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 #ifdef KSE 206 #if 0 /* for now do nothing */ 207 if (p->p_flag & P_SA) { 208 /* XXXKSE What to do? Should do more. */ 209 } 210 #endif 211 #endif 212 pstats = p->p_stats; 213 if (usermode && 214 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 215 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 216 p->p_sflag |= PS_ALRMPEND; 217 td->td_flags |= TDF_ASTPENDING; 218 } 219 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 220 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 221 p->p_sflag |= PS_PROFPEND; 222 td->td_flags |= TDF_ASTPENDING; 223 } 224 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 225 226 #ifdef HWPMC_HOOKS 227 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 228 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 229 #endif 230 } 231 232 /* 233 * The real-time timer, interrupting hz times per second. 234 */ 235 void 236 hardclock(int usermode, uintfptr_t pc) 237 { 238 int need_softclock = 0; 239 240 hardclock_cpu(usermode); 241 242 tc_ticktock(); 243 /* 244 * If no separate statistics clock is available, run it from here. 245 * 246 * XXX: this only works for UP 247 */ 248 if (stathz == 0) { 249 profclock(usermode, pc); 250 statclock(usermode); 251 } 252 253 #ifdef DEVICE_POLLING 254 hardclock_device_poll(); /* this is very short and quick */ 255 #endif /* DEVICE_POLLING */ 256 257 /* 258 * Process callouts at a very low cpu priority, so we don't keep the 259 * relatively high clock interrupt priority any longer than necessary. 260 */ 261 mtx_lock_spin_flags(&callout_lock, MTX_QUIET); 262 ticks++; 263 if (!TAILQ_EMPTY(&callwheel[ticks & callwheelmask])) { 264 need_softclock = 1; 265 } else if (softticks + 1 == ticks) 266 ++softticks; 267 mtx_unlock_spin_flags(&callout_lock, MTX_QUIET); 268 269 /* 270 * swi_sched acquires sched_lock, so we don't want to call it with 271 * callout_lock held; incorrect locking order. 272 */ 273 if (need_softclock) 274 swi_sched(softclock_ih, 0); 275 276 #ifdef SW_WATCHDOG 277 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 278 watchdog_fire(); 279 #endif /* SW_WATCHDOG */ 280 } 281 282 /* 283 * Compute number of ticks in the specified amount of time. 284 */ 285 int 286 tvtohz(tv) 287 struct timeval *tv; 288 { 289 register unsigned long ticks; 290 register long sec, usec; 291 292 /* 293 * If the number of usecs in the whole seconds part of the time 294 * difference fits in a long, then the total number of usecs will 295 * fit in an unsigned long. Compute the total and convert it to 296 * ticks, rounding up and adding 1 to allow for the current tick 297 * to expire. Rounding also depends on unsigned long arithmetic 298 * to avoid overflow. 299 * 300 * Otherwise, if the number of ticks in the whole seconds part of 301 * the time difference fits in a long, then convert the parts to 302 * ticks separately and add, using similar rounding methods and 303 * overflow avoidance. This method would work in the previous 304 * case but it is slightly slower and assumes that hz is integral. 305 * 306 * Otherwise, round the time difference down to the maximum 307 * representable value. 308 * 309 * If ints have 32 bits, then the maximum value for any timeout in 310 * 10ms ticks is 248 days. 311 */ 312 sec = tv->tv_sec; 313 usec = tv->tv_usec; 314 if (usec < 0) { 315 sec--; 316 usec += 1000000; 317 } 318 if (sec < 0) { 319 #ifdef DIAGNOSTIC 320 if (usec > 0) { 321 sec++; 322 usec -= 1000000; 323 } 324 printf("tvotohz: negative time difference %ld sec %ld usec\n", 325 sec, usec); 326 #endif 327 ticks = 1; 328 } else if (sec <= LONG_MAX / 1000000) 329 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 330 / tick + 1; 331 else if (sec <= LONG_MAX / hz) 332 ticks = sec * hz 333 + ((unsigned long)usec + (tick - 1)) / tick + 1; 334 else 335 ticks = LONG_MAX; 336 if (ticks > INT_MAX) 337 ticks = INT_MAX; 338 return ((int)ticks); 339 } 340 341 /* 342 * Start profiling on a process. 343 * 344 * Kernel profiling passes proc0 which never exits and hence 345 * keeps the profile clock running constantly. 346 */ 347 void 348 startprofclock(p) 349 register struct proc *p; 350 { 351 352 /* 353 * XXX; Right now sched_lock protects statclock(), but perhaps 354 * it should be protected later on by a time_lock, which would 355 * cover psdiv, etc. as well. 356 */ 357 PROC_LOCK_ASSERT(p, MA_OWNED); 358 if (p->p_flag & P_STOPPROF) 359 return; 360 if ((p->p_flag & P_PROFIL) == 0) { 361 mtx_lock_spin(&sched_lock); 362 p->p_flag |= P_PROFIL; 363 if (++profprocs == 1) 364 cpu_startprofclock(); 365 mtx_unlock_spin(&sched_lock); 366 } 367 } 368 369 /* 370 * Stop profiling on a process. 371 */ 372 void 373 stopprofclock(p) 374 register struct proc *p; 375 { 376 377 PROC_LOCK_ASSERT(p, MA_OWNED); 378 if (p->p_flag & P_PROFIL) { 379 if (p->p_profthreads != 0) { 380 p->p_flag |= P_STOPPROF; 381 while (p->p_profthreads != 0) 382 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 383 "stopprof", 0); 384 p->p_flag &= ~P_STOPPROF; 385 } 386 if ((p->p_flag & P_PROFIL) == 0) 387 return; 388 mtx_lock_spin(&sched_lock); 389 p->p_flag &= ~P_PROFIL; 390 if (--profprocs == 0) 391 cpu_stopprofclock(); 392 mtx_unlock_spin(&sched_lock); 393 } 394 } 395 396 /* 397 * Statistics clock. Grab profile sample, and if divider reaches 0, 398 * do process and kernel statistics. Most of the statistics are only 399 * used by user-level statistics programs. The main exceptions are 400 * ke->ke_uticks, p->p_rux.rux_sticks, p->p_rux.rux_iticks, and p->p_estcpu. 401 * This should be called by all active processors. 402 */ 403 void 404 statclock(int usermode) 405 { 406 struct rusage *ru; 407 struct vmspace *vm; 408 struct thread *td; 409 struct proc *p; 410 long rss; 411 412 td = curthread; 413 p = td->td_proc; 414 415 mtx_lock_spin_flags(&sched_lock, MTX_QUIET); 416 if (usermode) { 417 /* 418 * Charge the time as appropriate. 419 */ 420 #ifdef KSE 421 if (p->p_flag & P_SA) 422 thread_statclock(1); 423 #endif 424 td->td_uticks++; 425 if (p->p_nice > NZERO) 426 cp_time[CP_NICE]++; 427 else 428 cp_time[CP_USER]++; 429 } else { 430 /* 431 * Came from kernel mode, so we were: 432 * - handling an interrupt, 433 * - doing syscall or trap work on behalf of the current 434 * user process, or 435 * - spinning in the idle loop. 436 * Whichever it is, charge the time as appropriate. 437 * Note that we charge interrupts to the current process, 438 * regardless of whether they are ``for'' that process, 439 * so that we know how much of its real time was spent 440 * in ``non-process'' (i.e., interrupt) work. 441 */ 442 if ((td->td_pflags & TDP_ITHREAD) || 443 td->td_intr_nesting_level >= 2) { 444 td->td_iticks++; 445 cp_time[CP_INTR]++; 446 } else { 447 #ifdef KSE 448 if (p->p_flag & P_SA) 449 thread_statclock(0); 450 #endif 451 td->td_pticks++; 452 td->td_sticks++; 453 if (!TD_IS_IDLETHREAD(td)) 454 cp_time[CP_SYS]++; 455 else 456 cp_time[CP_IDLE]++; 457 } 458 } 459 CTR4(KTR_SCHED, "statclock: %p(%s) prio %d stathz %d", 460 td, td->td_proc->p_comm, td->td_priority, (stathz)?stathz:hz); 461 462 sched_clock(td); 463 464 /* Update resource usage integrals and maximums. */ 465 MPASS(p->p_stats != NULL); 466 MPASS(p->p_vmspace != NULL); 467 vm = p->p_vmspace; 468 ru = &p->p_stats->p_ru; 469 ru->ru_ixrss += pgtok(vm->vm_tsize); 470 ru->ru_idrss += pgtok(vm->vm_dsize); 471 ru->ru_isrss += pgtok(vm->vm_ssize); 472 rss = pgtok(vmspace_resident_count(vm)); 473 if (ru->ru_maxrss < rss) 474 ru->ru_maxrss = rss; 475 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 476 } 477 478 void 479 profclock(int usermode, uintfptr_t pc) 480 { 481 struct thread *td; 482 #ifdef GPROF 483 struct gmonparam *g; 484 uintfptr_t i; 485 #endif 486 487 td = curthread; 488 if (usermode) { 489 /* 490 * Came from user mode; CPU was in user state. 491 * If this process is being profiled, record the tick. 492 * if there is no related user location yet, don't 493 * bother trying to count it. 494 */ 495 if (td->td_proc->p_flag & P_PROFIL) 496 addupc_intr(td, pc, 1); 497 } 498 #ifdef GPROF 499 else { 500 /* 501 * Kernel statistics are just like addupc_intr, only easier. 502 */ 503 g = &_gmonparam; 504 if (g->state == GMON_PROF_ON && pc >= g->lowpc) { 505 i = PC_TO_I(g, pc); 506 if (i < g->textsize) { 507 KCOUNT(g, i)++; 508 } 509 } 510 } 511 #endif 512 } 513 514 /* 515 * Return information about system clocks. 516 */ 517 static int 518 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 519 { 520 struct clockinfo clkinfo; 521 /* 522 * Construct clockinfo structure. 523 */ 524 bzero(&clkinfo, sizeof(clkinfo)); 525 clkinfo.hz = hz; 526 clkinfo.tick = tick; 527 clkinfo.profhz = profhz; 528 clkinfo.stathz = stathz ? stathz : hz; 529 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 530 } 531 532 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 533 0, 0, sysctl_kern_clockrate, "S,clockinfo", 534 "Rate and period of various kernel clocks"); 535 536 #ifdef SW_WATCHDOG 537 538 static void 539 watchdog_config(void *unused __unused, u_int cmd, int *error) 540 { 541 u_int u; 542 543 u = cmd & WD_INTERVAL; 544 if (u >= WD_TO_1SEC) { 545 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 546 watchdog_enabled = 1; 547 *error = 0; 548 } else { 549 watchdog_enabled = 0; 550 } 551 } 552 553 /* 554 * Handle a watchdog timeout by dumping interrupt information and 555 * then either dropping to DDB or panicing. 556 */ 557 static void 558 watchdog_fire(void) 559 { 560 int nintr; 561 u_int64_t inttotal; 562 u_long *curintr; 563 char *curname; 564 565 curintr = intrcnt; 566 curname = intrnames; 567 inttotal = 0; 568 nintr = eintrcnt - intrcnt; 569 570 printf("interrupt total\n"); 571 while (--nintr >= 0) { 572 if (*curintr) 573 printf("%-12s %20lu\n", curname, *curintr); 574 curname += strlen(curname) + 1; 575 inttotal += *curintr++; 576 } 577 printf("Total %20ju\n", (uintmax_t)inttotal); 578 579 #ifdef KDB 580 kdb_backtrace(); 581 kdb_enter("watchdog timeout"); 582 #else 583 panic("watchdog timeout"); 584 #endif /* KDB */ 585 } 586 587 #endif /* SW_WATCHDOG */ 588