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_ntp.h" 41 #include "opt_watchdog.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/kdb.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/ktr.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/resource.h> 53 #include <sys/resourcevar.h> 54 #include <sys/sched.h> 55 #include <sys/signalvar.h> 56 #include <sys/smp.h> 57 #include <vm/vm.h> 58 #include <vm/pmap.h> 59 #include <vm/vm_map.h> 60 #include <sys/sysctl.h> 61 #include <sys/bus.h> 62 #include <sys/interrupt.h> 63 #include <sys/limits.h> 64 #include <sys/timetc.h> 65 66 #include <machine/cpu.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 SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time), 87 "LU", "CPU time statistics"); 88 89 #ifdef SW_WATCHDOG 90 #include <sys/watchdog.h> 91 92 static int watchdog_ticks; 93 static int watchdog_enabled; 94 static void watchdog_fire(void); 95 static void watchdog_config(void *, u_int, int *); 96 #endif /* SW_WATCHDOG */ 97 98 /* 99 * Clock handling routines. 100 * 101 * This code is written to operate with two timers that run independently of 102 * each other. 103 * 104 * The main timer, running hz times per second, is used to trigger interval 105 * timers, timeouts and rescheduling as needed. 106 * 107 * The second timer handles kernel and user profiling, 108 * and does resource use estimation. If the second timer is programmable, 109 * it is randomized to avoid aliasing between the two clocks. For example, 110 * the randomization prevents an adversary from always giving up the cpu 111 * just before its quantum expires. Otherwise, it would never accumulate 112 * cpu ticks. The mean frequency of the second timer is stathz. 113 * 114 * If no second timer exists, stathz will be zero; in this case we drive 115 * profiling and statistics off the main clock. This WILL NOT be accurate; 116 * do not do it unless absolutely necessary. 117 * 118 * The statistics clock may (or may not) be run at a higher rate while 119 * profiling. This profile clock runs at profhz. We require that profhz 120 * be an integral multiple of stathz. 121 * 122 * If the statistics clock is running fast, it must be divided by the ratio 123 * profhz/stathz for statistics. (For profiling, every tick counts.) 124 * 125 * Time-of-day is maintained using a "timecounter", which may or may 126 * not be related to the hardware generating the above mentioned 127 * interrupts. 128 */ 129 130 int stathz; 131 int profhz; 132 int profprocs; 133 int ticks; 134 int psratio; 135 136 /* 137 * Initialize clock frequencies and start both clocks running. 138 */ 139 /* ARGSUSED*/ 140 static void 141 initclocks(dummy) 142 void *dummy; 143 { 144 register int i; 145 146 /* 147 * Set divisors to 1 (normal case) and let the machine-specific 148 * code do its bit. 149 */ 150 cpu_initclocks(); 151 152 /* 153 * Compute profhz/stathz, and fix profhz if needed. 154 */ 155 i = stathz ? stathz : hz; 156 if (profhz == 0) 157 profhz = i; 158 psratio = profhz / i; 159 #ifdef SW_WATCHDOG 160 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); 161 #endif 162 } 163 164 /* 165 * Each time the real-time timer fires, this function is called on all CPUs. 166 * Note that hardclock() calls hardclock_process() for the boot CPU, so only 167 * the other CPUs in the system need to call this function. 168 */ 169 void 170 hardclock_process(frame) 171 register struct clockframe *frame; 172 { 173 struct pstats *pstats; 174 struct thread *td = curthread; 175 struct proc *p = td->td_proc; 176 177 /* 178 * Run current process's virtual and profile time, as needed. 179 */ 180 mtx_lock_spin_flags(&sched_lock, MTX_QUIET); 181 if (p->p_flag & P_SA) { 182 /* XXXKSE What to do? */ 183 } else { 184 pstats = p->p_stats; 185 if (CLKF_USERMODE(frame) && 186 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 187 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 188 p->p_sflag |= PS_ALRMPEND; 189 td->td_flags |= TDF_ASTPENDING; 190 } 191 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 192 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 193 p->p_sflag |= PS_PROFPEND; 194 td->td_flags |= TDF_ASTPENDING; 195 } 196 } 197 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 198 199 #ifdef HWPMC_HOOKS 200 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) 201 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); 202 #endif 203 } 204 205 /* 206 * The real-time timer, interrupting hz times per second. 207 */ 208 void 209 hardclock(frame) 210 register struct clockframe *frame; 211 { 212 int need_softclock = 0; 213 214 CTR0(KTR_CLK, "hardclock fired"); 215 hardclock_process(frame); 216 217 tc_ticktock(); 218 /* 219 * If no separate statistics clock is available, run it from here. 220 * 221 * XXX: this only works for UP 222 */ 223 if (stathz == 0) { 224 profclock(frame); 225 statclock(frame); 226 } 227 228 #ifdef DEVICE_POLLING 229 hardclock_device_poll(); /* this is very short and quick */ 230 #endif /* DEVICE_POLLING */ 231 232 /* 233 * Process callouts at a very low cpu priority, so we don't keep the 234 * relatively high clock interrupt priority any longer than necessary. 235 */ 236 mtx_lock_spin_flags(&callout_lock, MTX_QUIET); 237 ticks++; 238 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) { 239 need_softclock = 1; 240 } else if (softticks + 1 == ticks) 241 ++softticks; 242 mtx_unlock_spin_flags(&callout_lock, MTX_QUIET); 243 244 /* 245 * swi_sched acquires sched_lock, so we don't want to call it with 246 * callout_lock held; incorrect locking order. 247 */ 248 if (need_softclock) 249 swi_sched(softclock_ih, 0); 250 251 #ifdef SW_WATCHDOG 252 if (watchdog_enabled > 0 && --watchdog_ticks <= 0) 253 watchdog_fire(); 254 #endif /* SW_WATCHDOG */ 255 } 256 257 /* 258 * Compute number of ticks in the specified amount of time. 259 */ 260 int 261 tvtohz(tv) 262 struct timeval *tv; 263 { 264 register unsigned long ticks; 265 register long sec, usec; 266 267 /* 268 * If the number of usecs in the whole seconds part of the time 269 * difference fits in a long, then the total number of usecs will 270 * fit in an unsigned long. Compute the total and convert it to 271 * ticks, rounding up and adding 1 to allow for the current tick 272 * to expire. Rounding also depends on unsigned long arithmetic 273 * to avoid overflow. 274 * 275 * Otherwise, if the number of ticks in the whole seconds part of 276 * the time difference fits in a long, then convert the parts to 277 * ticks separately and add, using similar rounding methods and 278 * overflow avoidance. This method would work in the previous 279 * case but it is slightly slower and assumes that hz is integral. 280 * 281 * Otherwise, round the time difference down to the maximum 282 * representable value. 283 * 284 * If ints have 32 bits, then the maximum value for any timeout in 285 * 10ms ticks is 248 days. 286 */ 287 sec = tv->tv_sec; 288 usec = tv->tv_usec; 289 if (usec < 0) { 290 sec--; 291 usec += 1000000; 292 } 293 if (sec < 0) { 294 #ifdef DIAGNOSTIC 295 if (usec > 0) { 296 sec++; 297 usec -= 1000000; 298 } 299 printf("tvotohz: negative time difference %ld sec %ld usec\n", 300 sec, usec); 301 #endif 302 ticks = 1; 303 } else if (sec <= LONG_MAX / 1000000) 304 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 305 / tick + 1; 306 else if (sec <= LONG_MAX / hz) 307 ticks = sec * hz 308 + ((unsigned long)usec + (tick - 1)) / tick + 1; 309 else 310 ticks = LONG_MAX; 311 if (ticks > INT_MAX) 312 ticks = INT_MAX; 313 return ((int)ticks); 314 } 315 316 /* 317 * Start profiling on a process. 318 * 319 * Kernel profiling passes proc0 which never exits and hence 320 * keeps the profile clock running constantly. 321 */ 322 void 323 startprofclock(p) 324 register struct proc *p; 325 { 326 327 /* 328 * XXX; Right now sched_lock protects statclock(), but perhaps 329 * it should be protected later on by a time_lock, which would 330 * cover psdiv, etc. as well. 331 */ 332 PROC_LOCK_ASSERT(p, MA_OWNED); 333 if (p->p_flag & P_STOPPROF) 334 return; 335 if ((p->p_flag & P_PROFIL) == 0) { 336 mtx_lock_spin(&sched_lock); 337 p->p_flag |= P_PROFIL; 338 if (++profprocs == 1) 339 cpu_startprofclock(); 340 mtx_unlock_spin(&sched_lock); 341 } 342 } 343 344 /* 345 * Stop profiling on a process. 346 */ 347 void 348 stopprofclock(p) 349 register struct proc *p; 350 { 351 352 PROC_LOCK_ASSERT(p, MA_OWNED); 353 if (p->p_flag & P_PROFIL) { 354 if (p->p_profthreads != 0) { 355 p->p_flag |= P_STOPPROF; 356 while (p->p_profthreads != 0) 357 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, 358 "stopprof", 0); 359 p->p_flag &= ~P_STOPPROF; 360 } 361 if ((p->p_flag & P_PROFIL) == 0) 362 return; 363 mtx_lock_spin(&sched_lock); 364 p->p_flag &= ~P_PROFIL; 365 if (--profprocs == 0) 366 cpu_stopprofclock(); 367 mtx_unlock_spin(&sched_lock); 368 } 369 } 370 371 /* 372 * Statistics clock. Grab profile sample, and if divider reaches 0, 373 * do process and kernel statistics. Most of the statistics are only 374 * used by user-level statistics programs. The main exceptions are 375 * ke->ke_uticks, p->p_rux.rux_sticks, p->p_rux.rux_iticks, and p->p_estcpu. 376 * This should be called by all active processors. 377 */ 378 void 379 statclock(frame) 380 register struct clockframe *frame; 381 { 382 struct rusage *ru; 383 struct vmspace *vm; 384 struct thread *td; 385 struct proc *p; 386 long rss; 387 388 td = curthread; 389 p = td->td_proc; 390 391 mtx_lock_spin_flags(&sched_lock, MTX_QUIET); 392 if (CLKF_USERMODE(frame)) { 393 /* 394 * Charge the time as appropriate. 395 */ 396 if (p->p_flag & P_SA) 397 thread_statclock(1); 398 p->p_rux.rux_uticks++; 399 if (p->p_nice > NZERO) 400 cp_time[CP_NICE]++; 401 else 402 cp_time[CP_USER]++; 403 } else { 404 /* 405 * Came from kernel mode, so we were: 406 * - handling an interrupt, 407 * - doing syscall or trap work on behalf of the current 408 * user process, or 409 * - spinning in the idle loop. 410 * Whichever it is, charge the time as appropriate. 411 * Note that we charge interrupts to the current process, 412 * regardless of whether they are ``for'' that process, 413 * so that we know how much of its real time was spent 414 * in ``non-process'' (i.e., interrupt) work. 415 */ 416 if ((td->td_ithd != NULL) || td->td_intr_nesting_level >= 2) { 417 p->p_rux.rux_iticks++; 418 cp_time[CP_INTR]++; 419 } else { 420 if (p->p_flag & P_SA) 421 thread_statclock(0); 422 td->td_sticks++; 423 p->p_rux.rux_sticks++; 424 if (p != PCPU_GET(idlethread)->td_proc) 425 cp_time[CP_SYS]++; 426 else 427 cp_time[CP_IDLE]++; 428 } 429 } 430 CTR4(KTR_SCHED, "statclock: %p(%s) prio %d stathz %d", 431 td, td->td_proc->p_comm, td->td_priority, (stathz)?stathz:hz); 432 433 sched_clock(td); 434 435 /* Update resource usage integrals and maximums. */ 436 MPASS(p->p_stats != NULL); 437 MPASS(p->p_vmspace != NULL); 438 vm = p->p_vmspace; 439 ru = &p->p_stats->p_ru; 440 ru->ru_ixrss += pgtok(vm->vm_tsize); 441 ru->ru_idrss += pgtok(vm->vm_dsize); 442 ru->ru_isrss += pgtok(vm->vm_ssize); 443 rss = pgtok(vmspace_resident_count(vm)); 444 if (ru->ru_maxrss < rss) 445 ru->ru_maxrss = rss; 446 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET); 447 } 448 449 void 450 profclock(frame) 451 register struct clockframe *frame; 452 { 453 struct thread *td; 454 #ifdef GPROF 455 struct gmonparam *g; 456 int i; 457 #endif 458 459 td = curthread; 460 if (CLKF_USERMODE(frame)) { 461 /* 462 * Came from user mode; CPU was in user state. 463 * If this process is being profiled, record the tick. 464 * if there is no related user location yet, don't 465 * bother trying to count it. 466 */ 467 if (td->td_proc->p_flag & P_PROFIL) 468 addupc_intr(td, CLKF_PC(frame), 1); 469 } 470 #ifdef GPROF 471 else { 472 /* 473 * Kernel statistics are just like addupc_intr, only easier. 474 */ 475 g = &_gmonparam; 476 if (g->state == GMON_PROF_ON) { 477 i = CLKF_PC(frame) - g->lowpc; 478 if (i < g->textsize) { 479 i /= HISTFRACTION * sizeof(*g->kcount); 480 g->kcount[i]++; 481 } 482 } 483 } 484 #endif 485 } 486 487 /* 488 * Return information about system clocks. 489 */ 490 static int 491 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 492 { 493 struct clockinfo clkinfo; 494 /* 495 * Construct clockinfo structure. 496 */ 497 bzero(&clkinfo, sizeof(clkinfo)); 498 clkinfo.hz = hz; 499 clkinfo.tick = tick; 500 clkinfo.profhz = profhz; 501 clkinfo.stathz = stathz ? stathz : hz; 502 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 503 } 504 505 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 506 0, 0, sysctl_kern_clockrate, "S,clockinfo", 507 "Rate and period of various kernel clocks"); 508 509 #ifdef SW_WATCHDOG 510 511 static void 512 watchdog_config(void *unused __unused, u_int cmd, int *err) 513 { 514 u_int u; 515 516 u = cmd & WD_INTERVAL; 517 if (cmd && u >= WD_TO_1SEC) { 518 u = cmd & WD_INTERVAL; 519 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; 520 watchdog_enabled = 1; 521 *err = 0; 522 } else { 523 watchdog_enabled = 0; 524 } 525 } 526 527 /* 528 * Handle a watchdog timeout by dumping interrupt information and 529 * then either dropping to DDB or panicing. 530 */ 531 static void 532 watchdog_fire(void) 533 { 534 int nintr; 535 u_int64_t inttotal; 536 u_long *curintr; 537 char *curname; 538 539 curintr = intrcnt; 540 curname = intrnames; 541 inttotal = 0; 542 nintr = eintrcnt - intrcnt; 543 544 printf("interrupt total\n"); 545 while (--nintr >= 0) { 546 if (*curintr) 547 printf("%-12s %20lu\n", curname, *curintr); 548 curname += strlen(curname) + 1; 549 inttotal += *curintr++; 550 } 551 printf("Total %20ju\n", (uintmax_t)inttotal); 552 553 #ifdef KDB 554 kdb_backtrace(); 555 kdb_enter("watchdog timeout"); 556 #else 557 panic("watchdog timeout"); 558 #endif /* KDB */ 559 } 560 561 #endif /* SW_WATCHDOG */ 562