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 * $FreeBSD$ 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/ipl.h> 49 #include <sys/kernel.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/resourcevar.h> 53 #include <sys/signalvar.h> 54 #include <sys/timetc.h> 55 #include <sys/timepps.h> 56 #include <vm/vm.h> 57 #include <sys/lock.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 64 #include <machine/cpu.h> 65 #include <machine/limits.h> 66 #include <machine/smp.h> 67 68 #ifdef GPROF 69 #include <sys/gmon.h> 70 #endif 71 72 73 static void initclocks __P((void *dummy)); 74 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) 75 76 /* Some of these don't belong here, but it's easiest to concentrate them. */ 77 long cp_time[CPUSTATES]; 78 79 SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time), 80 "LU", "CPU time statistics"); 81 82 long tk_cancc; 83 long tk_nin; 84 long tk_nout; 85 long tk_rawcc; 86 87 /* 88 * Clock handling routines. 89 * 90 * This code is written to operate with two timers that run independently of 91 * each other. 92 * 93 * The main timer, running hz times per second, is used to trigger interval 94 * timers, timeouts and rescheduling as needed. 95 * 96 * The second timer handles kernel and user profiling, 97 * and does resource use estimation. If the second timer is programmable, 98 * it is randomized to avoid aliasing between the two clocks. For example, 99 * the randomization prevents an adversary from always giving up the cpu 100 * just before its quantum expires. Otherwise, it would never accumulate 101 * cpu ticks. The mean frequency of the second timer is stathz. 102 * 103 * If no second timer exists, stathz will be zero; in this case we drive 104 * profiling and statistics off the main clock. This WILL NOT be accurate; 105 * do not do it unless absolutely necessary. 106 * 107 * The statistics clock may (or may not) be run at a higher rate while 108 * profiling. This profile clock runs at profhz. We require that profhz 109 * be an integral multiple of stathz. 110 * 111 * If the statistics clock is running fast, it must be divided by the ratio 112 * profhz/stathz for statistics. (For profiling, every tick counts.) 113 * 114 * Time-of-day is maintained using a "timecounter", which may or may 115 * not be related to the hardware generating the above mentioned 116 * interrupts. 117 */ 118 119 int stathz; 120 int profhz; 121 static int profprocs; 122 int ticks; 123 static int psdiv, pscnt; /* prof => stat divider */ 124 int psratio; /* ratio: prof / stat */ 125 126 /* 127 * Initialize clock frequencies and start both clocks running. 128 */ 129 /* ARGSUSED*/ 130 static void 131 initclocks(dummy) 132 void *dummy; 133 { 134 register int i; 135 136 /* 137 * Set divisors to 1 (normal case) and let the machine-specific 138 * code do its bit. 139 */ 140 psdiv = pscnt = 1; 141 cpu_initclocks(); 142 143 /* 144 * Compute profhz/stathz, and fix profhz if needed. 145 */ 146 i = stathz ? stathz : hz; 147 if (profhz == 0) 148 profhz = i; 149 psratio = profhz / i; 150 } 151 152 /* 153 * The real-time timer, interrupting hz times per second. 154 */ 155 void 156 hardclock(frame) 157 register struct clockframe *frame; 158 { 159 register struct proc *p; 160 int need_softclock = 0; 161 162 p = curproc; 163 if (p != idleproc) { 164 register struct pstats *pstats; 165 166 /* 167 * Run current process's virtual and profile time, as needed. 168 */ 169 pstats = p->p_stats; 170 if (CLKF_USERMODE(frame) && 171 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 172 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { 173 p->p_flag |= P_ALRMPEND; 174 aston(); 175 } 176 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 177 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 178 p->p_flag |= P_PROFPEND; 179 aston(); 180 } 181 } 182 183 #if defined(SMP) && defined(BETTER_CLOCK) 184 forward_hardclock(pscnt); 185 #endif 186 187 /* 188 * If no separate statistics clock is available, run it from here. 189 */ 190 if (stathz == 0) 191 statclock(frame); 192 193 tc_windup(); 194 195 /* 196 * Process callouts at a very low cpu priority, so we don't keep the 197 * relatively high clock interrupt priority any longer than necessary. 198 */ 199 mtx_enter(&callout_lock, MTX_SPIN); 200 ticks++; 201 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) { 202 need_softclock = 1; 203 } else if (softticks + 1 == ticks) 204 ++softticks; 205 mtx_exit(&callout_lock, MTX_SPIN); 206 207 /* 208 * sched_swi acquires sched_lock, so we don't want to call it with 209 * callout_lock held; incorrect locking order. 210 */ 211 if (need_softclock) 212 sched_swi(softclock_ih, SWI_NOSWITCH); 213 } 214 215 /* 216 * Compute number of ticks in the specified amount of time. 217 */ 218 int 219 tvtohz(tv) 220 struct timeval *tv; 221 { 222 register unsigned long ticks; 223 register long sec, usec; 224 225 /* 226 * If the number of usecs in the whole seconds part of the time 227 * difference fits in a long, then the total number of usecs will 228 * fit in an unsigned long. Compute the total and convert it to 229 * ticks, rounding up and adding 1 to allow for the current tick 230 * to expire. Rounding also depends on unsigned long arithmetic 231 * to avoid overflow. 232 * 233 * Otherwise, if the number of ticks in the whole seconds part of 234 * the time difference fits in a long, then convert the parts to 235 * ticks separately and add, using similar rounding methods and 236 * overflow avoidance. This method would work in the previous 237 * case but it is slightly slower and assumes that hz is integral. 238 * 239 * Otherwise, round the time difference down to the maximum 240 * representable value. 241 * 242 * If ints have 32 bits, then the maximum value for any timeout in 243 * 10ms ticks is 248 days. 244 */ 245 sec = tv->tv_sec; 246 usec = tv->tv_usec; 247 if (usec < 0) { 248 sec--; 249 usec += 1000000; 250 } 251 if (sec < 0) { 252 #ifdef DIAGNOSTIC 253 if (usec > 0) { 254 sec++; 255 usec -= 1000000; 256 } 257 printf("tvotohz: negative time difference %ld sec %ld usec\n", 258 sec, usec); 259 #endif 260 ticks = 1; 261 } else if (sec <= LONG_MAX / 1000000) 262 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 263 / tick + 1; 264 else if (sec <= LONG_MAX / hz) 265 ticks = sec * hz 266 + ((unsigned long)usec + (tick - 1)) / tick + 1; 267 else 268 ticks = LONG_MAX; 269 if (ticks > INT_MAX) 270 ticks = INT_MAX; 271 return ((int)ticks); 272 } 273 274 /* 275 * Start profiling on a process. 276 * 277 * Kernel profiling passes proc0 which never exits and hence 278 * keeps the profile clock running constantly. 279 */ 280 void 281 startprofclock(p) 282 register struct proc *p; 283 { 284 int s; 285 286 if ((p->p_flag & P_PROFIL) == 0) { 287 p->p_flag |= P_PROFIL; 288 if (++profprocs == 1 && stathz != 0) { 289 s = splstatclock(); 290 psdiv = pscnt = psratio; 291 setstatclockrate(profhz); 292 splx(s); 293 } 294 } 295 } 296 297 /* 298 * Stop profiling on a process. 299 */ 300 void 301 stopprofclock(p) 302 register struct proc *p; 303 { 304 int s; 305 306 if (p->p_flag & P_PROFIL) { 307 p->p_flag &= ~P_PROFIL; 308 if (--profprocs == 0 && stathz != 0) { 309 s = splstatclock(); 310 psdiv = pscnt = 1; 311 setstatclockrate(stathz); 312 splx(s); 313 } 314 } 315 } 316 317 /* 318 * Statistics clock. Grab profile sample, and if divider reaches 0, 319 * do process and kernel statistics. Most of the statistics are only 320 * used by user-level statistics programs. The main exceptions are 321 * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. 322 */ 323 void 324 statclock(frame) 325 register struct clockframe *frame; 326 { 327 #ifdef GPROF 328 register struct gmonparam *g; 329 int i; 330 #endif 331 register struct proc *p; 332 struct pstats *pstats; 333 long rss; 334 struct rusage *ru; 335 struct vmspace *vm; 336 337 mtx_enter(&sched_lock, MTX_SPIN); 338 339 if (CLKF_USERMODE(frame)) { 340 /* 341 * Came from user mode; CPU was in user state. 342 * If this process is being profiled, record the tick. 343 */ 344 p = curproc; 345 if (p->p_flag & P_PROFIL) 346 addupc_intr(p, CLKF_PC(frame), 1); 347 #if defined(SMP) && defined(BETTER_CLOCK) 348 if (stathz != 0) 349 forward_statclock(pscnt); 350 #endif 351 if (--pscnt > 0) { 352 mtx_exit(&sched_lock, MTX_SPIN); 353 return; 354 } 355 /* 356 * Charge the time as appropriate. 357 */ 358 p->p_uticks++; 359 if (p->p_nice > NZERO) 360 cp_time[CP_NICE]++; 361 else 362 cp_time[CP_USER]++; 363 } else { 364 #ifdef GPROF 365 /* 366 * Kernel statistics are just like addupc_intr, only easier. 367 */ 368 g = &_gmonparam; 369 if (g->state == GMON_PROF_ON) { 370 i = CLKF_PC(frame) - g->lowpc; 371 if (i < g->textsize) { 372 i /= HISTFRACTION * sizeof(*g->kcount); 373 g->kcount[i]++; 374 } 375 } 376 #endif 377 #if defined(SMP) && defined(BETTER_CLOCK) 378 if (stathz != 0) 379 forward_statclock(pscnt); 380 #endif 381 if (--pscnt > 0) { 382 mtx_exit(&sched_lock, MTX_SPIN); 383 return; 384 } 385 /* 386 * Came from kernel mode, so we were: 387 * - handling an interrupt, 388 * - doing syscall or trap work on behalf of the current 389 * user process, or 390 * - spinning in the idle loop. 391 * Whichever it is, charge the time as appropriate. 392 * Note that we charge interrupts to the current process, 393 * regardless of whether they are ``for'' that process, 394 * so that we know how much of its real time was spent 395 * in ``non-process'' (i.e., interrupt) work. 396 */ 397 p = curproc; 398 if ((p->p_ithd != NULL) || CLKF_INTR(frame)) { 399 p->p_iticks++; 400 cp_time[CP_INTR]++; 401 } else { 402 p->p_sticks++; 403 if (p != idleproc) 404 cp_time[CP_SYS]++; 405 else 406 cp_time[CP_IDLE]++; 407 } 408 } 409 pscnt = psdiv; 410 411 schedclock(p); 412 413 /* Update resource usage integrals and maximums. */ 414 if ((pstats = p->p_stats) != NULL && 415 (ru = &pstats->p_ru) != NULL && 416 (vm = p->p_vmspace) != NULL) { 417 ru->ru_ixrss += pgtok(vm->vm_tsize); 418 ru->ru_idrss += pgtok(vm->vm_dsize); 419 ru->ru_isrss += pgtok(vm->vm_ssize); 420 rss = pgtok(vmspace_resident_count(vm)); 421 if (ru->ru_maxrss < rss) 422 ru->ru_maxrss = rss; 423 } 424 425 mtx_exit(&sched_lock, MTX_SPIN); 426 } 427 428 /* 429 * Return information about system clocks. 430 */ 431 static int 432 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 433 { 434 struct clockinfo clkinfo; 435 /* 436 * Construct clockinfo structure. 437 */ 438 clkinfo.hz = hz; 439 clkinfo.tick = tick; 440 clkinfo.tickadj = tickadj; 441 clkinfo.profhz = profhz; 442 clkinfo.stathz = stathz ? stathz : hz; 443 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 444 } 445 446 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 447 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); 448