kern_clock.c revision 65557
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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 39 * $FreeBSD: head/sys/kern/kern_clock.c 65557 2000-09-07 01:33:02Z jasone $ 40 */ 41 42#include "opt_ntp.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/dkstat.h> 47#include <sys/callout.h> 48#include <sys/kernel.h> 49#include <sys/proc.h> 50#include <sys/resourcevar.h> 51#include <sys/signalvar.h> 52#include <sys/timetc.h> 53#include <sys/timepps.h> 54#include <vm/vm.h> 55#include <sys/lock.h> 56#include <vm/pmap.h> 57#include <vm/vm_map.h> 58#include <sys/sysctl.h> 59 60#include <machine/cpu.h> 61#include <machine/limits.h> 62#include <machine/smp.h> 63 64#ifdef GPROF 65#include <sys/gmon.h> 66#endif 67 68 69static void initclocks __P((void *dummy)); 70SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) 71 72/* Some of these don't belong here, but it's easiest to concentrate them. */ 73long cp_time[CPUSTATES]; 74 75long tk_cancc; 76long tk_nin; 77long tk_nout; 78long tk_rawcc; 79 80/* 81 * Clock handling routines. 82 * 83 * This code is written to operate with two timers that run independently of 84 * each other. 85 * 86 * The main timer, running hz times per second, is used to trigger interval 87 * timers, timeouts and rescheduling as needed. 88 * 89 * The second timer handles kernel and user profiling, 90 * and does resource use estimation. If the second timer is programmable, 91 * it is randomized to avoid aliasing between the two clocks. For example, 92 * the randomization prevents an adversary from always giving up the cpu 93 * just before its quantum expires. Otherwise, it would never accumulate 94 * cpu ticks. The mean frequency of the second timer is stathz. 95 * 96 * If no second timer exists, stathz will be zero; in this case we drive 97 * profiling and statistics off the main clock. This WILL NOT be accurate; 98 * do not do it unless absolutely necessary. 99 * 100 * The statistics clock may (or may not) be run at a higher rate while 101 * profiling. This profile clock runs at profhz. We require that profhz 102 * be an integral multiple of stathz. 103 * 104 * If the statistics clock is running fast, it must be divided by the ratio 105 * profhz/stathz for statistics. (For profiling, every tick counts.) 106 * 107 * Time-of-day is maintained using a "timecounter", which may or may 108 * not be related to the hardware generating the above mentioned 109 * interrupts. 110 */ 111 112int stathz; 113int profhz; 114static int profprocs; 115int ticks; 116static int psdiv, pscnt; /* prof => stat divider */ 117int psratio; /* ratio: prof / stat */ 118 119/* 120 * Initialize clock frequencies and start both clocks running. 121 */ 122/* ARGSUSED*/ 123static void 124initclocks(dummy) 125 void *dummy; 126{ 127 register int i; 128 129 /* 130 * Set divisors to 1 (normal case) and let the machine-specific 131 * code do its bit. 132 */ 133 psdiv = pscnt = 1; 134 cpu_initclocks(); 135 136 /* 137 * Compute profhz/stathz, and fix profhz if needed. 138 */ 139 i = stathz ? stathz : hz; 140 if (profhz == 0) 141 profhz = i; 142 psratio = profhz / i; 143} 144 145/* 146 * The real-time timer, interrupting hz times per second. 147 */ 148void 149hardclock(frame) 150 register struct clockframe *frame; 151{ 152 register struct proc *p; 153 154 p = curproc; 155 if (p != idleproc) { 156 register struct pstats *pstats; 157 158 /* 159 * Run current process's virtual and profile time, as needed. 160 */ 161 pstats = p->p_stats; 162 if (CLKF_USERMODE(frame) && 163 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 164 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 165 psignal(p, SIGVTALRM); 166 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 167 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 168 psignal(p, SIGPROF); 169 } 170 171#if defined(SMP) && defined(BETTER_CLOCK) 172 forward_hardclock(pscnt); 173#endif 174 175 /* 176 * If no separate statistics clock is available, run it from here. 177 */ 178 if (stathz == 0) 179 statclock(frame); 180 181 tc_windup(); 182 ticks++; 183 184 /* 185 * Process callouts at a very low cpu priority, so we don't keep the 186 * relatively high clock interrupt priority any longer than necessary. 187 */ 188 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) { 189 if (CLKF_BASEPRI(frame)) { 190 /* 191 * Save the overhead of a software interrupt; 192 * it will happen as soon as we return, so do it now. 193 */ 194 (void)splsoftclock(); 195 softclock(); 196 } else 197 setsoftclock(); 198 } else if (softticks + 1 == ticks) 199 ++softticks; 200} 201 202/* 203 * Compute number of ticks in the specified amount of time. 204 */ 205int 206tvtohz(tv) 207 struct timeval *tv; 208{ 209 register unsigned long ticks; 210 register long sec, usec; 211 212 /* 213 * If the number of usecs in the whole seconds part of the time 214 * difference fits in a long, then the total number of usecs will 215 * fit in an unsigned long. Compute the total and convert it to 216 * ticks, rounding up and adding 1 to allow for the current tick 217 * to expire. Rounding also depends on unsigned long arithmetic 218 * to avoid overflow. 219 * 220 * Otherwise, if the number of ticks in the whole seconds part of 221 * the time difference fits in a long, then convert the parts to 222 * ticks separately and add, using similar rounding methods and 223 * overflow avoidance. This method would work in the previous 224 * case but it is slightly slower and assumes that hz is integral. 225 * 226 * Otherwise, round the time difference down to the maximum 227 * representable value. 228 * 229 * If ints have 32 bits, then the maximum value for any timeout in 230 * 10ms ticks is 248 days. 231 */ 232 sec = tv->tv_sec; 233 usec = tv->tv_usec; 234 if (usec < 0) { 235 sec--; 236 usec += 1000000; 237 } 238 if (sec < 0) { 239#ifdef DIAGNOSTIC 240 if (usec > 0) { 241 sec++; 242 usec -= 1000000; 243 } 244 printf("tvotohz: negative time difference %ld sec %ld usec\n", 245 sec, usec); 246#endif 247 ticks = 1; 248 } else if (sec <= LONG_MAX / 1000000) 249 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 250 / tick + 1; 251 else if (sec <= LONG_MAX / hz) 252 ticks = sec * hz 253 + ((unsigned long)usec + (tick - 1)) / tick + 1; 254 else 255 ticks = LONG_MAX; 256 if (ticks > INT_MAX) 257 ticks = INT_MAX; 258 return ((int)ticks); 259} 260 261/* 262 * Start profiling on a process. 263 * 264 * Kernel profiling passes proc0 which never exits and hence 265 * keeps the profile clock running constantly. 266 */ 267void 268startprofclock(p) 269 register struct proc *p; 270{ 271 int s; 272 273 if ((p->p_flag & P_PROFIL) == 0) { 274 p->p_flag |= P_PROFIL; 275 if (++profprocs == 1 && stathz != 0) { 276 s = splstatclock(); 277 psdiv = pscnt = psratio; 278 setstatclockrate(profhz); 279 splx(s); 280 } 281 } 282} 283 284/* 285 * Stop profiling on a process. 286 */ 287void 288stopprofclock(p) 289 register struct proc *p; 290{ 291 int s; 292 293 if (p->p_flag & P_PROFIL) { 294 p->p_flag &= ~P_PROFIL; 295 if (--profprocs == 0 && stathz != 0) { 296 s = splstatclock(); 297 psdiv = pscnt = 1; 298 setstatclockrate(stathz); 299 splx(s); 300 } 301 } 302} 303 304/* 305 * Statistics clock. Grab profile sample, and if divider reaches 0, 306 * do process and kernel statistics. Most of the statistics are only 307 * used by user-level statistics programs. The main exceptions are 308 * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. 309 */ 310void 311statclock(frame) 312 register struct clockframe *frame; 313{ 314#ifdef GPROF 315 register struct gmonparam *g; 316 int i; 317#endif 318 register struct proc *p; 319 struct pstats *pstats; 320 long rss; 321 struct rusage *ru; 322 struct vmspace *vm; 323 324 if (CLKF_USERMODE(frame)) { 325 /* 326 * Came from user mode; CPU was in user state. 327 * If this process is being profiled, record the tick. 328 */ 329 p = prevproc; 330 if (p->p_flag & P_PROFIL) 331 addupc_intr(p, CLKF_PC(frame), 1); 332#if defined(SMP) && defined(BETTER_CLOCK) 333 if (stathz != 0) 334 forward_statclock(pscnt); 335#endif 336 if (--pscnt > 0) 337 return; 338 /* 339 * Charge the time as appropriate. 340 */ 341 p->p_uticks++; 342 if (p->p_nice > NZERO) 343 cp_time[CP_NICE]++; 344 else 345 cp_time[CP_USER]++; 346 } else { 347#ifdef GPROF 348 /* 349 * Kernel statistics are just like addupc_intr, only easier. 350 */ 351 g = &_gmonparam; 352 if (g->state == GMON_PROF_ON) { 353 i = CLKF_PC(frame) - g->lowpc; 354 if (i < g->textsize) { 355 i /= HISTFRACTION * sizeof(*g->kcount); 356 g->kcount[i]++; 357 } 358 } 359#endif 360#if defined(SMP) && defined(BETTER_CLOCK) 361 if (stathz != 0) 362 forward_statclock(pscnt); 363#endif 364 if (--pscnt > 0) 365 return; 366 /* 367 * Came from kernel mode, so we were: 368 * - handling an interrupt, 369 * - doing syscall or trap work on behalf of the current 370 * user process, or 371 * - spinning in the idle loop. 372 * Whichever it is, charge the time as appropriate. 373 * Note that we charge interrupts to the current process, 374 * regardless of whether they are ``for'' that process, 375 * so that we know how much of its real time was spent 376 * in ``non-process'' (i.e., interrupt) work. 377 */ 378 p = prevproc; 379 if (p->p_ithd) { 380 p->p_iticks++; 381 cp_time[CP_INTR]++; 382 } else { 383 p->p_sticks++; 384 if (p != idleproc) 385 cp_time[CP_SYS]++; 386 else 387 cp_time[CP_IDLE]++; 388 } 389 } 390 pscnt = psdiv; 391 392 if (p != idleproc) { 393 schedclock(p); 394 395 /* Update resource usage integrals and maximums. */ 396 if ((pstats = p->p_stats) != NULL && 397 (ru = &pstats->p_ru) != NULL && 398 (vm = p->p_vmspace) != NULL) { 399 ru->ru_ixrss += pgtok(vm->vm_tsize); 400 ru->ru_idrss += pgtok(vm->vm_dsize); 401 ru->ru_isrss += pgtok(vm->vm_ssize); 402 rss = pgtok(vmspace_resident_count(vm)); 403 if (ru->ru_maxrss < rss) 404 ru->ru_maxrss = rss; 405 } 406 } 407} 408 409/* 410 * Return information about system clocks. 411 */ 412static int 413sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 414{ 415 struct clockinfo clkinfo; 416 /* 417 * Construct clockinfo structure. 418 */ 419 clkinfo.hz = hz; 420 clkinfo.tick = tick; 421 clkinfo.tickadj = tickadj; 422 clkinfo.profhz = profhz; 423 clkinfo.stathz = stathz ? stathz : hz; 424 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 425} 426 427SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 428 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); 429