1 /* $OpenBSD: kern_clock.c,v 1.41 2003/05/13 20:31:59 miod Exp $ */ 2 /* $NetBSD: kern_clock.c,v 1.34 1996/06/09 04:51:03 briggs Exp $ */ 3 4 /*- 5 * Copyright (c) 1982, 1986, 1991, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 42 */ 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/dkstat.h> 47 #include <sys/timeout.h> 48 #include <sys/kernel.h> 49 #include <sys/limits.h> 50 #include <sys/proc.h> 51 #include <sys/resourcevar.h> 52 #include <sys/signalvar.h> 53 #include <uvm/uvm_extern.h> 54 #include <sys/sysctl.h> 55 #include <sys/sched.h> 56 57 #include <machine/cpu.h> 58 59 #ifdef GPROF 60 #include <sys/gmon.h> 61 #endif 62 63 /* 64 * Clock handling routines. 65 * 66 * This code is written to operate with two timers that run independently of 67 * each other. The main clock, running hz times per second, is used to keep 68 * track of real time. The second timer handles kernel and user profiling, 69 * and does resource use estimation. If the second timer is programmable, 70 * it is randomized to avoid aliasing between the two clocks. For example, 71 * the randomization prevents an adversary from always giving up the cpu 72 * just before its quantum expires. Otherwise, it would never accumulate 73 * cpu ticks. The mean frequency of the second timer is stathz. 74 * 75 * If no second timer exists, stathz will be zero; in this case we drive 76 * profiling and statistics off the main clock. This WILL NOT be accurate; 77 * do not do it unless absolutely necessary. 78 * 79 * The statistics clock may (or may not) be run at a higher rate while 80 * profiling. This profile clock runs at profhz. We require that profhz 81 * be an integral multiple of stathz. 82 * 83 * If the statistics clock is running fast, it must be divided by the ratio 84 * profhz/stathz for statistics. (For profiling, every tick counts.) 85 */ 86 87 /* 88 * Bump a timeval by a small number of usec's. 89 */ 90 #define BUMPTIME(t, usec) { \ 91 register volatile struct timeval *tp = (t); \ 92 register long us; \ 93 \ 94 tp->tv_usec = us = tp->tv_usec + (usec); \ 95 if (us >= 1000000) { \ 96 tp->tv_usec = us - 1000000; \ 97 tp->tv_sec++; \ 98 } \ 99 } 100 101 int stathz; 102 int schedhz; 103 int profhz; 104 int profprocs; 105 int ticks; 106 static int psdiv, pscnt; /* prof => stat divider */ 107 int psratio; /* ratio: prof / stat */ 108 int tickfix, tickfixinterval; /* used if tick not really integral */ 109 static int tickfixcnt; /* accumulated fractional error */ 110 111 long cp_time[CPUSTATES]; 112 113 volatile struct timeval time 114 __attribute__((__aligned__(__alignof__(quad_t)))); 115 volatile struct timeval mono_time; 116 117 #ifdef __HAVE_GENERIC_SOFT_INTERRUPTS 118 void *softclock_si; 119 void generic_softclock(void *); 120 121 void 122 generic_softclock(void *ignore) 123 { 124 /* 125 * XXX - dont' commit, just a dummy wrapper until we learn everyone 126 * deal with a changed proto for softclock(). 127 */ 128 softclock(); 129 } 130 #endif 131 132 /* 133 * Initialize clock frequencies and start both clocks running. 134 */ 135 void 136 initclocks() 137 { 138 int i; 139 140 #ifdef __HAVE_GENERIC_SOFT_INTERRUPTS 141 softclock_si = softintr_establish(IPL_SOFTCLOCK, generic_softclock, NULL); 142 if (softclock_si == NULL) 143 panic("initclocks: unable to register softclock intr"); 144 #endif 145 146 /* 147 * Set divisors to 1 (normal case) and let the machine-specific 148 * code do its bit. 149 */ 150 psdiv = pscnt = 1; 151 cpu_initclocks(); 152 153 /* 154 * Compute profhz/stathz, and fix profhz if needed. 155 */ 156 i = stathz ? stathz : hz; 157 if (profhz == 0) 158 profhz = i; 159 psratio = profhz / i; 160 } 161 162 /* 163 * The real-time timer, interrupting hz times per second. 164 */ 165 void 166 hardclock(frame) 167 register struct clockframe *frame; 168 { 169 register struct proc *p; 170 register int delta; 171 extern int tickdelta; 172 extern long timedelta; 173 174 p = curproc; 175 if (p) { 176 register struct pstats *pstats; 177 178 /* 179 * Run current process's virtual and profile time, as needed. 180 */ 181 pstats = p->p_stats; 182 if (CLKF_USERMODE(frame) && 183 timerisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 184 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 185 psignal(p, SIGVTALRM); 186 if (timerisset(&pstats->p_timer[ITIMER_PROF].it_value) && 187 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 188 psignal(p, SIGPROF); 189 } 190 191 /* 192 * If no separate statistics clock is available, run it from here. 193 */ 194 if (stathz == 0) 195 statclock(frame); 196 197 /* 198 * Increment the time-of-day. The increment is normally just 199 * ``tick''. If the machine is one which has a clock frequency 200 * such that ``hz'' would not divide the second evenly into 201 * milliseconds, a periodic adjustment must be applied. Finally, 202 * if we are still adjusting the time (see adjtime()), 203 * ``tickdelta'' may also be added in. 204 */ 205 ticks++; 206 delta = tick; 207 208 if (tickfix) { 209 tickfixcnt += tickfix; 210 if (tickfixcnt >= tickfixinterval) { 211 delta++; 212 tickfixcnt -= tickfixinterval; 213 } 214 } 215 /* Imprecise 4bsd adjtime() handling */ 216 if (timedelta != 0) { 217 delta += tickdelta; 218 timedelta -= tickdelta; 219 } 220 221 #ifdef notyet 222 microset(); 223 #endif 224 225 BUMPTIME(&time, delta); 226 BUMPTIME(&mono_time, delta); 227 228 #ifdef CPU_CLOCKUPDATE 229 CPU_CLOCKUPDATE(); 230 #endif 231 232 /* 233 * Update real-time timeout queue. 234 * Process callouts at a very low cpu priority, so we don't keep the 235 * relatively high clock interrupt priority any longer than necessary. 236 */ 237 if (timeout_hardclock_update()) { 238 #ifdef __HAVE_GENERIC_SOFT_INTERRUPTS 239 softintr_schedule(softclock_si); 240 #else 241 setsoftclock(); 242 #endif 243 } 244 } 245 246 /* 247 * Compute number of hz until specified time. Used to 248 * compute the second argument to timeout_add() from an absolute time. 249 */ 250 int 251 hzto(tv) 252 struct timeval *tv; 253 { 254 unsigned long ticks; 255 long sec, usec; 256 int s; 257 258 /* 259 * If the number of usecs in the whole seconds part of the time 260 * difference fits in a long, then the total number of usecs will 261 * fit in an unsigned long. Compute the total and convert it to 262 * ticks, rounding up and adding 1 to allow for the current tick 263 * to expire. Rounding also depends on unsigned long arithmetic 264 * to avoid overflow. 265 * 266 * Otherwise, if the number of ticks in the whole seconds part of 267 * the time difference fits in a long, then convert the parts to 268 * ticks separately and add, using similar rounding methods and 269 * overflow avoidance. This method would work in the previous 270 * case but it is slightly slower and assumes that hz is integral. 271 * 272 * Otherwise, round the time difference down to the maximum 273 * representable value. 274 * 275 * If ints have 32 bits, then the maximum value for any timeout in 276 * 10ms ticks is 248 days. 277 */ 278 s = splhigh(); 279 sec = tv->tv_sec - time.tv_sec; 280 usec = tv->tv_usec - time.tv_usec; 281 splx(s); 282 if (usec < 0) { 283 sec--; 284 usec += 1000000; 285 } 286 if (sec < 0 || (sec == 0 && usec <= 0)) { 287 ticks = 0; 288 } else if (sec <= LONG_MAX / 1000000) 289 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 290 / tick + 1; 291 else if (sec <= LONG_MAX / hz) 292 ticks = sec * hz 293 + ((unsigned long)usec + (tick - 1)) / tick + 1; 294 else 295 ticks = LONG_MAX; 296 if (ticks > INT_MAX) 297 ticks = INT_MAX; 298 return ((int)ticks); 299 } 300 301 /* 302 * Compute number of hz in the specified amount of time. 303 */ 304 int 305 tvtohz(struct timeval *tv) 306 { 307 unsigned long ticks; 308 long sec, usec; 309 310 /* 311 * If the number of usecs in the whole seconds part of the time 312 * fits in a long, then the total number of usecs will 313 * fit in an unsigned long. Compute the total and convert it to 314 * ticks, rounding up and adding 1 to allow for the current tick 315 * to expire. Rounding also depends on unsigned long arithmetic 316 * to avoid overflow. 317 * 318 * Otherwise, if the number of ticks in the whole seconds part of 319 * the time fits in a long, then convert the parts to 320 * ticks separately and add, using similar rounding methods and 321 * overflow avoidance. This method would work in the previous 322 * case but it is slightly slower and assumes that hz is integral. 323 * 324 * Otherwise, round the time down to the maximum 325 * representable value. 326 * 327 * If ints have 32 bits, then the maximum value for any timeout in 328 * 10ms ticks is 248 days. 329 */ 330 sec = tv->tv_sec; 331 usec = tv->tv_usec; 332 if (sec < 0 || (sec == 0 && usec <= 0)) 333 ticks = 0; 334 else if (sec <= LONG_MAX / 1000000) 335 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 336 / tick + 1; 337 else if (sec <= LONG_MAX / hz) 338 ticks = sec * hz 339 + ((unsigned long)usec + (tick - 1)) / tick + 1; 340 else 341 ticks = LONG_MAX; 342 if (ticks > INT_MAX) 343 ticks = INT_MAX; 344 return ((int)ticks); 345 } 346 347 /* 348 * Start profiling on a process. 349 * 350 * Kernel profiling passes proc0 which never exits and hence 351 * keeps the profile clock running constantly. 352 */ 353 void 354 startprofclock(p) 355 register struct proc *p; 356 { 357 int s; 358 359 if ((p->p_flag & P_PROFIL) == 0) { 360 p->p_flag |= P_PROFIL; 361 if (++profprocs == 1 && stathz != 0) { 362 s = splstatclock(); 363 psdiv = pscnt = psratio; 364 setstatclockrate(profhz); 365 splx(s); 366 } 367 } 368 } 369 370 /* 371 * Stop profiling on a process. 372 */ 373 void 374 stopprofclock(p) 375 register struct proc *p; 376 { 377 int s; 378 379 if (p->p_flag & P_PROFIL) { 380 p->p_flag &= ~P_PROFIL; 381 if (--profprocs == 0 && stathz != 0) { 382 s = splstatclock(); 383 psdiv = pscnt = 1; 384 setstatclockrate(stathz); 385 splx(s); 386 } 387 } 388 } 389 390 /* 391 * Statistics clock. Grab profile sample, and if divider reaches 0, 392 * do process and kernel statistics. 393 */ 394 void 395 statclock(frame) 396 register struct clockframe *frame; 397 { 398 #ifdef GPROF 399 register struct gmonparam *g; 400 register int i; 401 #endif 402 static int schedclk; 403 register struct proc *p; 404 405 if (CLKF_USERMODE(frame)) { 406 p = curproc; 407 if (p->p_flag & P_PROFIL) 408 addupc_intr(p, CLKF_PC(frame)); 409 if (--pscnt > 0) 410 return; 411 /* 412 * Came from user mode; CPU was in user state. 413 * If this process is being profiled record the tick. 414 */ 415 p->p_uticks++; 416 if (p->p_nice > NZERO) 417 cp_time[CP_NICE]++; 418 else 419 cp_time[CP_USER]++; 420 } else { 421 #ifdef GPROF 422 /* 423 * Kernel statistics are just like addupc_intr, only easier. 424 */ 425 g = &_gmonparam; 426 if (g->state == GMON_PROF_ON) { 427 i = CLKF_PC(frame) - g->lowpc; 428 if (i < g->textsize) { 429 i /= HISTFRACTION * sizeof(*g->kcount); 430 g->kcount[i]++; 431 } 432 } 433 #endif 434 if (--pscnt > 0) 435 return; 436 /* 437 * Came from kernel mode, so we were: 438 * - handling an interrupt, 439 * - doing syscall or trap work on behalf of the current 440 * user process, or 441 * - spinning in the idle loop. 442 * Whichever it is, charge the time as appropriate. 443 * Note that we charge interrupts to the current process, 444 * regardless of whether they are ``for'' that process, 445 * so that we know how much of its real time was spent 446 * in ``non-process'' (i.e., interrupt) work. 447 */ 448 p = curproc; 449 if (CLKF_INTR(frame)) { 450 if (p != NULL) 451 p->p_iticks++; 452 cp_time[CP_INTR]++; 453 } else if (p != NULL) { 454 p->p_sticks++; 455 cp_time[CP_SYS]++; 456 } else 457 cp_time[CP_IDLE]++; 458 } 459 pscnt = psdiv; 460 461 if (p != NULL) { 462 p->p_cpticks++; 463 /* 464 * If no schedclock is provided, call it here at ~~12-25 Hz; 465 * ~~16 Hz is best 466 */ 467 if (schedhz == 0) 468 if ((++schedclk & 3) == 0) 469 schedclock(p); 470 } 471 } 472 473 /* 474 * Return information about system clocks. 475 */ 476 int 477 sysctl_clockrate(where, sizep) 478 register char *where; 479 size_t *sizep; 480 { 481 struct clockinfo clkinfo; 482 483 /* 484 * Construct clockinfo structure. 485 */ 486 clkinfo.tick = tick; 487 clkinfo.tickadj = tickadj; 488 clkinfo.hz = hz; 489 clkinfo.profhz = profhz; 490 clkinfo.stathz = stathz ? stathz : hz; 491 return (sysctl_rdstruct(where, sizep, NULL, &clkinfo, sizeof(clkinfo))); 492 } 493