 } 451 } 452 CTR4(KTR_SCHED, "statclock: %p(%s) prio %d stathz %d", 453 td, td->td_proc->p_comm, td->td_priority, (stathz)?stathz:hz); 454 455 sched_clock(td); 456 457 /* Update resource usage integrals and maximums. */ 458 MPASS(p->p_stats != NULL); 459 MPASS(p->p_vmspace != NULL); 460 vm = p->p_vmspace; 461 ru = &p->p_stats->p_ru; 462 ru->ru_ixrss += pgtok(vm->vm_tsize); 463 ru->ru_idrss += pgtok(vm->vm_dsize); 464 ru->ru_isrss += pgtok(vm->vm_ssize); 465 rss = pgtok(vmspace_resident_count(vm)); 466 if (ru->ru_maxrss < rss) 467 ru->ru_maxrss = rss; 468 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 469 } 470 471 void 472 profclock(int usermode, uintfptr_t pc) 473 { 474 struct thread *td; 475 #ifdef GPROF 476 struct gmonparam *g; 477 uintfptr_t i; 478 #endif 479 480 td = curthread; 481 if (usermode) { 482 /* 483 * Came from user mode; CPU was in user state. 484 * If this process is being profiled, record the tick. 485 * if there is no related user location yet, don't 486 * bother trying to count it. 487 */ 488 if (td->td_proc->p_flag & P_PROFIL) 489 addupc_intr(td, pc, 1); 490 } 491 #ifdef GPROF 492 else { 493 /* 494 * Kernel statistics are just like addupc_intr, only easier. 495 */ 496 g = &_gmonparam; 497 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 498 i = PC_TO_I(g, pc); 499 if (i < g->textsize) { 500 KCOUNT(g, i)++; 501 } 502 } 503 } 504 #endif 505 } 506 507 /* 508 * Return information about system clocks. 509 */ 510 static int 511 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 512 { 513 struct clockinfo clkinfo; 514 /* 515 * Construct clockinfo structure. 516 */ 517 bzero(&clkinfo, sizeof(clkinfo)); 518 clkinfo.hz = hz; 519 clkinfo.tick = tick; 520 clkinfo.profhz = profhz; 521 clkinfo.stathz = stathz ? stathz : hz; 522 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 523 } 524 525 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 526 0, 0, sysctl_kern_clockrate, "S,clockinfo", 527 "Rate and period of various kernel clocks"); 528 529 #ifdef SW_WATCHDOG 530 531 static void 532 watchdog_config(void *unused __unused, u_int cmd, int *err) 533 { 534 u_int u; 535 536 u = cmd & WD_INTERVAL; 537 if ((cmd & WD_ACTIVE) && u >= WD_TO_1SEC) { 538 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 539 watchdog_enabled = 1; 540 *err = 0; 541 } else { 542 watchdog_enabled = 0; 543 } 544 } 545 546 /* 547 * Handle a watchdog timeout by dumping interrupt information and 548 * then either dropping to DDB or panicing. 549 */ 550 static void 551 watchdog_fire(void) 552 { 553 int nintr; 554 u_int64_t inttotal; 555 u_long *curintr; 556 char *curname; 557 558 curintr = intrcnt; 559 curname = intrnames; 560 inttotal = 0; 561 nintr = eintrcnt - intrcnt; 562 563 printf("interrupt total\n"); 564 while (--nintr >= 0) { 565 if (*curintr) 566 printf("%-12s %20lu\n", curname, *curintr); 567 curname += strlen(curname) + 1; 568 inttotal += *curintr++; 569 } 570 printf("Total %20ju\n", (uintmax_t)inttotal); 571 572 #ifdef KDB 573 kdb_backtrace(); 574 kdb_enter("watchdog timeout"); 575 #else 576 panic("watchdog timeout"); 577 #endif /* KDB */ 578 } 579 580 #endif /* SW_WATCHDOG */ 581