kern_clock.c revision 44146
1/*- 2 * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org> 3 * Copyright (c) 1982, 1986, 1991, 1993 4 * The Regents of the University of California. All rights reserved. 5 * (c) UNIX System Laboratories, Inc. 6 * All or some portions of this file are derived from material licensed 7 * to the University of California by American Telephone and Telegraph 8 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 9 * the permission of UNIX System Laboratories, Inc. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 40 * $Id: kern_clock.c,v 1.86 1998/11/29 20:31:02 phk Exp $ 41 */ 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/dkstat.h> 46#include <sys/callout.h> 47#include <sys/kernel.h> 48#include <sys/proc.h> 49#include <sys/malloc.h> 50#include <sys/resourcevar.h> 51#include <sys/signalvar.h> 52#include <sys/timex.h> 53#include <vm/vm.h> 54#include <sys/lock.h> 55#include <vm/pmap.h> 56#include <vm/vm_map.h> 57#include <sys/sysctl.h> 58 59#include <machine/cpu.h> 60#include <machine/limits.h> 61 62#ifdef GPROF 63#include <sys/gmon.h> 64#endif 65 66#if defined(SMP) && defined(BETTER_CLOCK) 67#include <machine/smp.h> 68#endif 69 70/* This is where the NTIMECOUNTER option hangs out */ 71#include "opt_ntp.h" 72 73/* 74 * Number of timecounters used to implement stable storage 75 */ 76#ifndef NTIMECOUNTER 77#define NTIMECOUNTER 5 78#endif 79 80static MALLOC_DEFINE(M_TIMECOUNTER, "timecounter", 81 "Timecounter stable storage"); 82 83static void initclocks __P((void *dummy)); 84SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) 85 86static void tco_forward __P((int force)); 87static void tco_setscales __P((struct timecounter *tc)); 88static __inline unsigned tco_delta __P((struct timecounter *tc)); 89 90/* Some of these don't belong here, but it's easiest to concentrate them. */ 91#if defined(SMP) && defined(BETTER_CLOCK) 92long cp_time[CPUSTATES]; 93#else 94static long cp_time[CPUSTATES]; 95#endif 96 97long tk_cancc; 98long tk_nin; 99long tk_nout; 100long tk_rawcc; 101 102time_t time_second; 103 104/* 105 * Which update policy to use. 106 * 0 - every tick, bad hardware may fail with "calcru negative..." 107 * 1 - more resistent to the above hardware, but less efficient. 108 */ 109static int tco_method; 110 111/* 112 * Implement a dummy timecounter which we can use until we get a real one 113 * in the air. This allows the console and other early stuff to use 114 * timeservices. 115 */ 116 117static unsigned 118dummy_get_timecount(struct timecounter *tc) 119{ 120 static unsigned now; 121 return (++now); 122} 123 124static struct timecounter dummy_timecounter = { 125 dummy_get_timecount, 126 0, 127 ~0u, 128 1000000, 129 "dummy" 130}; 131 132struct timecounter *timecounter = &dummy_timecounter; 133 134/* 135 * Clock handling routines. 136 * 137 * This code is written to operate with two timers that run independently of 138 * each other. 139 * 140 * The main timer, running hz times per second, is used to trigger interval 141 * timers, timeouts and rescheduling as needed. 142 * 143 * The second timer handles kernel and user profiling, 144 * and does resource use estimation. If the second timer is programmable, 145 * it is randomized to avoid aliasing between the two clocks. For example, 146 * the randomization prevents an adversary from always giving up the cpu 147 * just before its quantum expires. Otherwise, it would never accumulate 148 * cpu ticks. The mean frequency of the second timer is stathz. 149 * 150 * If no second timer exists, stathz will be zero; in this case we drive 151 * profiling and statistics off the main clock. This WILL NOT be accurate; 152 * do not do it unless absolutely necessary. 153 * 154 * The statistics clock may (or may not) be run at a higher rate while 155 * profiling. This profile clock runs at profhz. We require that profhz 156 * be an integral multiple of stathz. 157 * 158 * If the statistics clock is running fast, it must be divided by the ratio 159 * profhz/stathz for statistics. (For profiling, every tick counts.) 160 * 161 * Time-of-day is maintained using a "timecounter", which may or may 162 * not be related to the hardware generating the above mentioned 163 * interrupts. 164 */ 165 166int stathz; 167int profhz; 168static int profprocs; 169int ticks; 170static int psdiv, pscnt; /* prof => stat divider */ 171int psratio; /* ratio: prof / stat */ 172 173/* 174 * Initialize clock frequencies and start both clocks running. 175 */ 176/* ARGSUSED*/ 177static void 178initclocks(dummy) 179 void *dummy; 180{ 181 register int i; 182 183 /* 184 * Set divisors to 1 (normal case) and let the machine-specific 185 * code do its bit. 186 */ 187 psdiv = pscnt = 1; 188 cpu_initclocks(); 189 190 /* 191 * Compute profhz/stathz, and fix profhz if needed. 192 */ 193 i = stathz ? stathz : hz; 194 if (profhz == 0) 195 profhz = i; 196 psratio = profhz / i; 197} 198 199/* 200 * The real-time timer, interrupting hz times per second. 201 */ 202void 203hardclock(frame) 204 register struct clockframe *frame; 205{ 206 register struct proc *p; 207 208 p = curproc; 209 if (p) { 210 register struct pstats *pstats; 211 212 /* 213 * Run current process's virtual and profile time, as needed. 214 */ 215 pstats = p->p_stats; 216 if (CLKF_USERMODE(frame) && 217 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 218 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 219 psignal(p, SIGVTALRM); 220 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 221 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 222 psignal(p, SIGPROF); 223 } 224 225#if defined(SMP) && defined(BETTER_CLOCK) 226 forward_hardclock(pscnt); 227#endif 228 229 /* 230 * If no separate statistics clock is available, run it from here. 231 */ 232 if (stathz == 0) 233 statclock(frame); 234 235 tco_forward(0); 236 ticks++; 237 238 /* 239 * Process callouts at a very low cpu priority, so we don't keep the 240 * relatively high clock interrupt priority any longer than necessary. 241 */ 242 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) { 243 if (CLKF_BASEPRI(frame)) { 244 /* 245 * Save the overhead of a software interrupt; 246 * it will happen as soon as we return, so do it now. 247 */ 248 (void)splsoftclock(); 249 softclock(); 250 } else 251 setsoftclock(); 252 } else if (softticks + 1 == ticks) 253 ++softticks; 254} 255 256/* 257 * Compute number of ticks in the specified amount of time. 258 */ 259int 260tvtohz(tv) 261 struct timeval *tv; 262{ 263 register unsigned long ticks; 264 register long sec, usec; 265 266 /* 267 * If the number of usecs in the whole seconds part of the time 268 * difference fits in a long, then the total number of usecs will 269 * fit in an unsigned long. Compute the total and convert it to 270 * ticks, rounding up and adding 1 to allow for the current tick 271 * to expire. Rounding also depends on unsigned long arithmetic 272 * to avoid overflow. 273 * 274 * Otherwise, if the number of ticks in the whole seconds part of 275 * the time difference fits in a long, then convert the parts to 276 * ticks separately and add, using similar rounding methods and 277 * overflow avoidance. This method would work in the previous 278 * case but it is slightly slower and assumes that hz is integral. 279 * 280 * Otherwise, round the time difference down to the maximum 281 * representable value. 282 * 283 * If ints have 32 bits, then the maximum value for any timeout in 284 * 10ms ticks is 248 days. 285 */ 286 sec = tv->tv_sec; 287 usec = tv->tv_usec; 288 if (usec < 0) { 289 sec--; 290 usec += 1000000; 291 } 292 if (sec < 0) { 293#ifdef DIAGNOSTIC 294 if (usec > 0) { 295 sec++; 296 usec -= 1000000; 297 } 298 printf("tvotohz: negative time difference %ld sec %ld usec\n", 299 sec, usec); 300#endif 301 ticks = 1; 302 } else if (sec <= LONG_MAX / 1000000) 303 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 304 / tick + 1; 305 else if (sec <= LONG_MAX / hz) 306 ticks = sec * hz 307 + ((unsigned long)usec + (tick - 1)) / tick + 1; 308 else 309 ticks = LONG_MAX; 310 if (ticks > INT_MAX) 311 ticks = INT_MAX; 312 return ((int)ticks); 313} 314 315/* 316 * Start profiling on a process. 317 * 318 * Kernel profiling passes proc0 which never exits and hence 319 * keeps the profile clock running constantly. 320 */ 321void 322startprofclock(p) 323 register struct proc *p; 324{ 325 int s; 326 327 if ((p->p_flag & P_PROFIL) == 0) { 328 p->p_flag |= P_PROFIL; 329 if (++profprocs == 1 && stathz != 0) { 330 s = splstatclock(); 331 psdiv = pscnt = psratio; 332 setstatclockrate(profhz); 333 splx(s); 334 } 335 } 336} 337 338/* 339 * Stop profiling on a process. 340 */ 341void 342stopprofclock(p) 343 register struct proc *p; 344{ 345 int s; 346 347 if (p->p_flag & P_PROFIL) { 348 p->p_flag &= ~P_PROFIL; 349 if (--profprocs == 0 && stathz != 0) { 350 s = splstatclock(); 351 psdiv = pscnt = 1; 352 setstatclockrate(stathz); 353 splx(s); 354 } 355 } 356} 357 358/* 359 * Statistics clock. Grab profile sample, and if divider reaches 0, 360 * do process and kernel statistics. 361 */ 362void 363statclock(frame) 364 register struct clockframe *frame; 365{ 366#ifdef GPROF 367 register struct gmonparam *g; 368 int i; 369#endif 370 register struct proc *p; 371 struct pstats *pstats; 372 long rss; 373 struct rusage *ru; 374 struct vmspace *vm; 375 376 if (curproc != NULL && CLKF_USERMODE(frame)) { 377 p = curproc; 378 if (p->p_flag & P_PROFIL) 379 addupc_intr(p, CLKF_PC(frame), 1); 380#if defined(SMP) && defined(BETTER_CLOCK) 381 if (stathz != 0) 382 forward_statclock(pscnt); 383#endif 384 if (--pscnt > 0) 385 return; 386 /* 387 * Came from user mode; CPU was in user state. 388 * If this process is being profiled record the tick. 389 */ 390 p->p_uticks++; 391 if (p->p_nice > NZERO) 392 cp_time[CP_NICE]++; 393 else 394 cp_time[CP_USER]++; 395 } else { 396#ifdef GPROF 397 /* 398 * Kernel statistics are just like addupc_intr, only easier. 399 */ 400 g = &_gmonparam; 401 if (g->state == GMON_PROF_ON) { 402 i = CLKF_PC(frame) - g->lowpc; 403 if (i < g->textsize) { 404 i /= HISTFRACTION * sizeof(*g->kcount); 405 g->kcount[i]++; 406 } 407 } 408#endif 409#if defined(SMP) && defined(BETTER_CLOCK) 410 if (stathz != 0) 411 forward_statclock(pscnt); 412#endif 413 if (--pscnt > 0) 414 return; 415 /* 416 * Came from kernel mode, so we were: 417 * - handling an interrupt, 418 * - doing syscall or trap work on behalf of the current 419 * user process, or 420 * - spinning in the idle loop. 421 * Whichever it is, charge the time as appropriate. 422 * Note that we charge interrupts to the current process, 423 * regardless of whether they are ``for'' that process, 424 * so that we know how much of its real time was spent 425 * in ``non-process'' (i.e., interrupt) work. 426 */ 427 p = curproc; 428 if (CLKF_INTR(frame)) { 429 if (p != NULL) 430 p->p_iticks++; 431 cp_time[CP_INTR]++; 432 } else if (p != NULL) { 433 p->p_sticks++; 434 cp_time[CP_SYS]++; 435 } else 436 cp_time[CP_IDLE]++; 437 } 438 pscnt = psdiv; 439 440 /* 441 * We maintain statistics shown by user-level statistics 442 * programs: the amount of time in each cpu state. 443 */ 444 445 /* 446 * We adjust the priority of the current process. The priority of 447 * a process gets worse as it accumulates CPU time. The cpu usage 448 * estimator (p_estcpu) is increased here. The formula for computing 449 * priorities (in kern_synch.c) will compute a different value each 450 * time p_estcpu increases by 4. The cpu usage estimator ramps up 451 * quite quickly when the process is running (linearly), and decays 452 * away exponentially, at a rate which is proportionally slower when 453 * the system is busy. The basic principal is that the system will 454 * 90% forget that the process used a lot of CPU time in 5 * loadav 455 * seconds. This causes the system to favor processes which haven't 456 * run much recently, and to round-robin among other processes. 457 */ 458 if (p != NULL) { 459 p->p_cpticks++; 460 if (++p->p_estcpu == 0) 461 p->p_estcpu--; 462 if ((p->p_estcpu & 3) == 0) { 463 resetpriority(p); 464 if (p->p_priority >= PUSER) 465 p->p_priority = p->p_usrpri; 466 } 467 468 /* Update resource usage integrals and maximums. */ 469 if ((pstats = p->p_stats) != NULL && 470 (ru = &pstats->p_ru) != NULL && 471 (vm = p->p_vmspace) != NULL) { 472 ru->ru_ixrss += vm->vm_tsize * PAGE_SIZE / 1024; 473 ru->ru_idrss += vm->vm_dsize * PAGE_SIZE / 1024; 474 ru->ru_isrss += vm->vm_ssize * PAGE_SIZE / 1024; 475 rss = vmspace_resident_count(vm) * PAGE_SIZE / 1024; 476 if (ru->ru_maxrss < rss) 477 ru->ru_maxrss = rss; 478 } 479 } 480} 481 482/* 483 * Return information about system clocks. 484 */ 485static int 486sysctl_kern_clockrate SYSCTL_HANDLER_ARGS 487{ 488 struct clockinfo clkinfo; 489 /* 490 * Construct clockinfo structure. 491 */ 492 clkinfo.hz = hz; 493 clkinfo.tick = tick; 494 clkinfo.tickadj = tickadj; 495 clkinfo.profhz = profhz; 496 clkinfo.stathz = stathz ? stathz : hz; 497 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 498} 499 500SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 501 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); 502 503static __inline unsigned 504tco_delta(struct timecounter *tc) 505{ 506 507 return ((tc->tc_get_timecount(tc) - tc->tc_offset_count) & 508 tc->tc_counter_mask); 509} 510 511/* 512 * We have four functions for looking at the clock, two for microseconds 513 * and two for nanoseconds. For each there is fast but less precise 514 * version "get{nano|micro}time" which will return a time which is up 515 * to 1/HZ previous to the call, whereas the raw version "{nano|micro}time" 516 * will return a timestamp which is as precise as possible. 517 */ 518 519void 520getmicrotime(struct timeval *tvp) 521{ 522 struct timecounter *tc; 523 524 if (!tco_method) { 525 tc = timecounter; 526 *tvp = tc->tc_microtime; 527 } else { 528 microtime(tvp); 529 } 530} 531 532void 533getnanotime(struct timespec *tsp) 534{ 535 struct timecounter *tc; 536 537 if (!tco_method) { 538 tc = timecounter; 539 *tsp = tc->tc_nanotime; 540 } else { 541 nanotime(tsp); 542 } 543} 544 545void 546microtime(struct timeval *tv) 547{ 548 struct timecounter *tc; 549 550 tc = (struct timecounter *)timecounter; 551 tv->tv_sec = tc->tc_offset_sec; 552 tv->tv_usec = tc->tc_offset_micro; 553 tv->tv_usec += ((u_int64_t)tco_delta(tc) * tc->tc_scale_micro) >> 32; 554 tv->tv_usec += boottime.tv_usec; 555 tv->tv_sec += boottime.tv_sec; 556 while (tv->tv_usec >= 1000000) { 557 tv->tv_usec -= 1000000; 558 tv->tv_sec++; 559 } 560} 561 562void 563nanotime(struct timespec *ts) 564{ 565 unsigned count; 566 u_int64_t delta; 567 struct timecounter *tc; 568 569 tc = (struct timecounter *)timecounter; 570 ts->tv_sec = tc->tc_offset_sec; 571 count = tco_delta(tc); 572 delta = tc->tc_offset_nano; 573 delta += ((u_int64_t)count * tc->tc_scale_nano_f); 574 delta >>= 32; 575 delta += ((u_int64_t)count * tc->tc_scale_nano_i); 576 delta += boottime.tv_usec * 1000; 577 ts->tv_sec += boottime.tv_sec; 578 while (delta >= 1000000000) { 579 delta -= 1000000000; 580 ts->tv_sec++; 581 } 582 ts->tv_nsec = delta; 583} 584 585void 586timecounter_timespec(unsigned count, struct timespec *ts) 587{ 588 u_int64_t delta; 589 struct timecounter *tc; 590 591 tc = (struct timecounter *)timecounter; 592 ts->tv_sec = tc->tc_offset_sec; 593 count -= tc->tc_offset_count; 594 count &= tc->tc_counter_mask; 595 delta = tc->tc_offset_nano; 596 delta += ((u_int64_t)count * tc->tc_scale_nano_f); 597 delta >>= 32; 598 delta += ((u_int64_t)count * tc->tc_scale_nano_i); 599 delta += boottime.tv_usec * 1000; 600 ts->tv_sec += boottime.tv_sec; 601 while (delta >= 1000000000) { 602 delta -= 1000000000; 603 ts->tv_sec++; 604 } 605 ts->tv_nsec = delta; 606} 607 608void 609getmicrouptime(struct timeval *tvp) 610{ 611 struct timecounter *tc; 612 613 if (!tco_method) { 614 tc = timecounter; 615 tvp->tv_sec = tc->tc_offset_sec; 616 tvp->tv_usec = tc->tc_offset_micro; 617 } else { 618 microuptime(tvp); 619 } 620} 621 622void 623getnanouptime(struct timespec *tsp) 624{ 625 struct timecounter *tc; 626 627 if (!tco_method) { 628 tc = timecounter; 629 tsp->tv_sec = tc->tc_offset_sec; 630 tsp->tv_nsec = tc->tc_offset_nano >> 32; 631 } else { 632 nanouptime(tsp); 633 } 634} 635 636void 637microuptime(struct timeval *tv) 638{ 639 struct timecounter *tc; 640 641 tc = (struct timecounter *)timecounter; 642 tv->tv_sec = tc->tc_offset_sec; 643 tv->tv_usec = tc->tc_offset_micro; 644 tv->tv_usec += ((u_int64_t)tco_delta(tc) * tc->tc_scale_micro) >> 32; 645 if (tv->tv_usec >= 1000000) { 646 tv->tv_usec -= 1000000; 647 tv->tv_sec++; 648 } 649} 650 651void 652nanouptime(struct timespec *ts) 653{ 654 unsigned count; 655 u_int64_t delta; 656 struct timecounter *tc; 657 658 tc = (struct timecounter *)timecounter; 659 ts->tv_sec = tc->tc_offset_sec; 660 count = tco_delta(tc); 661 delta = tc->tc_offset_nano; 662 delta += ((u_int64_t)count * tc->tc_scale_nano_f); 663 delta >>= 32; 664 delta += ((u_int64_t)count * tc->tc_scale_nano_i); 665 if (delta >= 1000000000) { 666 delta -= 1000000000; 667 ts->tv_sec++; 668 } 669 ts->tv_nsec = delta; 670} 671 672static void 673tco_setscales(struct timecounter *tc) 674{ 675 u_int64_t scale; 676 677 scale = 1000000000LL << 32; 678 if (tc->tc_adjustment > 0) 679 scale += (tc->tc_adjustment * 1000LL) << 10; 680 else 681 scale -= (-tc->tc_adjustment * 1000LL) << 10; 682 scale /= tc->tc_frequency; 683 tc->tc_scale_micro = scale / 1000; 684 tc->tc_scale_nano_f = scale & 0xffffffff; 685 tc->tc_scale_nano_i = scale >> 32; 686} 687 688void 689init_timecounter(struct timecounter *tc) 690{ 691 struct timespec ts1; 692 struct timecounter *t1, *t2, *t3; 693 int i; 694 695 tc->tc_adjustment = 0; 696 tco_setscales(tc); 697 tc->tc_offset_count = tc->tc_get_timecount(tc); 698 tc->tc_tweak = tc; 699 MALLOC(t1, struct timecounter *, sizeof *t1, M_TIMECOUNTER, M_WAITOK); 700 *t1 = *tc; 701 t2 = t1; 702 for (i = 1; i < NTIMECOUNTER; i++) { 703 MALLOC(t3, struct timecounter *, sizeof *t3, 704 M_TIMECOUNTER, M_WAITOK); 705 *t3 = *tc; 706 t3->tc_other = t2; 707 t2 = t3; 708 } 709 t1->tc_other = t3; 710 tc = t1; 711 712 printf("Timecounter \"%s\" frequency %lu Hz\n", 713 tc->tc_name, (u_long)tc->tc_frequency); 714 715 /* XXX: For now always start using the counter. */ 716 tc->tc_offset_count = tc->tc_get_timecount(tc); 717 nanouptime(&ts1); 718 tc->tc_offset_nano = (u_int64_t)ts1.tv_nsec << 32; 719 tc->tc_offset_micro = ts1.tv_nsec / 1000; 720 tc->tc_offset_sec = ts1.tv_sec; 721 timecounter = tc; 722} 723 724void 725set_timecounter(struct timespec *ts) 726{ 727 struct timespec ts2; 728 729 nanouptime(&ts2); 730 boottime.tv_sec = ts->tv_sec - ts2.tv_sec; 731 boottime.tv_usec = (ts->tv_nsec - ts2.tv_nsec) / 1000; 732 if (boottime.tv_usec < 0) { 733 boottime.tv_usec += 1000000; 734 boottime.tv_sec--; 735 } 736 /* fiddle all the little crinkly bits around the fiords... */ 737 tco_forward(1); 738} 739 740 741#if 0 /* Currently unused */ 742void 743switch_timecounter(struct timecounter *newtc) 744{ 745 int s; 746 struct timecounter *tc; 747 struct timespec ts; 748 749 s = splclock(); 750 tc = timecounter; 751 if (newtc == tc || newtc == tc->tc_other) { 752 splx(s); 753 return; 754 } 755 nanouptime(&ts); 756 newtc->tc_offset_sec = ts.tv_sec; 757 newtc->tc_offset_nano = (u_int64_t)ts.tv_nsec << 32; 758 newtc->tc_offset_micro = ts.tv_nsec / 1000; 759 newtc->tc_offset_count = newtc->tc_get_timecount(newtc); 760 timecounter = newtc; 761 splx(s); 762} 763#endif 764 765static struct timecounter * 766sync_other_counter(void) 767{ 768 struct timecounter *tc, *tcn, *tco; 769 unsigned delta; 770 771 tco = timecounter; 772 tc = tco->tc_other; 773 tcn = tc->tc_other; 774 *tc = *tco; 775 tc->tc_other = tcn; 776 delta = tco_delta(tc); 777 tc->tc_offset_count += delta; 778 tc->tc_offset_count &= tc->tc_counter_mask; 779 tc->tc_offset_nano += (u_int64_t)delta * tc->tc_scale_nano_f; 780 tc->tc_offset_nano += (u_int64_t)delta * tc->tc_scale_nano_i << 32; 781 return (tc); 782} 783 784static void 785tco_forward(int force) 786{ 787 struct timecounter *tc, *tco; 788 789 tco = timecounter; 790 tc = sync_other_counter(); 791 /* 792 * We may be inducing a tiny error here, the tc_poll_pps() may 793 * process a latched count which happens after the tco_delta() 794 * in sync_other_counter(), which would extend the previous 795 * counters parameters into the domain of this new one. 796 * Since the timewindow is very small for this, the error is 797 * going to be only a few weenieseconds (as Dave Mills would 798 * say), so lets just not talk more about it, OK ? 799 */ 800 if (tco->tc_poll_pps) 801 tco->tc_poll_pps(tco); 802 if (timedelta != 0) { 803 tc->tc_offset_nano += (u_int64_t)(tickdelta * 1000) << 32; 804 timedelta -= tickdelta; 805 force++; 806 } 807 808 while (tc->tc_offset_nano >= 1000000000ULL << 32) { 809 tc->tc_offset_nano -= 1000000000ULL << 32; 810 tc->tc_offset_sec++; 811 tc->tc_frequency = tc->tc_tweak->tc_frequency; 812 tc->tc_adjustment = tc->tc_tweak->tc_adjustment; 813 ntp_update_second(tc); /* XXX only needed if xntpd runs */ 814 tco_setscales(tc); 815 force++; 816 } 817 818 if (tco_method && !force) 819 return; 820 821 tc->tc_offset_micro = (tc->tc_offset_nano / 1000) >> 32; 822 823 /* Figure out the wall-clock time */ 824 tc->tc_nanotime.tv_sec = tc->tc_offset_sec + boottime.tv_sec; 825 tc->tc_nanotime.tv_nsec = 826 (tc->tc_offset_nano >> 32) + boottime.tv_usec * 1000; 827 tc->tc_microtime.tv_usec = tc->tc_offset_micro + boottime.tv_usec; 828 if (tc->tc_nanotime.tv_nsec >= 1000000000) { 829 tc->tc_nanotime.tv_nsec -= 1000000000; 830 tc->tc_microtime.tv_usec -= 1000000; 831 tc->tc_nanotime.tv_sec++; 832 } 833 time_second = tc->tc_microtime.tv_sec = tc->tc_nanotime.tv_sec; 834 835 timecounter = tc; 836} 837 838static int 839sysctl_kern_timecounter_frequency SYSCTL_HANDLER_ARGS 840{ 841 842 return (sysctl_handle_opaque(oidp, 843 &timecounter->tc_tweak->tc_frequency, 844 sizeof(timecounter->tc_tweak->tc_frequency), req)); 845} 846 847static int 848sysctl_kern_timecounter_adjustment SYSCTL_HANDLER_ARGS 849{ 850 851 return (sysctl_handle_opaque(oidp, 852 &timecounter->tc_tweak->tc_adjustment, 853 sizeof(timecounter->tc_tweak->tc_adjustment), req)); 854} 855 856SYSCTL_NODE(_kern, OID_AUTO, timecounter, CTLFLAG_RW, 0, ""); 857 858SYSCTL_INT(_kern_timecounter, KERN_ARGMAX, method, CTLFLAG_RW, &tco_method, 0, 859 "This variable determines the method used for updating timecounters. " 860 "If the default algorithm (0) fails with \"calcru negative...\" messages " 861 "try the alternate algorithm (1) which handles bad hardware better." 862 863); 864 865SYSCTL_PROC(_kern_timecounter, OID_AUTO, frequency, CTLTYPE_INT | CTLFLAG_RW, 866 0, sizeof(u_int), sysctl_kern_timecounter_frequency, "I", ""); 867 868SYSCTL_PROC(_kern_timecounter, OID_AUTO, adjustment, CTLTYPE_INT | CTLFLAG_RW, 869 0, sizeof(int), sysctl_kern_timecounter_adjustment, "I", ""); 870