1 /* $OpenBSD: kern_clock.c,v 1.119 2023/09/14 22:27:09 cheloha 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. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 38 */ 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/clockintr.h> 43 #include <sys/timeout.h> 44 #include <sys/kernel.h> 45 #include <sys/limits.h> 46 #include <sys/proc.h> 47 #include <sys/user.h> 48 #include <sys/resourcevar.h> 49 #include <sys/sysctl.h> 50 #include <sys/sched.h> 51 #include <sys/timetc.h> 52 53 #include "dt.h" 54 #if NDT > 0 55 #include <dev/dt/dtvar.h> 56 #endif 57 58 /* 59 * Clock handling routines. 60 * 61 * This code is written to operate with two timers that run independently of 62 * each other. The main clock, running hz times per second, is used to keep 63 * track of real time. The second timer handles kernel and user profiling, 64 * and does resource use estimation. If the second timer is programmable, 65 * it is randomized to avoid aliasing between the two clocks. For example, 66 * the randomization prevents an adversary from always giving up the cpu 67 * just before its quantum expires. Otherwise, it would never accumulate 68 * cpu ticks. The mean frequency of the second timer is stathz. 69 * 70 * If no second timer exists, stathz will be zero; in this case we drive 71 * profiling and statistics off the main clock. This WILL NOT be accurate; 72 * do not do it unless absolutely necessary. 73 * 74 * The statistics clock may (or may not) be run at a higher rate while 75 * profiling. This profile clock runs at profhz. We require that profhz 76 * be an integral multiple of stathz. 77 * 78 * If the statistics clock is running fast, it must be divided by the ratio 79 * profhz/stathz for statistics. (For profiling, every tick counts.) 80 */ 81 82 int stathz; 83 int profhz; 84 int profprocs; 85 int ticks = INT_MAX - (15 * 60 * HZ); 86 87 /* Don't force early wrap around, triggers bug in inteldrm */ 88 volatile unsigned long jiffies; 89 90 uint32_t hardclock_period; /* [I] hardclock period (ns) */ 91 uint32_t statclock_avg; /* [I] average statclock period (ns) */ 92 uint32_t statclock_min; /* [I] minimum statclock period (ns) */ 93 uint32_t statclock_mask; /* [I] set of allowed offsets */ 94 int statclock_is_randomized; /* [I] fixed or pseudorandom period? */ 95 96 /* 97 * Initialize clock frequencies and start both clocks running. 98 */ 99 void 100 initclocks(void) 101 { 102 uint32_t half_avg, var; 103 104 /* 105 * Let the machine-specific code do its bit. 106 */ 107 cpu_initclocks(); 108 109 KASSERT(hz > 0 && hz <= 1000000000); 110 hardclock_period = 1000000000 / hz; 111 roundrobin_period = hardclock_period * 10; 112 113 KASSERT(stathz >= 1 && stathz <= 1000000000); 114 115 /* 116 * Compute the average statclock() period. Then find var, the 117 * largest power of two such that var <= statclock_avg / 2. 118 */ 119 statclock_avg = 1000000000 / stathz; 120 half_avg = statclock_avg / 2; 121 for (var = 1U << 31; var > half_avg; var /= 2) 122 continue; 123 124 /* 125 * Set a lower bound for the range using statclock_avg and var. 126 * The mask for that range is just (var - 1). 127 */ 128 statclock_min = statclock_avg - (var / 2); 129 statclock_mask = var - 1; 130 131 KASSERT(profhz >= stathz && profhz <= 1000000000); 132 KASSERT(profhz % stathz == 0); 133 profclock_period = 1000000000 / profhz; 134 135 inittimecounter(); 136 137 /* Start dispatching clock interrupts on the primary CPU. */ 138 cpu_startclock(); 139 } 140 141 /* 142 * The real-time timer, interrupting hz times per second. 143 */ 144 void 145 hardclock(struct clockframe *frame) 146 { 147 #if NDT > 0 148 DT_ENTER(profile, NULL); 149 if (CPU_IS_PRIMARY(curcpu())) 150 DT_ENTER(interval, NULL); 151 #endif 152 153 /* 154 * If we are not the primary CPU, we're not allowed to do 155 * any more work. 156 */ 157 if (CPU_IS_PRIMARY(curcpu()) == 0) 158 return; 159 160 tc_ticktock(); 161 ticks++; 162 jiffies++; 163 164 /* 165 * Update the timeout wheel. 166 */ 167 timeout_hardclock_update(); 168 } 169 170 /* 171 * Compute number of hz in the specified amount of time. 172 */ 173 int 174 tvtohz(const struct timeval *tv) 175 { 176 unsigned long nticks; 177 time_t sec; 178 long usec; 179 180 /* 181 * If the number of usecs in the whole seconds part of the time 182 * fits in a long, then the total number of usecs will 183 * fit in an unsigned long. Compute the total and convert it to 184 * ticks, rounding up and adding 1 to allow for the current tick 185 * to expire. Rounding also depends on unsigned long arithmetic 186 * to avoid overflow. 187 * 188 * Otherwise, if the number of ticks in the whole seconds part of 189 * the time fits in a long, then convert the parts to 190 * ticks separately and add, using similar rounding methods and 191 * overflow avoidance. This method would work in the previous 192 * case but it is slightly slower and assumes that hz is integral. 193 * 194 * Otherwise, round the time down to the maximum 195 * representable value. 196 * 197 * If ints have 32 bits, then the maximum value for any timeout in 198 * 10ms ticks is 248 days. 199 */ 200 sec = tv->tv_sec; 201 usec = tv->tv_usec; 202 if (sec < 0 || (sec == 0 && usec <= 0)) 203 nticks = 0; 204 else if (sec <= LONG_MAX / 1000000) 205 nticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 206 / tick + 1; 207 else if (sec <= LONG_MAX / hz) 208 nticks = sec * hz 209 + ((unsigned long)usec + (tick - 1)) / tick + 1; 210 else 211 nticks = LONG_MAX; 212 if (nticks > INT_MAX) 213 nticks = INT_MAX; 214 return ((int)nticks); 215 } 216 217 int 218 tstohz(const struct timespec *ts) 219 { 220 struct timeval tv; 221 TIMESPEC_TO_TIMEVAL(&tv, ts); 222 223 /* Round up. */ 224 if ((ts->tv_nsec % 1000) != 0) { 225 tv.tv_usec += 1; 226 if (tv.tv_usec >= 1000000) { 227 tv.tv_usec -= 1000000; 228 tv.tv_sec += 1; 229 } 230 } 231 232 return (tvtohz(&tv)); 233 } 234 235 /* 236 * Start profiling on a process. 237 * 238 * Kernel profiling passes proc0 which never exits and hence 239 * keeps the profile clock running constantly. 240 */ 241 void 242 startprofclock(struct process *pr) 243 { 244 int s; 245 246 if ((pr->ps_flags & PS_PROFIL) == 0) { 247 atomic_setbits_int(&pr->ps_flags, PS_PROFIL); 248 if (++profprocs == 1) { 249 s = splstatclock(); 250 setstatclockrate(profhz); 251 splx(s); 252 } 253 } 254 } 255 256 /* 257 * Stop profiling on a process. 258 */ 259 void 260 stopprofclock(struct process *pr) 261 { 262 int s; 263 264 if (pr->ps_flags & PS_PROFIL) { 265 atomic_clearbits_int(&pr->ps_flags, PS_PROFIL); 266 if (--profprocs == 0) { 267 s = splstatclock(); 268 setstatclockrate(stathz); 269 splx(s); 270 } 271 } 272 } 273 274 /* 275 * Statistics clock. Grab profile sample, and if divider reaches 0, 276 * do process and kernel statistics. 277 */ 278 void 279 statclock(struct clockintr *cl, void *cf, void *arg) 280 { 281 uint64_t count, i; 282 struct clockframe *frame = cf; 283 struct cpu_info *ci = curcpu(); 284 struct schedstate_percpu *spc = &ci->ci_schedstate; 285 struct proc *p = curproc; 286 struct process *pr; 287 288 if (statclock_is_randomized) { 289 count = clockintr_advance_random(cl, statclock_min, 290 statclock_mask); 291 } else { 292 count = clockintr_advance(cl, statclock_avg); 293 } 294 295 if (CLKF_USERMODE(frame)) { 296 pr = p->p_p; 297 /* 298 * Came from user mode; CPU was in user state. 299 * If this process is being profiled record the tick. 300 */ 301 p->p_uticks += count; 302 if (pr->ps_nice > NZERO) 303 spc->spc_cp_time[CP_NICE] += count; 304 else 305 spc->spc_cp_time[CP_USER] += count; 306 } else { 307 /* 308 * Came from kernel mode, so we were: 309 * - spinning on a lock 310 * - handling an interrupt, 311 * - doing syscall or trap work on behalf of the current 312 * user process, or 313 * - spinning in the idle loop. 314 * Whichever it is, charge the time as appropriate. 315 * Note that we charge interrupts to the current process, 316 * regardless of whether they are ``for'' that process, 317 * so that we know how much of its real time was spent 318 * in ``non-process'' (i.e., interrupt) work. 319 */ 320 if (CLKF_INTR(frame)) { 321 if (p != NULL) 322 p->p_iticks += count; 323 spc->spc_cp_time[spc->spc_spinning ? 324 CP_SPIN : CP_INTR] += count; 325 } else if (p != NULL && p != spc->spc_idleproc) { 326 p->p_sticks += count; 327 spc->spc_cp_time[spc->spc_spinning ? 328 CP_SPIN : CP_SYS] += count; 329 } else 330 spc->spc_cp_time[spc->spc_spinning ? 331 CP_SPIN : CP_IDLE] += count; 332 } 333 334 if (p != NULL) { 335 p->p_cpticks += count; 336 /* 337 * schedclock() runs every fourth statclock(). 338 */ 339 for (i = 0; i < count; i++) { 340 if ((++spc->spc_schedticks & 3) == 0) 341 schedclock(p); 342 } 343 } 344 } 345 346 /* 347 * Return information about system clocks. 348 */ 349 int 350 sysctl_clockrate(char *where, size_t *sizep, void *newp) 351 { 352 struct clockinfo clkinfo; 353 354 /* 355 * Construct clockinfo structure. 356 */ 357 memset(&clkinfo, 0, sizeof clkinfo); 358 clkinfo.tick = tick; 359 clkinfo.hz = hz; 360 clkinfo.profhz = profhz; 361 clkinfo.stathz = stathz; 362 return (sysctl_rdstruct(where, sizep, newp, &clkinfo, sizeof(clkinfo))); 363 } 364