1 /* $OpenBSD: kern_tc.c,v 1.59 2020/06/26 18:48:31 cheloha Exp $ */ 2 3 /* 4 * Copyright (c) 2000 Poul-Henning Kamp <phk@FreeBSD.org> 5 * 6 * Permission to use, copy, modify, and distribute this software for any 7 * purpose with or without fee is hereby granted, provided that the above 8 * copyright notice and this permission notice appear in all copies. 9 * 10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 17 */ 18 19 /* 20 * If we meet some day, and you think this stuff is worth it, you 21 * can buy me a beer in return. Poul-Henning Kamp 22 */ 23 24 #include <sys/param.h> 25 #include <sys/atomic.h> 26 #include <sys/kernel.h> 27 #include <sys/mutex.h> 28 #include <sys/rwlock.h> 29 #include <sys/stdint.h> 30 #include <sys/timeout.h> 31 #include <sys/sysctl.h> 32 #include <sys/syslog.h> 33 #include <sys/systm.h> 34 #include <sys/timetc.h> 35 #include <sys/queue.h> 36 #include <sys/malloc.h> 37 38 /* 39 * A large step happens on boot. This constant detects such steps. 40 * It is relatively small so that ntp_update_second gets called enough 41 * in the typical 'missed a couple of seconds' case, but doesn't loop 42 * forever when the time step is large. 43 */ 44 #define LARGE_STEP 200 45 46 u_int dummy_get_timecount(struct timecounter *); 47 48 int sysctl_tc_hardware(void *, size_t *, void *, size_t); 49 int sysctl_tc_choice(void *, size_t *, void *, size_t); 50 51 /* 52 * Implement a dummy timecounter which we can use until we get a real one 53 * in the air. This allows the console and other early stuff to use 54 * time services. 55 */ 56 57 u_int 58 dummy_get_timecount(struct timecounter *tc) 59 { 60 static u_int now; 61 62 return (++now); 63 } 64 65 static struct timecounter dummy_timecounter = { 66 dummy_get_timecount, 0, ~0u, 1000000, "dummy", -1000000 67 }; 68 69 /* 70 * Locks used to protect struct members, global variables in this file: 71 * I immutable after initialization 72 * t tc_lock 73 * w windup_mtx 74 */ 75 76 struct timehands { 77 /* These fields must be initialized by the driver. */ 78 struct timecounter *th_counter; /* [w] */ 79 int64_t th_adjtimedelta; /* [tw] */ 80 int64_t th_adjustment; /* [w] */ 81 u_int64_t th_scale; /* [w] */ 82 u_int th_offset_count; /* [w] */ 83 struct bintime th_boottime; /* [tw] */ 84 struct bintime th_offset; /* [w] */ 85 struct bintime th_naptime; /* [w] */ 86 struct timeval th_microtime; /* [w] */ 87 struct timespec th_nanotime; /* [w] */ 88 /* Fields not to be copied in tc_windup start with th_generation. */ 89 volatile u_int th_generation; /* [w] */ 90 struct timehands *th_next; /* [I] */ 91 }; 92 93 static struct timehands th0; 94 static struct timehands th1 = { 95 .th_next = &th0 96 }; 97 static struct timehands th0 = { 98 .th_counter = &dummy_timecounter, 99 .th_scale = UINT64_MAX / 1000000, 100 .th_offset = { .sec = 1, .frac = 0 }, 101 .th_generation = 1, 102 .th_next = &th1 103 }; 104 105 struct rwlock tc_lock = RWLOCK_INITIALIZER("tc_lock"); 106 107 /* 108 * tc_windup() must be called before leaving this mutex. 109 */ 110 struct mutex windup_mtx = MUTEX_INITIALIZER(IPL_CLOCK); 111 112 static struct timehands *volatile timehands = &th0; /* [w] */ 113 struct timecounter *timecounter = &dummy_timecounter; /* [t] */ 114 static SLIST_HEAD(, timecounter) tc_list = SLIST_HEAD_INITIALIZER(tc_list); 115 116 /* 117 * These are updated from tc_windup(). They are useful when 118 * examining kernel core dumps. 119 */ 120 volatile time_t time_second = 1; 121 volatile time_t time_uptime = 0; 122 123 static int timestepwarnings; 124 125 void ntp_update_second(struct timehands *); 126 void tc_windup(struct bintime *, struct bintime *, int64_t *); 127 128 /* 129 * Return the difference between the timehands' counter value now and what 130 * was when we copied it to the timehands' offset_count. 131 */ 132 static __inline u_int 133 tc_delta(struct timehands *th) 134 { 135 struct timecounter *tc; 136 137 tc = th->th_counter; 138 return ((tc->tc_get_timecount(tc) - th->th_offset_count) & 139 tc->tc_counter_mask); 140 } 141 142 /* 143 * Functions for reading the time. We have to loop until we are sure that 144 * the timehands that we operated on was not updated under our feet. See 145 * the comment in <sys/time.h> for a description of these functions. 146 */ 147 148 void 149 binboottime(struct bintime *bt) 150 { 151 struct timehands *th; 152 u_int gen; 153 154 do { 155 th = timehands; 156 gen = th->th_generation; 157 membar_consumer(); 158 *bt = th->th_boottime; 159 membar_consumer(); 160 } while (gen == 0 || gen != th->th_generation); 161 } 162 163 void 164 microboottime(struct timeval *tvp) 165 { 166 struct bintime bt; 167 168 binboottime(&bt); 169 BINTIME_TO_TIMEVAL(&bt, tvp); 170 } 171 172 void 173 nanoboottime(struct timespec *tsp) 174 { 175 struct bintime bt; 176 177 binboottime(&bt); 178 BINTIME_TO_TIMESPEC(&bt, tsp); 179 } 180 181 void 182 binuptime(struct bintime *bt) 183 { 184 struct timehands *th; 185 u_int gen; 186 187 do { 188 th = timehands; 189 gen = th->th_generation; 190 membar_consumer(); 191 *bt = th->th_offset; 192 bintimeaddfrac(bt, th->th_scale * tc_delta(th), bt); 193 membar_consumer(); 194 } while (gen == 0 || gen != th->th_generation); 195 } 196 197 void 198 nanouptime(struct timespec *tsp) 199 { 200 struct bintime bt; 201 202 binuptime(&bt); 203 BINTIME_TO_TIMESPEC(&bt, tsp); 204 } 205 206 void 207 microuptime(struct timeval *tvp) 208 { 209 struct bintime bt; 210 211 binuptime(&bt); 212 BINTIME_TO_TIMEVAL(&bt, tvp); 213 } 214 215 time_t 216 getuptime(void) 217 { 218 #if defined(__LP64__) 219 return time_uptime; /* atomic */ 220 #else 221 time_t now; 222 struct timehands *th; 223 u_int gen; 224 225 do { 226 th = timehands; 227 gen = th->th_generation; 228 membar_consumer(); 229 now = th->th_offset.sec; 230 membar_consumer(); 231 } while (gen == 0 || gen != th->th_generation); 232 233 return now; 234 #endif 235 } 236 237 void 238 binruntime(struct bintime *bt) 239 { 240 struct timehands *th; 241 u_int gen; 242 243 do { 244 th = timehands; 245 gen = th->th_generation; 246 membar_consumer(); 247 bintimeaddfrac(&th->th_offset, th->th_scale * tc_delta(th), bt); 248 bintimesub(bt, &th->th_naptime, bt); 249 membar_consumer(); 250 } while (gen == 0 || gen != th->th_generation); 251 } 252 253 void 254 nanoruntime(struct timespec *ts) 255 { 256 struct bintime bt; 257 258 binruntime(&bt); 259 BINTIME_TO_TIMESPEC(&bt, ts); 260 } 261 262 void 263 bintime(struct bintime *bt) 264 { 265 struct timehands *th; 266 u_int gen; 267 268 do { 269 th = timehands; 270 gen = th->th_generation; 271 membar_consumer(); 272 *bt = th->th_offset; 273 bintimeaddfrac(bt, th->th_scale * tc_delta(th), bt); 274 bintimeadd(bt, &th->th_boottime, bt); 275 membar_consumer(); 276 } while (gen == 0 || gen != th->th_generation); 277 } 278 279 void 280 nanotime(struct timespec *tsp) 281 { 282 struct bintime bt; 283 284 bintime(&bt); 285 BINTIME_TO_TIMESPEC(&bt, tsp); 286 } 287 288 void 289 microtime(struct timeval *tvp) 290 { 291 struct bintime bt; 292 293 bintime(&bt); 294 BINTIME_TO_TIMEVAL(&bt, tvp); 295 } 296 297 time_t 298 gettime(void) 299 { 300 #if defined(__LP64__) 301 return time_second; /* atomic */ 302 #else 303 time_t now; 304 struct timehands *th; 305 u_int gen; 306 307 do { 308 th = timehands; 309 gen = th->th_generation; 310 membar_consumer(); 311 now = th->th_microtime.tv_sec; 312 membar_consumer(); 313 } while (gen == 0 || gen != th->th_generation); 314 315 return now; 316 #endif 317 } 318 319 void 320 getnanouptime(struct timespec *tsp) 321 { 322 struct timehands *th; 323 u_int gen; 324 325 do { 326 th = timehands; 327 gen = th->th_generation; 328 membar_consumer(); 329 BINTIME_TO_TIMESPEC(&th->th_offset, tsp); 330 membar_consumer(); 331 } while (gen == 0 || gen != th->th_generation); 332 } 333 334 void 335 getmicrouptime(struct timeval *tvp) 336 { 337 struct timehands *th; 338 u_int gen; 339 340 do { 341 th = timehands; 342 gen = th->th_generation; 343 membar_consumer(); 344 BINTIME_TO_TIMEVAL(&th->th_offset, tvp); 345 membar_consumer(); 346 } while (gen == 0 || gen != th->th_generation); 347 } 348 349 void 350 getnanotime(struct timespec *tsp) 351 { 352 struct timehands *th; 353 u_int gen; 354 355 do { 356 th = timehands; 357 gen = th->th_generation; 358 membar_consumer(); 359 *tsp = th->th_nanotime; 360 membar_consumer(); 361 } while (gen == 0 || gen != th->th_generation); 362 } 363 364 void 365 getmicrotime(struct timeval *tvp) 366 { 367 struct timehands *th; 368 u_int gen; 369 370 do { 371 th = timehands; 372 gen = th->th_generation; 373 membar_consumer(); 374 *tvp = th->th_microtime; 375 membar_consumer(); 376 } while (gen == 0 || gen != th->th_generation); 377 } 378 379 /* 380 * Initialize a new timecounter and possibly use it. 381 */ 382 void 383 tc_init(struct timecounter *tc) 384 { 385 u_int64_t tmp; 386 u_int u; 387 388 u = tc->tc_frequency / tc->tc_counter_mask; 389 /* XXX: We need some margin here, 10% is a guess */ 390 u *= 11; 391 u /= 10; 392 if (tc->tc_quality >= 0) { 393 if (u > hz) { 394 tc->tc_quality = -2000; 395 printf("Timecounter \"%s\" frequency %lu Hz", 396 tc->tc_name, (unsigned long)tc->tc_frequency); 397 printf(" -- Insufficient hz, needs at least %u\n", u); 398 } 399 } 400 401 /* Determine the counter's precision. */ 402 for (tmp = 1; (tmp & tc->tc_counter_mask) == 0; tmp <<= 1) 403 continue; 404 tc->tc_precision = tmp; 405 406 SLIST_INSERT_HEAD(&tc_list, tc, tc_next); 407 408 /* 409 * Never automatically use a timecounter with negative quality. 410 * Even though we run on the dummy counter, switching here may be 411 * worse since this timecounter may not be monotonic. 412 */ 413 if (tc->tc_quality < 0) 414 return; 415 if (tc->tc_quality < timecounter->tc_quality) 416 return; 417 if (tc->tc_quality == timecounter->tc_quality && 418 tc->tc_frequency < timecounter->tc_frequency) 419 return; 420 (void)tc->tc_get_timecount(tc); 421 enqueue_randomness(tc->tc_get_timecount(tc)); 422 423 timecounter = tc; 424 } 425 426 /* Report the frequency of the current timecounter. */ 427 u_int64_t 428 tc_getfrequency(void) 429 { 430 return (timehands->th_counter->tc_frequency); 431 } 432 433 /* Report the precision of the current timecounter. */ 434 u_int64_t 435 tc_getprecision(void) 436 { 437 return (timehands->th_counter->tc_precision); 438 } 439 440 /* 441 * Step our concept of UTC, aka the realtime clock. 442 * This is done by modifying our estimate of when we booted. 443 * 444 * Any ongoing adjustment is meaningless after a clock jump, 445 * so we zero adjtimedelta here as well. 446 */ 447 void 448 tc_setrealtimeclock(const struct timespec *ts) 449 { 450 struct timespec ts2; 451 struct bintime bt, bt2; 452 int64_t zero = 0; 453 454 rw_enter_write(&tc_lock); 455 mtx_enter(&windup_mtx); 456 binuptime(&bt2); 457 TIMESPEC_TO_BINTIME(ts, &bt); 458 bintimesub(&bt, &bt2, &bt); 459 bintimeadd(&bt2, &timehands->th_boottime, &bt2); 460 461 /* XXX fiddle all the little crinkly bits around the fiords... */ 462 tc_windup(&bt, NULL, &zero); 463 mtx_leave(&windup_mtx); 464 rw_exit_write(&tc_lock); 465 466 enqueue_randomness(ts->tv_sec); 467 468 if (timestepwarnings) { 469 BINTIME_TO_TIMESPEC(&bt2, &ts2); 470 log(LOG_INFO, "Time stepped from %lld.%09ld to %lld.%09ld\n", 471 (long long)ts2.tv_sec, ts2.tv_nsec, 472 (long long)ts->tv_sec, ts->tv_nsec); 473 } 474 } 475 476 /* 477 * Step the monotonic and realtime clocks, triggering any timeouts that 478 * should have occurred across the interval. 479 */ 480 void 481 tc_setclock(const struct timespec *ts) 482 { 483 struct bintime bt, old_naptime, naptime; 484 struct timespec earlier; 485 static int first = 1; 486 #ifndef SMALL_KERNEL 487 long long adj_ticks; 488 #endif 489 490 /* 491 * When we're called for the first time, during boot when 492 * the root partition is mounted, we need to set boottime. 493 */ 494 if (first) { 495 tc_setrealtimeclock(ts); 496 first = 0; 497 return; 498 } 499 500 enqueue_randomness(ts->tv_sec); 501 502 mtx_enter(&windup_mtx); 503 TIMESPEC_TO_BINTIME(ts, &bt); 504 bintimesub(&bt, &timehands->th_boottime, &bt); 505 old_naptime = timehands->th_naptime; 506 /* XXX fiddle all the little crinkly bits around the fiords... */ 507 tc_windup(NULL, &bt, NULL); 508 naptime = timehands->th_naptime; 509 mtx_leave(&windup_mtx); 510 511 if (bintimecmp(&old_naptime, &naptime, ==)) { 512 BINTIME_TO_TIMESPEC(&bt, &earlier); 513 printf("%s: cannot rewind uptime to %lld.%09ld\n", 514 __func__, (long long)earlier.tv_sec, earlier.tv_nsec); 515 } 516 517 #ifndef SMALL_KERNEL 518 /* convert the bintime to ticks */ 519 bintimesub(&naptime, &old_naptime, &bt); 520 adj_ticks = (uint64_t)hz * bt.sec + 521 (((uint64_t)1000000 * (uint32_t)(bt.frac >> 32)) >> 32) / tick; 522 if (adj_ticks > 0) { 523 if (adj_ticks > INT_MAX) 524 adj_ticks = INT_MAX; 525 timeout_adjust_ticks(adj_ticks); 526 } 527 #endif 528 } 529 530 /* 531 * Initialize the next struct timehands in the ring and make 532 * it the active timehands. Along the way we might switch to a different 533 * timecounter and/or do seconds processing in NTP. Slightly magic. 534 */ 535 void 536 tc_windup(struct bintime *new_boottime, struct bintime *new_offset, 537 int64_t *new_adjtimedelta) 538 { 539 struct bintime bt; 540 struct timecounter *active_tc; 541 struct timehands *th, *tho; 542 u_int64_t scale; 543 u_int delta, ncount, ogen; 544 int i; 545 546 if (new_boottime != NULL || new_adjtimedelta != NULL) 547 rw_assert_wrlock(&tc_lock); 548 MUTEX_ASSERT_LOCKED(&windup_mtx); 549 550 active_tc = timecounter; 551 552 /* 553 * Make the next timehands a copy of the current one, but do not 554 * overwrite the generation or next pointer. While we update 555 * the contents, the generation must be zero. 556 */ 557 tho = timehands; 558 th = tho->th_next; 559 ogen = th->th_generation; 560 th->th_generation = 0; 561 membar_producer(); 562 memcpy(th, tho, offsetof(struct timehands, th_generation)); 563 564 /* 565 * Capture a timecounter delta on the current timecounter and if 566 * changing timecounters, a counter value from the new timecounter. 567 * Update the offset fields accordingly. 568 */ 569 delta = tc_delta(th); 570 if (th->th_counter != active_tc) 571 ncount = active_tc->tc_get_timecount(active_tc); 572 else 573 ncount = 0; 574 th->th_offset_count += delta; 575 th->th_offset_count &= th->th_counter->tc_counter_mask; 576 bintimeaddfrac(&th->th_offset, th->th_scale * delta, &th->th_offset); 577 578 /* 579 * Ignore new offsets that predate the current offset. 580 * If changing the offset, first increase the naptime 581 * accordingly. 582 */ 583 if (new_offset != NULL && bintimecmp(&th->th_offset, new_offset, <)) { 584 bintimesub(new_offset, &th->th_offset, &bt); 585 bintimeadd(&th->th_naptime, &bt, &th->th_naptime); 586 th->th_offset = *new_offset; 587 } 588 589 #ifdef notyet 590 /* 591 * Hardware latching timecounters may not generate interrupts on 592 * PPS events, so instead we poll them. There is a finite risk that 593 * the hardware might capture a count which is later than the one we 594 * got above, and therefore possibly in the next NTP second which might 595 * have a different rate than the current NTP second. It doesn't 596 * matter in practice. 597 */ 598 if (tho->th_counter->tc_poll_pps) 599 tho->th_counter->tc_poll_pps(tho->th_counter); 600 #endif 601 602 /* 603 * If changing the boot time or clock adjustment, do so before 604 * NTP processing. 605 */ 606 if (new_boottime != NULL) 607 th->th_boottime = *new_boottime; 608 if (new_adjtimedelta != NULL) 609 th->th_adjtimedelta = *new_adjtimedelta; 610 611 /* 612 * Deal with NTP second processing. The for loop normally 613 * iterates at most once, but in extreme situations it might 614 * keep NTP sane if timeouts are not run for several seconds. 615 * At boot, the time step can be large when the TOD hardware 616 * has been read, so on really large steps, we call 617 * ntp_update_second only twice. We need to call it twice in 618 * case we missed a leap second. 619 */ 620 bt = th->th_offset; 621 bintimeadd(&bt, &th->th_boottime, &bt); 622 i = bt.sec - tho->th_microtime.tv_sec; 623 if (i > LARGE_STEP) 624 i = 2; 625 for (; i > 0; i--) 626 ntp_update_second(th); 627 628 /* Update the UTC timestamps used by the get*() functions. */ 629 /* XXX shouldn't do this here. Should force non-`get' versions. */ 630 BINTIME_TO_TIMEVAL(&bt, &th->th_microtime); 631 BINTIME_TO_TIMESPEC(&bt, &th->th_nanotime); 632 633 /* Now is a good time to change timecounters. */ 634 if (th->th_counter != active_tc) { 635 th->th_counter = active_tc; 636 th->th_offset_count = ncount; 637 } 638 639 /*- 640 * Recalculate the scaling factor. We want the number of 1/2^64 641 * fractions of a second per period of the hardware counter, taking 642 * into account the th_adjustment factor which the NTP PLL/adjtime(2) 643 * processing provides us with. 644 * 645 * The th_adjustment is nanoseconds per second with 32 bit binary 646 * fraction and we want 64 bit binary fraction of second: 647 * 648 * x = a * 2^32 / 10^9 = a * 4.294967296 649 * 650 * The range of th_adjustment is +/- 5000PPM so inside a 64bit int 651 * we can only multiply by about 850 without overflowing, but that 652 * leaves suitably precise fractions for multiply before divide. 653 * 654 * Divide before multiply with a fraction of 2199/512 results in a 655 * systematic undercompensation of 10PPM of th_adjustment. On a 656 * 5000PPM adjustment this is a 0.05PPM error. This is acceptable. 657 * 658 * We happily sacrifice the lowest of the 64 bits of our result 659 * to the goddess of code clarity. 660 * 661 */ 662 scale = (u_int64_t)1 << 63; 663 scale += \ 664 ((th->th_adjustment + th->th_counter->tc_freq_adj) / 1024) * 2199; 665 scale /= th->th_counter->tc_frequency; 666 th->th_scale = scale * 2; 667 668 /* 669 * Now that the struct timehands is again consistent, set the new 670 * generation number, making sure to not make it zero. 671 */ 672 if (++ogen == 0) 673 ogen = 1; 674 membar_producer(); 675 th->th_generation = ogen; 676 677 /* Go live with the new struct timehands. */ 678 time_second = th->th_microtime.tv_sec; 679 time_uptime = th->th_offset.sec; 680 membar_producer(); 681 timehands = th; 682 } 683 684 /* Report or change the active timecounter hardware. */ 685 int 686 sysctl_tc_hardware(void *oldp, size_t *oldlenp, void *newp, size_t newlen) 687 { 688 char newname[32]; 689 struct timecounter *newtc, *tc; 690 int error; 691 692 tc = timecounter; 693 strlcpy(newname, tc->tc_name, sizeof(newname)); 694 695 error = sysctl_string(oldp, oldlenp, newp, newlen, newname, sizeof(newname)); 696 if (error != 0 || strcmp(newname, tc->tc_name) == 0) 697 return (error); 698 SLIST_FOREACH(newtc, &tc_list, tc_next) { 699 if (strcmp(newname, newtc->tc_name) != 0) 700 continue; 701 702 /* Warm up new timecounter. */ 703 (void)newtc->tc_get_timecount(newtc); 704 (void)newtc->tc_get_timecount(newtc); 705 706 rw_enter_write(&tc_lock); 707 timecounter = newtc; 708 rw_exit_write(&tc_lock); 709 710 return (0); 711 } 712 return (EINVAL); 713 } 714 715 /* Report or change the active timecounter hardware. */ 716 int 717 sysctl_tc_choice(void *oldp, size_t *oldlenp, void *newp, size_t newlen) 718 { 719 char buf[32], *spc, *choices; 720 struct timecounter *tc; 721 int error, maxlen; 722 723 if (SLIST_EMPTY(&tc_list)) 724 return (sysctl_rdstring(oldp, oldlenp, newp, "")); 725 726 spc = ""; 727 maxlen = 0; 728 SLIST_FOREACH(tc, &tc_list, tc_next) 729 maxlen += sizeof(buf); 730 choices = malloc(maxlen, M_TEMP, M_WAITOK); 731 *choices = '\0'; 732 SLIST_FOREACH(tc, &tc_list, tc_next) { 733 snprintf(buf, sizeof(buf), "%s%s(%d)", 734 spc, tc->tc_name, tc->tc_quality); 735 spc = " "; 736 strlcat(choices, buf, maxlen); 737 } 738 error = sysctl_rdstring(oldp, oldlenp, newp, choices); 739 free(choices, M_TEMP, maxlen); 740 return (error); 741 } 742 743 /* 744 * Timecounters need to be updated every so often to prevent the hardware 745 * counter from overflowing. Updating also recalculates the cached values 746 * used by the get*() family of functions, so their precision depends on 747 * the update frequency. 748 */ 749 static int tc_tick; 750 751 void 752 tc_ticktock(void) 753 { 754 static int count; 755 756 if (++count < tc_tick) 757 return; 758 if (!mtx_enter_try(&windup_mtx)) 759 return; 760 count = 0; 761 tc_windup(NULL, NULL, NULL); 762 mtx_leave(&windup_mtx); 763 } 764 765 void 766 inittimecounter(void) 767 { 768 #ifdef DEBUG 769 u_int p; 770 #endif 771 772 /* 773 * Set the initial timeout to 774 * max(1, <approx. number of hardclock ticks in a millisecond>). 775 * People should probably not use the sysctl to set the timeout 776 * to smaller than its initial value, since that value is the 777 * smallest reasonable one. If they want better timestamps they 778 * should use the non-"get"* functions. 779 */ 780 if (hz > 1000) 781 tc_tick = (hz + 500) / 1000; 782 else 783 tc_tick = 1; 784 #ifdef DEBUG 785 p = (tc_tick * 1000000) / hz; 786 printf("Timecounters tick every %d.%03u msec\n", p / 1000, p % 1000); 787 #endif 788 789 /* warm up new timecounter (again) and get rolling. */ 790 (void)timecounter->tc_get_timecount(timecounter); 791 (void)timecounter->tc_get_timecount(timecounter); 792 } 793 794 /* 795 * Return timecounter-related information. 796 */ 797 int 798 sysctl_tc(int *name, u_int namelen, void *oldp, size_t *oldlenp, 799 void *newp, size_t newlen) 800 { 801 if (namelen != 1) 802 return (ENOTDIR); 803 804 switch (name[0]) { 805 case KERN_TIMECOUNTER_TICK: 806 return (sysctl_rdint(oldp, oldlenp, newp, tc_tick)); 807 case KERN_TIMECOUNTER_TIMESTEPWARNINGS: 808 return (sysctl_int(oldp, oldlenp, newp, newlen, 809 ×tepwarnings)); 810 case KERN_TIMECOUNTER_HARDWARE: 811 return (sysctl_tc_hardware(oldp, oldlenp, newp, newlen)); 812 case KERN_TIMECOUNTER_CHOICE: 813 return (sysctl_tc_choice(oldp, oldlenp, newp, newlen)); 814 default: 815 return (EOPNOTSUPP); 816 } 817 /* NOTREACHED */ 818 } 819 820 /* 821 * Skew the timehands according to any adjtime(2) adjustment. 822 */ 823 void 824 ntp_update_second(struct timehands *th) 825 { 826 int64_t adj; 827 828 MUTEX_ASSERT_LOCKED(&windup_mtx); 829 830 if (th->th_adjtimedelta > 0) 831 adj = MIN(5000, th->th_adjtimedelta); 832 else 833 adj = MAX(-5000, th->th_adjtimedelta); 834 th->th_adjtimedelta -= adj; 835 th->th_adjustment = (adj * 1000) << 32; 836 } 837 838 void 839 tc_adjfreq(int64_t *old, int64_t *new) 840 { 841 if (old != NULL) { 842 rw_assert_anylock(&tc_lock); 843 *old = timecounter->tc_freq_adj; 844 } 845 if (new != NULL) { 846 rw_assert_wrlock(&tc_lock); 847 mtx_enter(&windup_mtx); 848 timecounter->tc_freq_adj = *new; 849 tc_windup(NULL, NULL, NULL); 850 mtx_leave(&windup_mtx); 851 } 852 } 853 854 void 855 tc_adjtime(int64_t *old, int64_t *new) 856 { 857 struct timehands *th; 858 u_int gen; 859 860 if (old != NULL) { 861 do { 862 th = timehands; 863 gen = th->th_generation; 864 membar_consumer(); 865 *old = th->th_adjtimedelta; 866 membar_consumer(); 867 } while (gen == 0 || gen != th->th_generation); 868 } 869 if (new != NULL) { 870 rw_assert_wrlock(&tc_lock); 871 mtx_enter(&windup_mtx); 872 tc_windup(NULL, NULL, new); 873 mtx_leave(&windup_mtx); 874 } 875 } 876