1 /* $OpenBSD: kern_tc.c,v 1.58 2020/06/22 21:16:07 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 volatile time_t time_second = 1; 117 volatile time_t time_uptime = 0; 118 119 static int timestepwarnings; 120 121 void ntp_update_second(struct timehands *); 122 void tc_windup(struct bintime *, struct bintime *, int64_t *); 123 124 /* 125 * Return the difference between the timehands' counter value now and what 126 * was when we copied it to the timehands' offset_count. 127 */ 128 static __inline u_int 129 tc_delta(struct timehands *th) 130 { 131 struct timecounter *tc; 132 133 tc = th->th_counter; 134 return ((tc->tc_get_timecount(tc) - th->th_offset_count) & 135 tc->tc_counter_mask); 136 } 137 138 /* 139 * Functions for reading the time. We have to loop until we are sure that 140 * the timehands that we operated on was not updated under our feet. See 141 * the comment in <sys/time.h> for a description of these functions. 142 */ 143 144 void 145 binboottime(struct bintime *bt) 146 { 147 struct timehands *th; 148 u_int gen; 149 150 do { 151 th = timehands; 152 gen = th->th_generation; 153 membar_consumer(); 154 *bt = th->th_boottime; 155 membar_consumer(); 156 } while (gen == 0 || gen != th->th_generation); 157 } 158 159 void 160 microboottime(struct timeval *tvp) 161 { 162 struct bintime bt; 163 164 binboottime(&bt); 165 BINTIME_TO_TIMEVAL(&bt, tvp); 166 } 167 168 void 169 nanoboottime(struct timespec *tsp) 170 { 171 struct bintime bt; 172 173 binboottime(&bt); 174 BINTIME_TO_TIMESPEC(&bt, tsp); 175 } 176 177 void 178 binuptime(struct bintime *bt) 179 { 180 struct timehands *th; 181 u_int gen; 182 183 do { 184 th = timehands; 185 gen = th->th_generation; 186 membar_consumer(); 187 *bt = th->th_offset; 188 bintimeaddfrac(bt, th->th_scale * tc_delta(th), bt); 189 membar_consumer(); 190 } while (gen == 0 || gen != th->th_generation); 191 } 192 193 void 194 nanouptime(struct timespec *tsp) 195 { 196 struct bintime bt; 197 198 binuptime(&bt); 199 BINTIME_TO_TIMESPEC(&bt, tsp); 200 } 201 202 void 203 microuptime(struct timeval *tvp) 204 { 205 struct bintime bt; 206 207 binuptime(&bt); 208 BINTIME_TO_TIMEVAL(&bt, tvp); 209 } 210 211 time_t 212 getuptime(void) 213 { 214 #if defined(__LP64__) 215 return time_uptime; /* atomic */ 216 #else 217 time_t now; 218 struct timehands *th; 219 u_int gen; 220 221 do { 222 th = timehands; 223 gen = th->th_generation; 224 membar_consumer(); 225 now = th->th_offset.sec; 226 membar_consumer(); 227 } while (gen == 0 || gen != th->th_generation); 228 229 return now; 230 #endif 231 } 232 233 void 234 binruntime(struct bintime *bt) 235 { 236 struct timehands *th; 237 u_int gen; 238 239 do { 240 th = timehands; 241 gen = th->th_generation; 242 membar_consumer(); 243 bintimeaddfrac(&th->th_offset, th->th_scale * tc_delta(th), bt); 244 bintimesub(bt, &th->th_naptime, bt); 245 membar_consumer(); 246 } while (gen == 0 || gen != th->th_generation); 247 } 248 249 void 250 nanoruntime(struct timespec *ts) 251 { 252 struct bintime bt; 253 254 binruntime(&bt); 255 BINTIME_TO_TIMESPEC(&bt, ts); 256 } 257 258 void 259 bintime(struct bintime *bt) 260 { 261 struct timehands *th; 262 u_int gen; 263 264 do { 265 th = timehands; 266 gen = th->th_generation; 267 membar_consumer(); 268 *bt = th->th_offset; 269 bintimeaddfrac(bt, th->th_scale * tc_delta(th), bt); 270 bintimeadd(bt, &th->th_boottime, bt); 271 membar_consumer(); 272 } while (gen == 0 || gen != th->th_generation); 273 } 274 275 void 276 nanotime(struct timespec *tsp) 277 { 278 struct bintime bt; 279 280 bintime(&bt); 281 BINTIME_TO_TIMESPEC(&bt, tsp); 282 } 283 284 void 285 microtime(struct timeval *tvp) 286 { 287 struct bintime bt; 288 289 bintime(&bt); 290 BINTIME_TO_TIMEVAL(&bt, tvp); 291 } 292 293 time_t 294 gettime(void) 295 { 296 #if defined(__LP64__) 297 return time_second; /* atomic */ 298 #else 299 time_t now; 300 struct timehands *th; 301 u_int gen; 302 303 do { 304 th = timehands; 305 gen = th->th_generation; 306 membar_consumer(); 307 now = th->th_microtime.tv_sec; 308 membar_consumer(); 309 } while (gen == 0 || gen != th->th_generation); 310 311 return now; 312 #endif 313 } 314 315 void 316 getnanouptime(struct timespec *tsp) 317 { 318 struct timehands *th; 319 u_int gen; 320 321 do { 322 th = timehands; 323 gen = th->th_generation; 324 membar_consumer(); 325 BINTIME_TO_TIMESPEC(&th->th_offset, tsp); 326 membar_consumer(); 327 } while (gen == 0 || gen != th->th_generation); 328 } 329 330 void 331 getmicrouptime(struct timeval *tvp) 332 { 333 struct timehands *th; 334 u_int gen; 335 336 do { 337 th = timehands; 338 gen = th->th_generation; 339 membar_consumer(); 340 BINTIME_TO_TIMEVAL(&th->th_offset, tvp); 341 membar_consumer(); 342 } while (gen == 0 || gen != th->th_generation); 343 } 344 345 void 346 getnanotime(struct timespec *tsp) 347 { 348 struct timehands *th; 349 u_int gen; 350 351 do { 352 th = timehands; 353 gen = th->th_generation; 354 membar_consumer(); 355 *tsp = th->th_nanotime; 356 membar_consumer(); 357 } while (gen == 0 || gen != th->th_generation); 358 } 359 360 void 361 getmicrotime(struct timeval *tvp) 362 { 363 struct timehands *th; 364 u_int gen; 365 366 do { 367 th = timehands; 368 gen = th->th_generation; 369 membar_consumer(); 370 *tvp = th->th_microtime; 371 membar_consumer(); 372 } while (gen == 0 || gen != th->th_generation); 373 } 374 375 /* 376 * Initialize a new timecounter and possibly use it. 377 */ 378 void 379 tc_init(struct timecounter *tc) 380 { 381 u_int64_t tmp; 382 u_int u; 383 384 u = tc->tc_frequency / tc->tc_counter_mask; 385 /* XXX: We need some margin here, 10% is a guess */ 386 u *= 11; 387 u /= 10; 388 if (tc->tc_quality >= 0) { 389 if (u > hz) { 390 tc->tc_quality = -2000; 391 printf("Timecounter \"%s\" frequency %lu Hz", 392 tc->tc_name, (unsigned long)tc->tc_frequency); 393 printf(" -- Insufficient hz, needs at least %u\n", u); 394 } 395 } 396 397 /* Determine the counter's precision. */ 398 for (tmp = 1; (tmp & tc->tc_counter_mask) == 0; tmp <<= 1) 399 continue; 400 tc->tc_precision = tmp; 401 402 SLIST_INSERT_HEAD(&tc_list, tc, tc_next); 403 404 /* 405 * Never automatically use a timecounter with negative quality. 406 * Even though we run on the dummy counter, switching here may be 407 * worse since this timecounter may not be monotonic. 408 */ 409 if (tc->tc_quality < 0) 410 return; 411 if (tc->tc_quality < timecounter->tc_quality) 412 return; 413 if (tc->tc_quality == timecounter->tc_quality && 414 tc->tc_frequency < timecounter->tc_frequency) 415 return; 416 (void)tc->tc_get_timecount(tc); 417 enqueue_randomness(tc->tc_get_timecount(tc)); 418 419 timecounter = tc; 420 } 421 422 /* Report the frequency of the current timecounter. */ 423 u_int64_t 424 tc_getfrequency(void) 425 { 426 return (timehands->th_counter->tc_frequency); 427 } 428 429 /* Report the precision of the current timecounter. */ 430 u_int64_t 431 tc_getprecision(void) 432 { 433 return (timehands->th_counter->tc_precision); 434 } 435 436 /* 437 * Step our concept of UTC, aka the realtime clock. 438 * This is done by modifying our estimate of when we booted. 439 * 440 * Any ongoing adjustment is meaningless after a clock jump, 441 * so we zero adjtimedelta here as well. 442 */ 443 void 444 tc_setrealtimeclock(const struct timespec *ts) 445 { 446 struct timespec ts2; 447 struct bintime bt, bt2; 448 int64_t zero = 0; 449 450 rw_enter_write(&tc_lock); 451 mtx_enter(&windup_mtx); 452 binuptime(&bt2); 453 TIMESPEC_TO_BINTIME(ts, &bt); 454 bintimesub(&bt, &bt2, &bt); 455 bintimeadd(&bt2, &timehands->th_boottime, &bt2); 456 457 /* XXX fiddle all the little crinkly bits around the fiords... */ 458 tc_windup(&bt, NULL, &zero); 459 mtx_leave(&windup_mtx); 460 rw_exit_write(&tc_lock); 461 462 enqueue_randomness(ts->tv_sec); 463 464 if (timestepwarnings) { 465 BINTIME_TO_TIMESPEC(&bt2, &ts2); 466 log(LOG_INFO, "Time stepped from %lld.%09ld to %lld.%09ld\n", 467 (long long)ts2.tv_sec, ts2.tv_nsec, 468 (long long)ts->tv_sec, ts->tv_nsec); 469 } 470 } 471 472 /* 473 * Step the monotonic and realtime clocks, triggering any timeouts that 474 * should have occurred across the interval. 475 */ 476 void 477 tc_setclock(const struct timespec *ts) 478 { 479 struct bintime bt, old_naptime, naptime; 480 struct timespec earlier; 481 static int first = 1; 482 #ifndef SMALL_KERNEL 483 long long adj_ticks; 484 #endif 485 486 /* 487 * When we're called for the first time, during boot when 488 * the root partition is mounted, we need to set boottime. 489 */ 490 if (first) { 491 tc_setrealtimeclock(ts); 492 first = 0; 493 return; 494 } 495 496 enqueue_randomness(ts->tv_sec); 497 498 mtx_enter(&windup_mtx); 499 TIMESPEC_TO_BINTIME(ts, &bt); 500 bintimesub(&bt, &timehands->th_boottime, &bt); 501 old_naptime = timehands->th_naptime; 502 /* XXX fiddle all the little crinkly bits around the fiords... */ 503 tc_windup(NULL, &bt, NULL); 504 naptime = timehands->th_naptime; 505 mtx_leave(&windup_mtx); 506 507 if (bintimecmp(&old_naptime, &naptime, ==)) { 508 BINTIME_TO_TIMESPEC(&bt, &earlier); 509 printf("%s: cannot rewind uptime to %lld.%09ld\n", 510 __func__, (long long)earlier.tv_sec, earlier.tv_nsec); 511 } 512 513 #ifndef SMALL_KERNEL 514 /* convert the bintime to ticks */ 515 bintimesub(&naptime, &old_naptime, &bt); 516 adj_ticks = (uint64_t)hz * bt.sec + 517 (((uint64_t)1000000 * (uint32_t)(bt.frac >> 32)) >> 32) / tick; 518 if (adj_ticks > 0) { 519 if (adj_ticks > INT_MAX) 520 adj_ticks = INT_MAX; 521 timeout_adjust_ticks(adj_ticks); 522 } 523 #endif 524 } 525 526 /* 527 * Initialize the next struct timehands in the ring and make 528 * it the active timehands. Along the way we might switch to a different 529 * timecounter and/or do seconds processing in NTP. Slightly magic. 530 */ 531 void 532 tc_windup(struct bintime *new_boottime, struct bintime *new_offset, 533 int64_t *new_adjtimedelta) 534 { 535 struct bintime bt; 536 struct timecounter *active_tc; 537 struct timehands *th, *tho; 538 u_int64_t scale; 539 u_int delta, ncount, ogen; 540 int i; 541 542 if (new_boottime != NULL || new_adjtimedelta != NULL) 543 rw_assert_wrlock(&tc_lock); 544 MUTEX_ASSERT_LOCKED(&windup_mtx); 545 546 active_tc = timecounter; 547 548 /* 549 * Make the next timehands a copy of the current one, but do not 550 * overwrite the generation or next pointer. While we update 551 * the contents, the generation must be zero. 552 */ 553 tho = timehands; 554 th = tho->th_next; 555 ogen = th->th_generation; 556 th->th_generation = 0; 557 membar_producer(); 558 memcpy(th, tho, offsetof(struct timehands, th_generation)); 559 560 /* 561 * Capture a timecounter delta on the current timecounter and if 562 * changing timecounters, a counter value from the new timecounter. 563 * Update the offset fields accordingly. 564 */ 565 delta = tc_delta(th); 566 if (th->th_counter != active_tc) 567 ncount = active_tc->tc_get_timecount(active_tc); 568 else 569 ncount = 0; 570 th->th_offset_count += delta; 571 th->th_offset_count &= th->th_counter->tc_counter_mask; 572 bintimeaddfrac(&th->th_offset, th->th_scale * delta, &th->th_offset); 573 574 /* 575 * Ignore new offsets that predate the current offset. 576 * If changing the offset, first increase the naptime 577 * accordingly. 578 */ 579 if (new_offset != NULL && bintimecmp(&th->th_offset, new_offset, <)) { 580 bintimesub(new_offset, &th->th_offset, &bt); 581 bintimeadd(&th->th_naptime, &bt, &th->th_naptime); 582 th->th_offset = *new_offset; 583 } 584 585 #ifdef notyet 586 /* 587 * Hardware latching timecounters may not generate interrupts on 588 * PPS events, so instead we poll them. There is a finite risk that 589 * the hardware might capture a count which is later than the one we 590 * got above, and therefore possibly in the next NTP second which might 591 * have a different rate than the current NTP second. It doesn't 592 * matter in practice. 593 */ 594 if (tho->th_counter->tc_poll_pps) 595 tho->th_counter->tc_poll_pps(tho->th_counter); 596 #endif 597 598 /* 599 * If changing the boot time or clock adjustment, do so before 600 * NTP processing. 601 */ 602 if (new_boottime != NULL) 603 th->th_boottime = *new_boottime; 604 if (new_adjtimedelta != NULL) 605 th->th_adjtimedelta = *new_adjtimedelta; 606 607 /* 608 * Deal with NTP second processing. The for loop normally 609 * iterates at most once, but in extreme situations it might 610 * keep NTP sane if timeouts are not run for several seconds. 611 * At boot, the time step can be large when the TOD hardware 612 * has been read, so on really large steps, we call 613 * ntp_update_second only twice. We need to call it twice in 614 * case we missed a leap second. 615 */ 616 bt = th->th_offset; 617 bintimeadd(&bt, &th->th_boottime, &bt); 618 i = bt.sec - tho->th_microtime.tv_sec; 619 if (i > LARGE_STEP) 620 i = 2; 621 for (; i > 0; i--) 622 ntp_update_second(th); 623 624 /* Update the UTC timestamps used by the get*() functions. */ 625 /* XXX shouldn't do this here. Should force non-`get' versions. */ 626 BINTIME_TO_TIMEVAL(&bt, &th->th_microtime); 627 BINTIME_TO_TIMESPEC(&bt, &th->th_nanotime); 628 629 /* Now is a good time to change timecounters. */ 630 if (th->th_counter != active_tc) { 631 th->th_counter = active_tc; 632 th->th_offset_count = ncount; 633 } 634 635 /*- 636 * Recalculate the scaling factor. We want the number of 1/2^64 637 * fractions of a second per period of the hardware counter, taking 638 * into account the th_adjustment factor which the NTP PLL/adjtime(2) 639 * processing provides us with. 640 * 641 * The th_adjustment is nanoseconds per second with 32 bit binary 642 * fraction and we want 64 bit binary fraction of second: 643 * 644 * x = a * 2^32 / 10^9 = a * 4.294967296 645 * 646 * The range of th_adjustment is +/- 5000PPM so inside a 64bit int 647 * we can only multiply by about 850 without overflowing, but that 648 * leaves suitably precise fractions for multiply before divide. 649 * 650 * Divide before multiply with a fraction of 2199/512 results in a 651 * systematic undercompensation of 10PPM of th_adjustment. On a 652 * 5000PPM adjustment this is a 0.05PPM error. This is acceptable. 653 * 654 * We happily sacrifice the lowest of the 64 bits of our result 655 * to the goddess of code clarity. 656 * 657 */ 658 scale = (u_int64_t)1 << 63; 659 scale += \ 660 ((th->th_adjustment + th->th_counter->tc_freq_adj) / 1024) * 2199; 661 scale /= th->th_counter->tc_frequency; 662 th->th_scale = scale * 2; 663 664 /* 665 * Now that the struct timehands is again consistent, set the new 666 * generation number, making sure to not make it zero. 667 */ 668 if (++ogen == 0) 669 ogen = 1; 670 membar_producer(); 671 th->th_generation = ogen; 672 673 /* Go live with the new struct timehands. */ 674 time_second = th->th_microtime.tv_sec; 675 time_uptime = th->th_offset.sec; 676 membar_producer(); 677 timehands = th; 678 } 679 680 /* Report or change the active timecounter hardware. */ 681 int 682 sysctl_tc_hardware(void *oldp, size_t *oldlenp, void *newp, size_t newlen) 683 { 684 char newname[32]; 685 struct timecounter *newtc, *tc; 686 int error; 687 688 tc = timecounter; 689 strlcpy(newname, tc->tc_name, sizeof(newname)); 690 691 error = sysctl_string(oldp, oldlenp, newp, newlen, newname, sizeof(newname)); 692 if (error != 0 || strcmp(newname, tc->tc_name) == 0) 693 return (error); 694 SLIST_FOREACH(newtc, &tc_list, tc_next) { 695 if (strcmp(newname, newtc->tc_name) != 0) 696 continue; 697 698 /* Warm up new timecounter. */ 699 (void)newtc->tc_get_timecount(newtc); 700 (void)newtc->tc_get_timecount(newtc); 701 702 rw_enter_write(&tc_lock); 703 timecounter = newtc; 704 rw_exit_write(&tc_lock); 705 706 return (0); 707 } 708 return (EINVAL); 709 } 710 711 /* Report or change the active timecounter hardware. */ 712 int 713 sysctl_tc_choice(void *oldp, size_t *oldlenp, void *newp, size_t newlen) 714 { 715 char buf[32], *spc, *choices; 716 struct timecounter *tc; 717 int error, maxlen; 718 719 if (SLIST_EMPTY(&tc_list)) 720 return (sysctl_rdstring(oldp, oldlenp, newp, "")); 721 722 spc = ""; 723 maxlen = 0; 724 SLIST_FOREACH(tc, &tc_list, tc_next) 725 maxlen += sizeof(buf); 726 choices = malloc(maxlen, M_TEMP, M_WAITOK); 727 *choices = '\0'; 728 SLIST_FOREACH(tc, &tc_list, tc_next) { 729 snprintf(buf, sizeof(buf), "%s%s(%d)", 730 spc, tc->tc_name, tc->tc_quality); 731 spc = " "; 732 strlcat(choices, buf, maxlen); 733 } 734 error = sysctl_rdstring(oldp, oldlenp, newp, choices); 735 free(choices, M_TEMP, maxlen); 736 return (error); 737 } 738 739 /* 740 * Timecounters need to be updated every so often to prevent the hardware 741 * counter from overflowing. Updating also recalculates the cached values 742 * used by the get*() family of functions, so their precision depends on 743 * the update frequency. 744 */ 745 static int tc_tick; 746 747 void 748 tc_ticktock(void) 749 { 750 static int count; 751 752 if (++count < tc_tick) 753 return; 754 if (!mtx_enter_try(&windup_mtx)) 755 return; 756 count = 0; 757 tc_windup(NULL, NULL, NULL); 758 mtx_leave(&windup_mtx); 759 } 760 761 void 762 inittimecounter(void) 763 { 764 #ifdef DEBUG 765 u_int p; 766 #endif 767 768 /* 769 * Set the initial timeout to 770 * max(1, <approx. number of hardclock ticks in a millisecond>). 771 * People should probably not use the sysctl to set the timeout 772 * to smaller than its initial value, since that value is the 773 * smallest reasonable one. If they want better timestamps they 774 * should use the non-"get"* functions. 775 */ 776 if (hz > 1000) 777 tc_tick = (hz + 500) / 1000; 778 else 779 tc_tick = 1; 780 #ifdef DEBUG 781 p = (tc_tick * 1000000) / hz; 782 printf("Timecounters tick every %d.%03u msec\n", p / 1000, p % 1000); 783 #endif 784 785 /* warm up new timecounter (again) and get rolling. */ 786 (void)timecounter->tc_get_timecount(timecounter); 787 (void)timecounter->tc_get_timecount(timecounter); 788 } 789 790 /* 791 * Return timecounter-related information. 792 */ 793 int 794 sysctl_tc(int *name, u_int namelen, void *oldp, size_t *oldlenp, 795 void *newp, size_t newlen) 796 { 797 if (namelen != 1) 798 return (ENOTDIR); 799 800 switch (name[0]) { 801 case KERN_TIMECOUNTER_TICK: 802 return (sysctl_rdint(oldp, oldlenp, newp, tc_tick)); 803 case KERN_TIMECOUNTER_TIMESTEPWARNINGS: 804 return (sysctl_int(oldp, oldlenp, newp, newlen, 805 ×tepwarnings)); 806 case KERN_TIMECOUNTER_HARDWARE: 807 return (sysctl_tc_hardware(oldp, oldlenp, newp, newlen)); 808 case KERN_TIMECOUNTER_CHOICE: 809 return (sysctl_tc_choice(oldp, oldlenp, newp, newlen)); 810 default: 811 return (EOPNOTSUPP); 812 } 813 /* NOTREACHED */ 814 } 815 816 /* 817 * Skew the timehands according to any adjtime(2) adjustment. 818 */ 819 void 820 ntp_update_second(struct timehands *th) 821 { 822 int64_t adj; 823 824 MUTEX_ASSERT_LOCKED(&windup_mtx); 825 826 if (th->th_adjtimedelta > 0) 827 adj = MIN(5000, th->th_adjtimedelta); 828 else 829 adj = MAX(-5000, th->th_adjtimedelta); 830 th->th_adjtimedelta -= adj; 831 th->th_adjustment = (adj * 1000) << 32; 832 } 833 834 void 835 tc_adjfreq(int64_t *old, int64_t *new) 836 { 837 if (old != NULL) { 838 rw_assert_anylock(&tc_lock); 839 *old = timecounter->tc_freq_adj; 840 } 841 if (new != NULL) { 842 rw_assert_wrlock(&tc_lock); 843 mtx_enter(&windup_mtx); 844 timecounter->tc_freq_adj = *new; 845 tc_windup(NULL, NULL, NULL); 846 mtx_leave(&windup_mtx); 847 } 848 } 849 850 void 851 tc_adjtime(int64_t *old, int64_t *new) 852 { 853 struct timehands *th; 854 u_int gen; 855 856 if (old != NULL) { 857 do { 858 th = timehands; 859 gen = th->th_generation; 860 membar_consumer(); 861 *old = th->th_adjtimedelta; 862 membar_consumer(); 863 } while (gen == 0 || gen != th->th_generation); 864 } 865 if (new != NULL) { 866 rw_assert_wrlock(&tc_lock); 867 mtx_enter(&windup_mtx); 868 tc_windup(NULL, NULL, new); 869 mtx_leave(&windup_mtx); 870 } 871 } 872