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 != PCPU_GET(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 mtx_lock_spin(&sched_lock); 174 p->p_sflag |= PS_ALRMPEND; 175 aston(); 176 mtx_unlock_spin(&sched_lock); 177 } 178 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 179 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { 180 mtx_lock_spin(&sched_lock); 181 p->p_sflag |= PS_PROFPEND; 182 aston(); 183 mtx_unlock_spin(&sched_lock); 184 } 185 } 186 187 #if defined(SMP) && defined(BETTER_CLOCK) 188 forward_hardclock(pscnt); 189 #endif 190 191 /* 192 * If no separate statistics clock is available, run it from here. 193 */ 194 if (stathz == 0) 195 statclock(frame); 196 197 tc_windup(); 198 199 /* 200 * Process callouts at a very low cpu priority, so we don't keep the 201 * relatively high clock interrupt priority any longer than necessary. 202 */ 203 mtx_lock_spin(&callout_lock); 204 ticks++; 205 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) { 206 need_softclock = 1; 207 } else if (softticks + 1 == ticks) 208 ++softticks; 209 mtx_unlock_spin(&callout_lock); 210 211 /* 212 * swi_sched acquires sched_lock, so we don't want to call it with 213 * callout_lock held; incorrect locking order. 214 */ 215 if (need_softclock) 216 swi_sched(softclock_ih, SWI_NOSWITCH); 217 } 218 219 /* 220 * Compute number of ticks in the specified amount of time. 221 */ 222 int 223 tvtohz(tv) 224 struct timeval *tv; 225 { 226 register unsigned long ticks; 227 register long sec, usec; 228 229 /* 230 * If the number of usecs in the whole seconds part of the time 231 * difference fits in a long, then the total number of usecs will 232 * fit in an unsigned long. Compute the total and convert it to 233 * ticks, rounding up and adding 1 to allow for the current tick 234 * to expire. Rounding also depends on unsigned long arithmetic 235 * to avoid overflow. 236 * 237 * Otherwise, if the number of ticks in the whole seconds part of 238 * the time difference fits in a long, then convert the parts to 239 * ticks separately and add, using similar rounding methods and 240 * overflow avoidance. This method would work in the previous 241 * case but it is slightly slower and assumes that hz is integral. 242 * 243 * Otherwise, round the time difference down to the maximum 244 * representable value. 245 * 246 * If ints have 32 bits, then the maximum value for any timeout in 247 * 10ms ticks is 248 days. 248 */ 249 sec = tv->tv_sec; 250 usec = tv->tv_usec; 251 if (usec < 0) { 252 sec--; 253 usec += 1000000; 254 } 255 if (sec < 0) { 256 #ifdef DIAGNOSTIC 257 if (usec > 0) { 258 sec++; 259 usec -= 1000000; 260 } 261 printf("tvotohz: negative time difference %ld sec %ld usec\n", 262 sec, usec); 263 #endif 264 ticks = 1; 265 } else if (sec <= LONG_MAX / 1000000) 266 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 267 / tick + 1; 268 else if (sec <= LONG_MAX / hz) 269 ticks = sec * hz 270 + ((unsigned long)usec + (tick - 1)) / tick + 1; 271 else 272 ticks = LONG_MAX; 273 if (ticks > INT_MAX) 274 ticks = INT_MAX; 275 return ((int)ticks); 276 } 277 278 /* 279 * Start profiling on a process. 280 * 281 * Kernel profiling passes proc0 which never exits and hence 282 * keeps the profile clock running constantly. 283 */ 284 void 285 startprofclock(p) 286 register struct proc *p; 287 { 288 int s; 289 290 /* 291 * XXX; Right now sched_lock protects statclock(), but perhaps 292 * it should be protected later on by a time_lock, which would 293 * cover psdiv, etc. as well. 294 */ 295 mtx_lock_spin(&sched_lock); 296 if ((p->p_sflag & PS_PROFIL) == 0) { 297 p->p_sflag |= PS_PROFIL; 298 if (++profprocs == 1 && stathz != 0) { 299 s = splstatclock(); 300 psdiv = pscnt = psratio; 301 setstatclockrate(profhz); 302 splx(s); 303 } 304 } 305 mtx_unlock_spin(&sched_lock); 306 } 307 308 /* 309 * Stop profiling on a process. 310 */ 311 void 312 stopprofclock(p) 313 register struct proc *p; 314 { 315 int s; 316 317 mtx_lock_spin(&sched_lock); 318 if (p->p_sflag & PS_PROFIL) { 319 p->p_sflag &= ~PS_PROFIL; 320 if (--profprocs == 0 && stathz != 0) { 321 s = splstatclock(); 322 psdiv = pscnt = 1; 323 setstatclockrate(stathz); 324 splx(s); 325 } 326 } 327 mtx_unlock_spin(&sched_lock); 328 } 329 330 /* 331 * Statistics clock. Grab profile sample, and if divider reaches 0, 332 * do process and kernel statistics. Most of the statistics are only 333 * used by user-level statistics programs. The main exceptions are 334 * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. 335 */ 336 void 337 statclock(frame) 338 register struct clockframe *frame; 339 { 340 #ifdef GPROF 341 register struct gmonparam *g; 342 int i; 343 #endif 344 register struct proc *p; 345 struct pstats *pstats; 346 long rss; 347 struct rusage *ru; 348 struct vmspace *vm; 349 350 mtx_lock_spin(&sched_lock); 351 352 if (CLKF_USERMODE(frame)) { 353 /* 354 * Came from user mode; CPU was in user state. 355 * If this process is being profiled, record the tick. 356 */ 357 p = curproc; 358 if (p->p_sflag & PS_PROFIL) 359 addupc_intr(p, CLKF_PC(frame), 1); 360 #if defined(SMP) && defined(BETTER_CLOCK) 361 if (stathz != 0) 362 forward_statclock(pscnt); 363 #endif 364 if (--pscnt > 0) { 365 mtx_unlock_spin(&sched_lock); 366 return; 367 } 368 /* 369 * Charge the time as appropriate. 370 */ 371 p->p_uticks++; 372 if (p->p_nice > NZERO) 373 cp_time[CP_NICE]++; 374 else 375 cp_time[CP_USER]++; 376 } else { 377 #ifdef GPROF 378 /* 379 * Kernel statistics are just like addupc_intr, only easier. 380 */ 381 g = &_gmonparam; 382 if (g->state == GMON_PROF_ON) { 383 i = CLKF_PC(frame) - g->lowpc; 384 if (i < g->textsize) { 385 i /= HISTFRACTION * sizeof(*g->kcount); 386 g->kcount[i]++; 387 } 388 } 389 #endif 390 #if defined(SMP) && defined(BETTER_CLOCK) 391 if (stathz != 0) 392 forward_statclock(pscnt); 393 #endif 394 if (--pscnt > 0) { 395 mtx_unlock_spin(&sched_lock); 396 return; 397 } 398 /* 399 * Came from kernel mode, so we were: 400 * - handling an interrupt, 401 * - doing syscall or trap work on behalf of the current 402 * user process, or 403 * - spinning in the idle loop. 404 * Whichever it is, charge the time as appropriate. 405 * Note that we charge interrupts to the current process, 406 * regardless of whether they are ``for'' that process, 407 * so that we know how much of its real time was spent 408 * in ``non-process'' (i.e., interrupt) work. 409 */ 410 p = curproc; 411 if ((p->p_ithd != NULL) || CLKF_INTR(frame)) { 412 p->p_iticks++; 413 cp_time[CP_INTR]++; 414 } else { 415 p->p_sticks++; 416 if (p != PCPU_GET(idleproc)) 417 cp_time[CP_SYS]++; 418 else 419 cp_time[CP_IDLE]++; 420 } 421 } 422 pscnt = psdiv; 423 424 schedclock(p); 425 426 /* Update resource usage integrals and maximums. */ 427 if ((pstats = p->p_stats) != NULL && 428 (ru = &pstats->p_ru) != NULL && 429 (vm = p->p_vmspace) != NULL) { 430 ru->ru_ixrss += pgtok(vm->vm_tsize); 431 ru->ru_idrss += pgtok(vm->vm_dsize); 432 ru->ru_isrss += pgtok(vm->vm_ssize); 433 rss = pgtok(vmspace_resident_count(vm)); 434 if (ru->ru_maxrss < rss) 435 ru->ru_maxrss = rss; 436 } 437 438 mtx_unlock_spin(&sched_lock); 439 } 440 441 /* 442 * Return information about system clocks. 443 */ 444 static int 445 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 446 { 447 struct clockinfo clkinfo; 448 /* 449 * Construct clockinfo structure. 450 */ 451 clkinfo.hz = hz; 452 clkinfo.tick = tick; 453 clkinfo.tickadj = tickadj; 454 clkinfo.profhz = profhz; 455 clkinfo.stathz = stathz ? stathz : hz; 456 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 457 } 458 459 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 460 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); 461