| 1 | /* |
| 2 | * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved. |
| 3 | * |
| 4 | * This code is derived from software contributed to The DragonFly Project |
| 5 | * by Matthew Dillon <dillon@backplane.com> |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in |
| 15 | * the documentation and/or other materials provided with the |
| 16 | * distribution. |
| 17 | * 3. Neither the name of The DragonFly Project nor the names of its |
| 18 | * contributors may be used to endorse or promote products derived |
| 19 | * from this software without specific, prior written permission. |
| 20 | * |
| 21 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 22 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 23 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 24 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 25 | * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 26 | * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 27 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 28 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 29 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 30 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 31 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 32 | * SUCH DAMAGE. |
| 33 | * |
| 34 | * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org> |
| 35 | * Copyright (c) 1982, 1986, 1991, 1993 |
| 36 | * The Regents of the University of California. All rights reserved. |
| 37 | * (c) UNIX System Laboratories, Inc. |
| 38 | * All or some portions of this file are derived from material licensed |
| 39 | * to the University of California by American Telephone and Telegraph |
| 40 | * Co. or Unix System Laboratories, Inc. and are reproduced herein with |
| 41 | * the permission of UNIX System Laboratories, Inc. |
| 42 | * |
| 43 | * Redistribution and use in source and binary forms, with or without |
| 44 | * modification, are permitted provided that the following conditions |
| 45 | * are met: |
| 46 | * 1. Redistributions of source code must retain the above copyright |
| 47 | * notice, this list of conditions and the following disclaimer. |
| 48 | * 2. Redistributions in binary form must reproduce the above copyright |
| 49 | * notice, this list of conditions and the following disclaimer in the |
| 50 | * documentation and/or other materials provided with the distribution. |
| 51 | * 3. All advertising materials mentioning features or use of this software |
| 52 | * must display the following acknowledgement: |
| 53 | * This product includes software developed by the University of |
| 54 | * California, Berkeley and its contributors. |
| 55 | * 4. Neither the name of the University nor the names of its contributors |
| 56 | * may be used to endorse or promote products derived from this software |
| 57 | * without specific prior written permission. |
| 58 | * |
| 59 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 60 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 61 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 62 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 63 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 64 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 65 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 66 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 67 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 68 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 69 | * SUCH DAMAGE. |
| 70 | * |
| 71 | * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 |
| 72 | * $FreeBSD: src/sys/kern/kern_clock.c,v 1.105.2.10 2002/10/17 13:19:40 maxim Exp $ |
| 73 | * $DragonFly: src/sys/kern/kern_clock.c,v 1.30 2005/01/31 21:37:52 joerg Exp $ |
| 74 | */ |
| 75 | |
| 76 | #include "opt_ntp.h" |
| 77 | |
| 78 | #include <sys/param.h> |
| 79 | #include <sys/systm.h> |
| 80 | #include <sys/dkstat.h> |
| 81 | #include <sys/callout.h> |
| 82 | #include <sys/kernel.h> |
| 83 | #include <sys/kinfo.h> |
| 84 | #include <sys/proc.h> |
| 85 | #include <sys/malloc.h> |
| 86 | #include <sys/resourcevar.h> |
| 87 | #include <sys/signalvar.h> |
| 88 | #include <sys/timex.h> |
| 89 | #include <sys/timepps.h> |
| 90 | #include <vm/vm.h> |
| 91 | #include <sys/lock.h> |
| 92 | #include <vm/pmap.h> |
| 93 | #include <vm/vm_map.h> |
| 94 | #include <sys/sysctl.h> |
| 95 | #include <sys/thread2.h> |
| 96 | |
| 97 | #include <machine/cpu.h> |
| 98 | #include <machine/limits.h> |
| 99 | #include <machine/smp.h> |
| 100 | |
| 101 | #ifdef GPROF |
| 102 | #include <sys/gmon.h> |
| 103 | #endif |
| 104 | |
| 105 | #ifdef DEVICE_POLLING |
| 106 | extern void init_device_poll(void); |
| 107 | extern void hardclock_device_poll(void); |
| 108 | #endif /* DEVICE_POLLING */ |
| 109 | |
| 110 | static void initclocks (void *dummy); |
| 111 | SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) |
| 112 | |
| 113 | /* |
| 114 | * Some of these don't belong here, but it's easiest to concentrate them. |
| 115 | * Note that cp_time counts in microseconds, but most userland programs |
| 116 | * just compare relative times against the total by delta. |
| 117 | */ |
| 118 | struct cp_time cp_time; |
| 119 | |
| 120 | SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time), |
| 121 | "LU", "CPU time statistics"); |
| 122 | |
| 123 | /* |
| 124 | * boottime is used to calculate the 'real' uptime. Do not confuse this with |
| 125 | * microuptime(). microtime() is not drift compensated. The real uptime |
| 126 | * with compensation is nanotime() - bootime. boottime is recalculated |
| 127 | * whenever the real time is set based on the compensated elapsed time |
| 128 | * in seconds (gd->gd_time_seconds). |
| 129 | * |
| 130 | * basetime is used to calculate the compensated real time of day. Chunky |
| 131 | * changes to the time, aka settimeofday(), are made by modifying basetime. |
| 132 | * |
| 133 | * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic. |
| 134 | * Slight adjustments to gd_cpuclock_base are made to phase-lock it to |
| 135 | * the real time. |
| 136 | */ |
| 137 | struct timespec boottime; /* boot time (realtime) for reference only */ |
| 138 | struct timespec basetime; /* base time adjusts uptime -> realtime */ |
| 139 | time_t time_second; /* read-only 'passive' uptime in seconds */ |
| 140 | |
| 141 | SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD, |
| 142 | &boottime, timeval, "System boottime"); |
| 143 | SYSCTL_STRUCT(_kern, OID_AUTO, basetime, CTLFLAG_RD, |
| 144 | &basetime, timeval, "System basetime"); |
| 145 | |
| 146 | static void hardclock(systimer_t info, struct intrframe *frame); |
| 147 | static void statclock(systimer_t info, struct intrframe *frame); |
| 148 | static void schedclock(systimer_t info, struct intrframe *frame); |
| 149 | |
| 150 | int ticks; /* system master ticks at hz */ |
| 151 | int clocks_running; /* tsleep/timeout clocks operational */ |
| 152 | int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */ |
| 153 | int64_t nsec_acc; /* accumulator */ |
| 154 | |
| 155 | /* |
| 156 | * Finish initializing clock frequencies and start all clocks running. |
| 157 | */ |
| 158 | /* ARGSUSED*/ |
| 159 | static void |
| 160 | initclocks(void *dummy) |
| 161 | { |
| 162 | cpu_initclocks(); |
| 163 | #ifdef DEVICE_POLLING |
| 164 | init_device_poll(); |
| 165 | #endif |
| 166 | /*psratio = profhz / stathz;*/ |
| 167 | initclocks_pcpu(); |
| 168 | clocks_running = 1; |
| 169 | } |
| 170 | |
| 171 | /* |
| 172 | * Called on a per-cpu basis |
| 173 | */ |
| 174 | void |
| 175 | initclocks_pcpu(void) |
| 176 | { |
| 177 | struct globaldata *gd = mycpu; |
| 178 | |
| 179 | crit_enter(); |
| 180 | if (gd->gd_cpuid == 0) { |
| 181 | gd->gd_time_seconds = 1; |
| 182 | gd->gd_cpuclock_base = cputimer_count(); |
| 183 | } else { |
| 184 | /* XXX */ |
| 185 | gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds; |
| 186 | gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base; |
| 187 | } |
| 188 | |
| 189 | /* |
| 190 | * Use a non-queued periodic systimer to prevent multiple ticks from |
| 191 | * building up if the sysclock jumps forward (8254 gets reset). The |
| 192 | * sysclock will never jump backwards. Our time sync is based on |
| 193 | * the actual sysclock, not the ticks count. |
| 194 | */ |
| 195 | systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz); |
| 196 | systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz); |
| 197 | /* XXX correct the frequency for scheduler / estcpu tests */ |
| 198 | systimer_init_periodic_nq(&gd->gd_schedclock, schedclock, |
| 199 | NULL, ESTCPUFREQ); |
| 200 | crit_exit(); |
| 201 | } |
| 202 | |
| 203 | /* |
| 204 | * This sets the current real time of day. Timespecs are in seconds and |
| 205 | * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base, |
| 206 | * instead we adjust basetime so basetime + gd_* results in the current |
| 207 | * time of day. This way the gd_* fields are guarenteed to represent |
| 208 | * a monotonically increasing 'uptime' value. |
| 209 | */ |
| 210 | void |
| 211 | set_timeofday(struct timespec *ts) |
| 212 | { |
| 213 | struct timespec ts2; |
| 214 | |
| 215 | /* |
| 216 | * XXX SMP / non-atomic basetime updates |
| 217 | */ |
| 218 | crit_enter(); |
| 219 | nanouptime(&ts2); |
| 220 | basetime.tv_sec = ts->tv_sec - ts2.tv_sec; |
| 221 | basetime.tv_nsec = ts->tv_nsec - ts2.tv_nsec; |
| 222 | if (basetime.tv_nsec < 0) { |
| 223 | basetime.tv_nsec += 1000000000; |
| 224 | --basetime.tv_sec; |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * Note that basetime diverges from boottime as the clock drift is |
| 229 | * compensated for, so we cannot do away with boottime. When setting |
| 230 | * the absolute time of day the drift is 0 (for an instant) and we |
| 231 | * can simply assign boottime to basetime. |
| 232 | * |
| 233 | * Note that nanouptime() is based on gd_time_seconds which is drift |
| 234 | * compensated up to a point (it is guarenteed to remain monotonically |
| 235 | * increasing). gd_time_seconds is thus our best uptime guess and |
| 236 | * suitable for use in the boottime calculation. It is already taken |
| 237 | * into account in the basetime calculation above. |
| 238 | */ |
| 239 | boottime.tv_sec = basetime.tv_sec; |
| 240 | timedelta = 0; |
| 241 | crit_exit(); |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * Each cpu has its own hardclock, but we only increments ticks and softticks |
| 246 | * on cpu #0. |
| 247 | * |
| 248 | * NOTE! systimer! the MP lock might not be held here. We can only safely |
| 249 | * manipulate objects owned by the current cpu. |
| 250 | */ |
| 251 | static void |
| 252 | hardclock(systimer_t info, struct intrframe *frame) |
| 253 | { |
| 254 | sysclock_t cputicks; |
| 255 | struct proc *p; |
| 256 | struct pstats *pstats; |
| 257 | struct globaldata *gd = mycpu; |
| 258 | |
| 259 | /* |
| 260 | * Realtime updates are per-cpu. Note that timer corrections as |
| 261 | * returned by microtime() and friends make an additional adjustment |
| 262 | * using a system-wise 'basetime', but the running time is always |
| 263 | * taken from the per-cpu globaldata area. Since the same clock |
| 264 | * is distributing (XXX SMP) to all cpus, the per-cpu timebases |
| 265 | * stay in synch. |
| 266 | * |
| 267 | * Note that we never allow info->time (aka gd->gd_hardclock.time) |
| 268 | * to reverse index gd_cpuclock_base, but that it is possible for |
| 269 | * it to temporarily get behind in the seconds if something in the |
| 270 | * system locks interrupts for a long period of time. Since periodic |
| 271 | * timers count events, though everything should resynch again |
| 272 | * immediately. |
| 273 | */ |
| 274 | cputicks = info->time - gd->gd_cpuclock_base; |
| 275 | if (cputicks >= cputimer_freq) { |
| 276 | ++gd->gd_time_seconds; |
| 277 | gd->gd_cpuclock_base += cputimer_freq; |
| 278 | } |
| 279 | |
| 280 | /* |
| 281 | * The system-wide ticks counter and NTP related timedelta/tickdelta |
| 282 | * adjustments only occur on cpu #0. NTP adjustments are accomplished |
| 283 | * by updating basetime. |
| 284 | */ |
| 285 | if (gd->gd_cpuid == 0) { |
| 286 | struct timespec nts; |
| 287 | int leap; |
| 288 | |
| 289 | ++ticks; |
| 290 | |
| 291 | #ifdef DEVICE_POLLING |
| 292 | hardclock_device_poll(); /* mpsafe, short and quick */ |
| 293 | #endif /* DEVICE_POLLING */ |
| 294 | |
| 295 | #if 0 |
| 296 | if (tco->tc_poll_pps) |
| 297 | tco->tc_poll_pps(tco); |
| 298 | #endif |
| 299 | /* |
| 300 | * Apply adjtime corrections. At the moment only do this if |
| 301 | * we can get the MP lock to interlock with adjtime's modification |
| 302 | * of these variables. Note that basetime adjustments are not |
| 303 | * MP safe either XXX. |
| 304 | */ |
| 305 | if (timedelta != 0 && try_mplock()) { |
| 306 | basetime.tv_nsec += tickdelta * 1000; |
| 307 | if (basetime.tv_nsec >= 1000000000) { |
| 308 | basetime.tv_nsec -= 1000000000; |
| 309 | ++basetime.tv_sec; |
| 310 | } else if (basetime.tv_nsec < 0) { |
| 311 | basetime.tv_nsec += 1000000000; |
| 312 | --basetime.tv_sec; |
| 313 | } |
| 314 | timedelta -= tickdelta; |
| 315 | rel_mplock(); |
| 316 | } |
| 317 | |
| 318 | /* |
| 319 | * Apply per-tick compensation. ticks_adj adjusts for both |
| 320 | * offset and frequency, and could be negative. |
| 321 | */ |
| 322 | if (nsec_adj != 0 && try_mplock()) { |
| 323 | nsec_acc += nsec_adj; |
| 324 | if (nsec_acc >= 0x100000000LL) { |
| 325 | basetime.tv_nsec += nsec_acc >> 32; |
| 326 | nsec_acc = (nsec_acc & 0xFFFFFFFFLL); |
| 327 | } else if (nsec_acc <= -0x100000000LL) { |
| 328 | basetime.tv_nsec -= -nsec_acc >> 32; |
| 329 | nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL); |
| 330 | } |
| 331 | if (basetime.tv_nsec >= 1000000000) { |
| 332 | basetime.tv_nsec -= 1000000000; |
| 333 | ++basetime.tv_sec; |
| 334 | } else if (basetime.tv_nsec < 0) { |
| 335 | basetime.tv_nsec += 1000000000; |
| 336 | --basetime.tv_sec; |
| 337 | } |
| 338 | rel_mplock(); |
| 339 | } |
| 340 | |
| 341 | /* |
| 342 | * If the realtime-adjusted seconds hand rolls over then tell |
| 343 | * ntp_update_second() what we did in the last second so it can |
| 344 | * calculate what to do in the next second. It may also add |
| 345 | * or subtract a leap second. |
| 346 | */ |
| 347 | getnanotime(&nts); |
| 348 | if (time_second != nts.tv_sec) { |
| 349 | leap = ntp_update_second(time_second, &nsec_adj); |
| 350 | basetime.tv_sec += leap; |
| 351 | time_second = nts.tv_sec + leap; |
| 352 | nsec_adj /= hz; |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * softticks are handled for all cpus |
| 358 | */ |
| 359 | hardclock_softtick(gd); |
| 360 | |
| 361 | /* |
| 362 | * ITimer handling is per-tick, per-cpu. I don't think psignal() |
| 363 | * is mpsafe on curproc, so XXX get the mplock. |
| 364 | */ |
| 365 | if ((p = curproc) != NULL && try_mplock()) { |
| 366 | pstats = p->p_stats; |
| 367 | if (frame && CLKF_USERMODE(frame) && |
| 368 | timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && |
| 369 | itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) |
| 370 | psignal(p, SIGVTALRM); |
| 371 | if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && |
| 372 | itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) |
| 373 | psignal(p, SIGPROF); |
| 374 | rel_mplock(); |
| 375 | } |
| 376 | setdelayed(); |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * The statistics clock typically runs at a 125Hz rate, and is intended |
| 381 | * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu. |
| 382 | * |
| 383 | * NOTE! systimer! the MP lock might not be held here. We can only safely |
| 384 | * manipulate objects owned by the current cpu. |
| 385 | * |
| 386 | * The stats clock is responsible for grabbing a profiling sample. |
| 387 | * Most of the statistics are only used by user-level statistics programs. |
| 388 | * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and |
| 389 | * p->p_estcpu. |
| 390 | * |
| 391 | * Like the other clocks, the stat clock is called from what is effectively |
| 392 | * a fast interrupt, so the context should be the thread/process that got |
| 393 | * interrupted. |
| 394 | */ |
| 395 | static void |
| 396 | statclock(systimer_t info, struct intrframe *frame) |
| 397 | { |
| 398 | #ifdef GPROF |
| 399 | struct gmonparam *g; |
| 400 | int i; |
| 401 | #endif |
| 402 | thread_t td; |
| 403 | struct proc *p; |
| 404 | int bump; |
| 405 | struct timeval tv; |
| 406 | struct timeval *stv; |
| 407 | |
| 408 | /* |
| 409 | * How big was our timeslice relative to the last time? |
| 410 | */ |
| 411 | microuptime(&tv); /* mpsafe */ |
| 412 | stv = &mycpu->gd_stattv; |
| 413 | if (stv->tv_sec == 0) { |
| 414 | bump = 1; |
| 415 | } else { |
| 416 | bump = tv.tv_usec - stv->tv_usec + |
| 417 | (tv.tv_sec - stv->tv_sec) * 1000000; |
| 418 | if (bump < 0) |
| 419 | bump = 0; |
| 420 | if (bump > 1000000) |
| 421 | bump = 1000000; |
| 422 | } |
| 423 | *stv = tv; |
| 424 | |
| 425 | td = curthread; |
| 426 | p = td->td_proc; |
| 427 | |
| 428 | if (frame && CLKF_USERMODE(frame)) { |
| 429 | /* |
| 430 | * Came from userland, handle user time and deal with |
| 431 | * possible process. |
| 432 | */ |
| 433 | if (p && (p->p_flag & P_PROFIL)) |
| 434 | addupc_intr(p, CLKF_PC(frame), 1); |
| 435 | td->td_uticks += bump; |
| 436 | |
| 437 | /* |
| 438 | * Charge the time as appropriate |
| 439 | */ |
| 440 | if (p && p->p_nice > NZERO) |
| 441 | cp_time.cp_nice += bump; |
| 442 | else |
| 443 | cp_time.cp_user += bump; |
| 444 | } else { |
| 445 | #ifdef GPROF |
| 446 | /* |
| 447 | * Kernel statistics are just like addupc_intr, only easier. |
| 448 | */ |
| 449 | g = &_gmonparam; |
| 450 | if (g->state == GMON_PROF_ON && frame) { |
| 451 | i = CLKF_PC(frame) - g->lowpc; |
| 452 | if (i < g->textsize) { |
| 453 | i /= HISTFRACTION * sizeof(*g->kcount); |
| 454 | g->kcount[i]++; |
| 455 | } |
| 456 | } |
| 457 | #endif |
| 458 | /* |
| 459 | * Came from kernel mode, so we were: |
| 460 | * - handling an interrupt, |
| 461 | * - doing syscall or trap work on behalf of the current |
| 462 | * user process, or |
| 463 | * - spinning in the idle loop. |
| 464 | * Whichever it is, charge the time as appropriate. |
| 465 | * Note that we charge interrupts to the current process, |
| 466 | * regardless of whether they are ``for'' that process, |
| 467 | * so that we know how much of its real time was spent |
| 468 | * in ``non-process'' (i.e., interrupt) work. |
| 469 | * |
| 470 | * XXX assume system if frame is NULL. A NULL frame |
| 471 | * can occur if ipi processing is done from an splx(). |
| 472 | */ |
| 473 | if (frame && CLKF_INTR(frame)) |
| 474 | td->td_iticks += bump; |
| 475 | else |
| 476 | td->td_sticks += bump; |
| 477 | |
| 478 | if (frame && CLKF_INTR(frame)) { |
| 479 | cp_time.cp_intr += bump; |
| 480 | } else { |
| 481 | if (td == &mycpu->gd_idlethread) |
| 482 | cp_time.cp_idle += bump; |
| 483 | else |
| 484 | cp_time.cp_sys += bump; |
| 485 | } |
| 486 | } |
| 487 | } |
| 488 | |
| 489 | /* |
| 490 | * The scheduler clock typically runs at a 20Hz rate. NOTE! systimer, |
| 491 | * the MP lock might not be held. We can safely manipulate parts of curproc |
| 492 | * but that's about it. |
| 493 | */ |
| 494 | static void |
| 495 | schedclock(systimer_t info, struct intrframe *frame) |
| 496 | { |
| 497 | struct proc *p; |
| 498 | struct pstats *pstats; |
| 499 | struct rusage *ru; |
| 500 | struct vmspace *vm; |
| 501 | long rss; |
| 502 | |
| 503 | schedulerclock(NULL); /* mpsafe */ |
| 504 | if ((p = curproc) != NULL) { |
| 505 | /* Update resource usage integrals and maximums. */ |
| 506 | if ((pstats = p->p_stats) != NULL && |
| 507 | (ru = &pstats->p_ru) != NULL && |
| 508 | (vm = p->p_vmspace) != NULL) { |
| 509 | ru->ru_ixrss += pgtok(vm->vm_tsize); |
| 510 | ru->ru_idrss += pgtok(vm->vm_dsize); |
| 511 | ru->ru_isrss += pgtok(vm->vm_ssize); |
| 512 | rss = pgtok(vmspace_resident_count(vm)); |
| 513 | if (ru->ru_maxrss < rss) |
| 514 | ru->ru_maxrss = rss; |
| 515 | } |
| 516 | } |
| 517 | } |
| 518 | |
| 519 | /* |
| 520 | * Compute number of ticks for the specified amount of time. The |
| 521 | * return value is intended to be used in a clock interrupt timed |
| 522 | * operation and guarenteed to meet or exceed the requested time. |
| 523 | * If the representation overflows, return INT_MAX. The minimum return |
| 524 | * value is 1 ticks and the function will average the calculation up. |
| 525 | * If any value greater then 0 microseconds is supplied, a value |
| 526 | * of at least 2 will be returned to ensure that a near-term clock |
| 527 | * interrupt does not cause the timeout to occur (degenerately) early. |
| 528 | * |
| 529 | * Note that limit checks must take into account microseconds, which is |
| 530 | * done simply by using the smaller signed long maximum instead of |
| 531 | * the unsigned long maximum. |
| 532 | * |
| 533 | * If ints have 32 bits, then the maximum value for any timeout in |
| 534 | * 10ms ticks is 248 days. |
| 535 | */ |
| 536 | int |
| 537 | tvtohz_high(struct timeval *tv) |
| 538 | { |
| 539 | int ticks; |
| 540 | long sec, usec; |
| 541 | |
| 542 | sec = tv->tv_sec; |
| 543 | usec = tv->tv_usec; |
| 544 | if (usec < 0) { |
| 545 | sec--; |
| 546 | usec += 1000000; |
| 547 | } |
| 548 | if (sec < 0) { |
| 549 | #ifdef DIAGNOSTIC |
| 550 | if (usec > 0) { |
| 551 | sec++; |
| 552 | usec -= 1000000; |
| 553 | } |
| 554 | printf("tvotohz: negative time difference %ld sec %ld usec\n", |
| 555 | sec, usec); |
| 556 | #endif |
| 557 | ticks = 1; |
| 558 | } else if (sec <= INT_MAX / hz) { |
| 559 | ticks = (int)(sec * hz + |
| 560 | ((u_long)usec + (tick - 1)) / tick) + 1; |
| 561 | } else { |
| 562 | ticks = INT_MAX; |
| 563 | } |
| 564 | return (ticks); |
| 565 | } |
| 566 | |
| 567 | /* |
| 568 | * Compute number of ticks for the specified amount of time, erroring on |
| 569 | * the side of it being too low to ensure that sleeping the returned number |
| 570 | * of ticks will not result in a late return. |
| 571 | * |
| 572 | * The supplied timeval may not be negative and should be normalized. A |
| 573 | * return value of 0 is possible if the timeval converts to less then |
| 574 | * 1 tick. |
| 575 | * |
| 576 | * If ints have 32 bits, then the maximum value for any timeout in |
| 577 | * 10ms ticks is 248 days. |
| 578 | */ |
| 579 | int |
| 580 | tvtohz_low(struct timeval *tv) |
| 581 | { |
| 582 | int ticks; |
| 583 | long sec; |
| 584 | |
| 585 | sec = tv->tv_sec; |
| 586 | if (sec <= INT_MAX / hz) |
| 587 | ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick); |
| 588 | else |
| 589 | ticks = INT_MAX; |
| 590 | return (ticks); |
| 591 | } |
| 592 | |
| 593 | |
| 594 | /* |
| 595 | * Start profiling on a process. |
| 596 | * |
| 597 | * Kernel profiling passes proc0 which never exits and hence |
| 598 | * keeps the profile clock running constantly. |
| 599 | */ |
| 600 | void |
| 601 | startprofclock(struct proc *p) |
| 602 | { |
| 603 | if ((p->p_flag & P_PROFIL) == 0) { |
| 604 | p->p_flag |= P_PROFIL; |
| 605 | #if 0 /* XXX */ |
| 606 | if (++profprocs == 1 && stathz != 0) { |
| 607 | s = splstatclock(); |
| 608 | psdiv = psratio; |
| 609 | setstatclockrate(profhz); |
| 610 | splx(s); |
| 611 | } |
| 612 | #endif |
| 613 | } |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * Stop profiling on a process. |
| 618 | */ |
| 619 | void |
| 620 | stopprofclock(struct proc *p) |
| 621 | { |
| 622 | if (p->p_flag & P_PROFIL) { |
| 623 | p->p_flag &= ~P_PROFIL; |
| 624 | #if 0 /* XXX */ |
| 625 | if (--profprocs == 0 && stathz != 0) { |
| 626 | s = splstatclock(); |
| 627 | psdiv = 1; |
| 628 | setstatclockrate(stathz); |
| 629 | splx(s); |
| 630 | } |
| 631 | #endif |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | /* |
| 636 | * Return information about system clocks. |
| 637 | */ |
| 638 | static int |
| 639 | sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) |
| 640 | { |
| 641 | struct kinfo_clockinfo clkinfo; |
| 642 | /* |
| 643 | * Construct clockinfo structure. |
| 644 | */ |
| 645 | clkinfo.ci_hz = hz; |
| 646 | clkinfo.ci_tick = tick; |
| 647 | clkinfo.ci_tickadj = tickadj; |
| 648 | clkinfo.ci_profhz = profhz; |
| 649 | clkinfo.ci_stathz = stathz ? stathz : hz; |
| 650 | return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); |
| 651 | } |
| 652 | |
| 653 | SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, |
| 654 | 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); |
| 655 | |
| 656 | /* |
| 657 | * We have eight functions for looking at the clock, four for |
| 658 | * microseconds and four for nanoseconds. For each there is fast |
| 659 | * but less precise version "get{nano|micro}[up]time" which will |
| 660 | * return a time which is up to 1/HZ previous to the call, whereas |
| 661 | * the raw version "{nano|micro}[up]time" will return a timestamp |
| 662 | * which is as precise as possible. The "up" variants return the |
| 663 | * time relative to system boot, these are well suited for time |
| 664 | * interval measurements. |
| 665 | * |
| 666 | * Each cpu independantly maintains the current time of day, so all |
| 667 | * we need to do to protect ourselves from changes is to do a loop |
| 668 | * check on the seconds field changing out from under us. |
| 669 | * |
| 670 | * The system timer maintains a 32 bit count and due to various issues |
| 671 | * it is possible for the calculated delta to occassionally exceed |
| 672 | * cputimer_freq. If this occurs the cputimer_freq64_nsec multiplication |
| 673 | * can easily overflow, so we deal with the case. For uniformity we deal |
| 674 | * with the case in the usec case too. |
| 675 | */ |
| 676 | void |
| 677 | getmicrouptime(struct timeval *tvp) |
| 678 | { |
| 679 | struct globaldata *gd = mycpu; |
| 680 | sysclock_t delta; |
| 681 | |
| 682 | do { |
| 683 | tvp->tv_sec = gd->gd_time_seconds; |
| 684 | delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; |
| 685 | } while (tvp->tv_sec != gd->gd_time_seconds); |
| 686 | |
| 687 | if (delta >= cputimer_freq) { |
| 688 | tvp->tv_sec += delta / cputimer_freq; |
| 689 | delta %= cputimer_freq; |
| 690 | } |
| 691 | tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; |
| 692 | if (tvp->tv_usec >= 1000000) { |
| 693 | tvp->tv_usec -= 1000000; |
| 694 | ++tvp->tv_sec; |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | void |
| 699 | getnanouptime(struct timespec *tsp) |
| 700 | { |
| 701 | struct globaldata *gd = mycpu; |
| 702 | sysclock_t delta; |
| 703 | |
| 704 | do { |
| 705 | tsp->tv_sec = gd->gd_time_seconds; |
| 706 | delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; |
| 707 | } while (tsp->tv_sec != gd->gd_time_seconds); |
| 708 | |
| 709 | if (delta >= cputimer_freq) { |
| 710 | tsp->tv_sec += delta / cputimer_freq; |
| 711 | delta %= cputimer_freq; |
| 712 | } |
| 713 | tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; |
| 714 | } |
| 715 | |
| 716 | void |
| 717 | microuptime(struct timeval *tvp) |
| 718 | { |
| 719 | struct globaldata *gd = mycpu; |
| 720 | sysclock_t delta; |
| 721 | |
| 722 | do { |
| 723 | tvp->tv_sec = gd->gd_time_seconds; |
| 724 | delta = cputimer_count() - gd->gd_cpuclock_base; |
| 725 | } while (tvp->tv_sec != gd->gd_time_seconds); |
| 726 | |
| 727 | if (delta >= cputimer_freq) { |
| 728 | tvp->tv_sec += delta / cputimer_freq; |
| 729 | delta %= cputimer_freq; |
| 730 | } |
| 731 | tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; |
| 732 | } |
| 733 | |
| 734 | void |
| 735 | nanouptime(struct timespec *tsp) |
| 736 | { |
| 737 | struct globaldata *gd = mycpu; |
| 738 | sysclock_t delta; |
| 739 | |
| 740 | do { |
| 741 | tsp->tv_sec = gd->gd_time_seconds; |
| 742 | delta = cputimer_count() - gd->gd_cpuclock_base; |
| 743 | } while (tsp->tv_sec != gd->gd_time_seconds); |
| 744 | |
| 745 | if (delta >= cputimer_freq) { |
| 746 | tsp->tv_sec += delta / cputimer_freq; |
| 747 | delta %= cputimer_freq; |
| 748 | } |
| 749 | tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; |
| 750 | } |
| 751 | |
| 752 | /* |
| 753 | * realtime routines |
| 754 | */ |
| 755 | |
| 756 | void |
| 757 | getmicrotime(struct timeval *tvp) |
| 758 | { |
| 759 | struct globaldata *gd = mycpu; |
| 760 | sysclock_t delta; |
| 761 | |
| 762 | do { |
| 763 | tvp->tv_sec = gd->gd_time_seconds; |
| 764 | delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; |
| 765 | } while (tvp->tv_sec != gd->gd_time_seconds); |
| 766 | |
| 767 | if (delta >= cputimer_freq) { |
| 768 | tvp->tv_sec += delta / cputimer_freq; |
| 769 | delta %= cputimer_freq; |
| 770 | } |
| 771 | tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; |
| 772 | |
| 773 | tvp->tv_sec += basetime.tv_sec; |
| 774 | tvp->tv_usec += basetime.tv_nsec / 1000; |
| 775 | while (tvp->tv_usec >= 1000000) { |
| 776 | tvp->tv_usec -= 1000000; |
| 777 | ++tvp->tv_sec; |
| 778 | } |
| 779 | } |
| 780 | |
| 781 | void |
| 782 | getnanotime(struct timespec *tsp) |
| 783 | { |
| 784 | struct globaldata *gd = mycpu; |
| 785 | sysclock_t delta; |
| 786 | |
| 787 | do { |
| 788 | tsp->tv_sec = gd->gd_time_seconds; |
| 789 | delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; |
| 790 | } while (tsp->tv_sec != gd->gd_time_seconds); |
| 791 | |
| 792 | if (delta >= cputimer_freq) { |
| 793 | tsp->tv_sec += delta / cputimer_freq; |
| 794 | delta %= cputimer_freq; |
| 795 | } |
| 796 | tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; |
| 797 | |
| 798 | tsp->tv_sec += basetime.tv_sec; |
| 799 | tsp->tv_nsec += basetime.tv_nsec; |
| 800 | while (tsp->tv_nsec >= 1000000000) { |
| 801 | tsp->tv_nsec -= 1000000000; |
| 802 | ++tsp->tv_sec; |
| 803 | } |
| 804 | } |
| 805 | |
| 806 | void |
| 807 | microtime(struct timeval *tvp) |
| 808 | { |
| 809 | struct globaldata *gd = mycpu; |
| 810 | sysclock_t delta; |
| 811 | |
| 812 | do { |
| 813 | tvp->tv_sec = gd->gd_time_seconds; |
| 814 | delta = cputimer_count() - gd->gd_cpuclock_base; |
| 815 | } while (tvp->tv_sec != gd->gd_time_seconds); |
| 816 | |
| 817 | if (delta >= cputimer_freq) { |
| 818 | tvp->tv_sec += delta / cputimer_freq; |
| 819 | delta %= cputimer_freq; |
| 820 | } |
| 821 | tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; |
| 822 | |
| 823 | tvp->tv_sec += basetime.tv_sec; |
| 824 | tvp->tv_usec += basetime.tv_nsec / 1000; |
| 825 | while (tvp->tv_usec >= 1000000) { |
| 826 | tvp->tv_usec -= 1000000; |
| 827 | ++tvp->tv_sec; |
| 828 | } |
| 829 | } |
| 830 | |
| 831 | void |
| 832 | nanotime(struct timespec *tsp) |
| 833 | { |
| 834 | struct globaldata *gd = mycpu; |
| 835 | sysclock_t delta; |
| 836 | |
| 837 | do { |
| 838 | tsp->tv_sec = gd->gd_time_seconds; |
| 839 | delta = cputimer_count() - gd->gd_cpuclock_base; |
| 840 | } while (tsp->tv_sec != gd->gd_time_seconds); |
| 841 | |
| 842 | if (delta >= cputimer_freq) { |
| 843 | tsp->tv_sec += delta / cputimer_freq; |
| 844 | delta %= cputimer_freq; |
| 845 | } |
| 846 | tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; |
| 847 | |
| 848 | tsp->tv_sec += basetime.tv_sec; |
| 849 | tsp->tv_nsec += basetime.tv_nsec; |
| 850 | while (tsp->tv_nsec >= 1000000000) { |
| 851 | tsp->tv_nsec -= 1000000000; |
| 852 | ++tsp->tv_sec; |
| 853 | } |
| 854 | } |
| 855 | |
| 856 | int |
| 857 | pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps) |
| 858 | { |
| 859 | pps_params_t *app; |
| 860 | struct pps_fetch_args *fapi; |
| 861 | #ifdef PPS_SYNC |
| 862 | struct pps_kcbind_args *kapi; |
| 863 | #endif |
| 864 | |
| 865 | switch (cmd) { |
| 866 | case PPS_IOC_CREATE: |
| 867 | return (0); |
| 868 | case PPS_IOC_DESTROY: |
| 869 | return (0); |
| 870 | case PPS_IOC_SETPARAMS: |
| 871 | app = (pps_params_t *)data; |
| 872 | if (app->mode & ~pps->ppscap) |
| 873 | return (EINVAL); |
| 874 | pps->ppsparam = *app; |
| 875 | return (0); |
| 876 | case PPS_IOC_GETPARAMS: |
| 877 | app = (pps_params_t *)data; |
| 878 | *app = pps->ppsparam; |
| 879 | app->api_version = PPS_API_VERS_1; |
| 880 | return (0); |
| 881 | case PPS_IOC_GETCAP: |
| 882 | *(int*)data = pps->ppscap; |
| 883 | return (0); |
| 884 | case PPS_IOC_FETCH: |
| 885 | fapi = (struct pps_fetch_args *)data; |
| 886 | if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC) |
| 887 | return (EINVAL); |
| 888 | if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec) |
| 889 | return (EOPNOTSUPP); |
| 890 | pps->ppsinfo.current_mode = pps->ppsparam.mode; |
| 891 | fapi->pps_info_buf = pps->ppsinfo; |
| 892 | return (0); |
| 893 | case PPS_IOC_KCBIND: |
| 894 | #ifdef PPS_SYNC |
| 895 | kapi = (struct pps_kcbind_args *)data; |
| 896 | /* XXX Only root should be able to do this */ |
| 897 | if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC) |
| 898 | return (EINVAL); |
| 899 | if (kapi->kernel_consumer != PPS_KC_HARDPPS) |
| 900 | return (EINVAL); |
| 901 | if (kapi->edge & ~pps->ppscap) |
| 902 | return (EINVAL); |
| 903 | pps->kcmode = kapi->edge; |
| 904 | return (0); |
| 905 | #else |
| 906 | return (EOPNOTSUPP); |
| 907 | #endif |
| 908 | default: |
| 909 | return (ENOTTY); |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | void |
| 914 | pps_init(struct pps_state *pps) |
| 915 | { |
| 916 | pps->ppscap |= PPS_TSFMT_TSPEC; |
| 917 | if (pps->ppscap & PPS_CAPTUREASSERT) |
| 918 | pps->ppscap |= PPS_OFFSETASSERT; |
| 919 | if (pps->ppscap & PPS_CAPTURECLEAR) |
| 920 | pps->ppscap |= PPS_OFFSETCLEAR; |
| 921 | } |
| 922 | |
| 923 | void |
| 924 | pps_event(struct pps_state *pps, sysclock_t count, int event) |
| 925 | { |
| 926 | struct globaldata *gd; |
| 927 | struct timespec *tsp; |
| 928 | struct timespec *osp; |
| 929 | struct timespec ts; |
| 930 | sysclock_t *pcount; |
| 931 | #ifdef PPS_SYNC |
| 932 | sysclock_t tcount; |
| 933 | #endif |
| 934 | sysclock_t delta; |
| 935 | pps_seq_t *pseq; |
| 936 | int foff; |
| 937 | int fhard; |
| 938 | |
| 939 | gd = mycpu; |
| 940 | |
| 941 | /* Things would be easier with arrays... */ |
| 942 | if (event == PPS_CAPTUREASSERT) { |
| 943 | tsp = &pps->ppsinfo.assert_timestamp; |
| 944 | osp = &pps->ppsparam.assert_offset; |
| 945 | foff = pps->ppsparam.mode & PPS_OFFSETASSERT; |
| 946 | fhard = pps->kcmode & PPS_CAPTUREASSERT; |
| 947 | pcount = &pps->ppscount[0]; |
| 948 | pseq = &pps->ppsinfo.assert_sequence; |
| 949 | } else { |
| 950 | tsp = &pps->ppsinfo.clear_timestamp; |
| 951 | osp = &pps->ppsparam.clear_offset; |
| 952 | foff = pps->ppsparam.mode & PPS_OFFSETCLEAR; |
| 953 | fhard = pps->kcmode & PPS_CAPTURECLEAR; |
| 954 | pcount = &pps->ppscount[1]; |
| 955 | pseq = &pps->ppsinfo.clear_sequence; |
| 956 | } |
| 957 | |
| 958 | /* Nothing really happened */ |
| 959 | if (*pcount == count) |
| 960 | return; |
| 961 | |
| 962 | *pcount = count; |
| 963 | |
| 964 | do { |
| 965 | ts.tv_sec = gd->gd_time_seconds; |
| 966 | delta = count - gd->gd_cpuclock_base; |
| 967 | } while (ts.tv_sec != gd->gd_time_seconds); |
| 968 | |
| 969 | if (delta >= cputimer_freq) { |
| 970 | ts.tv_sec += delta / cputimer_freq; |
| 971 | delta %= cputimer_freq; |
| 972 | } |
| 973 | ts.tv_nsec = (cputimer_freq64_nsec * delta) >> 32; |
| 974 | ts.tv_sec += basetime.tv_sec; |
| 975 | ts.tv_nsec += basetime.tv_nsec; |
| 976 | while (ts.tv_nsec >= 1000000000) { |
| 977 | ts.tv_nsec -= 1000000000; |
| 978 | ++ts.tv_sec; |
| 979 | } |
| 980 | |
| 981 | (*pseq)++; |
| 982 | *tsp = ts; |
| 983 | |
| 984 | if (foff) { |
| 985 | timespecadd(tsp, osp); |
| 986 | if (tsp->tv_nsec < 0) { |
| 987 | tsp->tv_nsec += 1000000000; |
| 988 | tsp->tv_sec -= 1; |
| 989 | } |
| 990 | } |
| 991 | #ifdef PPS_SYNC |
| 992 | if (fhard) { |
| 993 | /* magic, at its best... */ |
| 994 | tcount = count - pps->ppscount[2]; |
| 995 | pps->ppscount[2] = count; |
| 996 | if (tcount >= cputimer_freq) { |
| 997 | delta = (1000000000 * (tcount / cputimer_freq) + |
| 998 | cputimer_freq64_nsec * |
| 999 | (tcount % cputimer_freq)) >> 32; |
| 1000 | } else { |
| 1001 | delta = (cputimer_freq64_nsec * tcount) >> 32; |
| 1002 | } |
| 1003 | hardpps(tsp, delta); |
| 1004 | } |
| 1005 | #endif |
| 1006 | } |
| 1007 | |