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 1.26 2004/11/12 17:50:56 dillon 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/proc.h> 84 #include <sys/malloc.h> 85 #include <sys/resourcevar.h> 86 #include <sys/signalvar.h> 87 #include <sys/timex.h> 88 #include <sys/timepps.h> 89 #include <vm/vm.h> 90 #include <sys/lock.h> 91 #include <vm/pmap.h> 92 #include <vm/vm_map.h> 93 #include <sys/sysctl.h> 94 #include <sys/thread2.h> 95 96 #include <machine/cpu.h> 97 #include <machine/limits.h> 98 #include <machine/smp.h> 99 100 #ifdef GPROF 101 #include <sys/gmon.h> 102 #endif 103 104 #ifdef DEVICE_POLLING 105 extern void init_device_poll(void); 106 extern void hardclock_device_poll(void); 107 #endif /* DEVICE_POLLING */ 108 109 static void initclocks (void *dummy); 110 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) 111 112 /* 113 * Some of these don't belong here, but it's easiest to concentrate them. 114 * Note that cp_time[] counts in microseconds, but most userland programs 115 * just compare relative times against the total by delta. 116 */ 117 long cp_time[CPUSTATES]; 118 119 SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time), 120 "LU", "CPU time statistics"); 121 122 long tk_cancc; 123 long tk_nin; 124 long tk_nout; 125 long tk_rawcc; 126 127 /* 128 * boottime is used to calculate the 'real' uptime. Do not confuse this with 129 * microuptime(). microtime() is not drift compensated. The real uptime 130 * with compensation is nanotime() - bootime. boottime is recalculated 131 * whenever the real time is set based on the compensated elapsed time 132 * in seconds (gd->gd_time_seconds). 133 * 134 * basetime is used to calculate the compensated real time of day. Chunky 135 * changes to the time, aka settimeofday(), are made by modifying basetime. 136 * 137 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic. 138 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to 139 * the real time. 140 */ 141 struct timespec boottime; /* boot time (realtime) for reference only */ 142 struct timespec basetime; /* base time adjusts uptime -> realtime */ 143 time_t time_second; /* read-only 'passive' uptime in seconds */ 144 145 SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD, 146 &boottime, timeval, "System boottime"); 147 SYSCTL_STRUCT(_kern, OID_AUTO, basetime, CTLFLAG_RD, 148 &basetime, timeval, "System basetime"); 149 150 static void hardclock(systimer_t info, struct intrframe *frame); 151 static void statclock(systimer_t info, struct intrframe *frame); 152 static void schedclock(systimer_t info, struct intrframe *frame); 153 154 int ticks; /* system master ticks at hz */ 155 int clocks_running; /* tsleep/timeout clocks operational */ 156 int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */ 157 int64_t nsec_acc; /* accumulator */ 158 159 /* 160 * Finish initializing clock frequencies and start all clocks running. 161 */ 162 /* ARGSUSED*/ 163 static void 164 initclocks(void *dummy) 165 { 166 cpu_initclocks(); 167 #ifdef DEVICE_POLLING 168 init_device_poll(); 169 #endif 170 /*psratio = profhz / stathz;*/ 171 initclocks_pcpu(); 172 clocks_running = 1; 173 } 174 175 /* 176 * Called on a per-cpu basis 177 */ 178 void 179 initclocks_pcpu(void) 180 { 181 struct globaldata *gd = mycpu; 182 183 crit_enter(); 184 if (gd->gd_cpuid == 0) { 185 gd->gd_time_seconds = 1; 186 gd->gd_cpuclock_base = cputimer_count(); 187 } else { 188 /* XXX */ 189 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds; 190 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base; 191 } 192 systimer_init_periodic(&gd->gd_hardclock, hardclock, NULL, hz); 193 systimer_init_periodic(&gd->gd_statclock, statclock, NULL, stathz); 194 /* XXX correct the frequency for scheduler / estcpu tests */ 195 systimer_init_periodic(&gd->gd_schedclock, schedclock, 196 NULL, ESTCPUFREQ); 197 crit_exit(); 198 } 199 200 /* 201 * Resynchronize gd_cpuclock_base after the system has been woken up from 202 * a sleep. It is absolutely essential that all the cpus be properly 203 * synchronized. Resynching is required because nanouptime() and friends 204 * will overflow intermediate multiplications if more then 2 seconds 205 * worth of cputimer_cont() delta has built up. 206 */ 207 #ifdef SMP 208 209 static 210 void 211 restoreclocks_remote(lwkt_cpusync_t poll) 212 { 213 mycpu->gd_cpuclock_base = *(sysclock_t *)poll->cs_data; 214 mycpu->gd_time_seconds = globaldata_find(0)->gd_time_seconds; 215 } 216 217 #endif 218 219 void 220 restoreclocks(void) 221 { 222 sysclock_t base = cputimer_count(); 223 #ifdef SMP 224 lwkt_cpusync_simple(-1, restoreclocks_remote, &base); 225 #else 226 mycpu->gd_cpuclock_base = base; 227 #endif 228 } 229 230 /* 231 * This sets the current real time of day. Timespecs are in seconds and 232 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base, 233 * instead we adjust basetime so basetime + gd_* results in the current 234 * time of day. This way the gd_* fields are guarenteed to represent 235 * a monotonically increasing 'uptime' value. 236 */ 237 void 238 set_timeofday(struct timespec *ts) 239 { 240 struct timespec ts2; 241 242 /* 243 * XXX SMP / non-atomic basetime updates 244 */ 245 crit_enter(); 246 nanouptime(&ts2); 247 basetime.tv_sec = ts->tv_sec - ts2.tv_sec; 248 basetime.tv_nsec = ts->tv_nsec - ts2.tv_nsec; 249 if (basetime.tv_nsec < 0) { 250 basetime.tv_nsec += 1000000000; 251 --basetime.tv_sec; 252 } 253 boottime.tv_sec = basetime.tv_sec - mycpu->gd_time_seconds; 254 timedelta = 0; 255 crit_exit(); 256 } 257 258 /* 259 * Each cpu has its own hardclock, but we only increments ticks and softticks 260 * on cpu #0. 261 * 262 * NOTE! systimer! the MP lock might not be held here. We can only safely 263 * manipulate objects owned by the current cpu. 264 */ 265 static void 266 hardclock(systimer_t info, struct intrframe *frame) 267 { 268 sysclock_t cputicks; 269 struct proc *p; 270 struct pstats *pstats; 271 struct globaldata *gd = mycpu; 272 273 /* 274 * Realtime updates are per-cpu. Note that timer corrections as 275 * returned by microtime() and friends make an additional adjustment 276 * using a system-wise 'basetime', but the running time is always 277 * taken from the per-cpu globaldata area. Since the same clock 278 * is distributing (XXX SMP) to all cpus, the per-cpu timebases 279 * stay in synch. 280 * 281 * Note that we never allow info->time (aka gd->gd_hardclock.time) 282 * to reverse index gd_cpuclock_base, but that it is possible for 283 * it to temporarily get behind in the seconds if something in the 284 * system locks interrupts for a long period of time. Since periodic 285 * timers count events, though everything should resynch again 286 * immediately. 287 */ 288 cputicks = info->time - gd->gd_cpuclock_base; 289 if (cputicks >= cputimer_freq) { 290 ++gd->gd_time_seconds; 291 gd->gd_cpuclock_base += cputimer_freq; 292 } 293 294 /* 295 * The system-wide ticks counter and NTP related timedelta/tickdelta 296 * adjustments only occur on cpu #0. NTP adjustments are accomplished 297 * by updating basetime. 298 */ 299 if (gd->gd_cpuid == 0) { 300 struct timespec nts; 301 int leap; 302 303 ++ticks; 304 305 #ifdef DEVICE_POLLING 306 hardclock_device_poll(); /* mpsafe, short and quick */ 307 #endif /* DEVICE_POLLING */ 308 309 #if 0 310 if (tco->tc_poll_pps) 311 tco->tc_poll_pps(tco); 312 #endif 313 /* 314 * Apply adjtime corrections. At the moment only do this if 315 * we can get the MP lock to interlock with adjtime's modification 316 * of these variables. Note that basetime adjustments are not 317 * MP safe either XXX. 318 */ 319 if (timedelta != 0 && try_mplock()) { 320 basetime.tv_nsec += tickdelta * 1000; 321 if (basetime.tv_nsec >= 1000000000) { 322 basetime.tv_nsec -= 1000000000; 323 ++basetime.tv_sec; 324 } else if (basetime.tv_nsec < 0) { 325 basetime.tv_nsec += 1000000000; 326 --basetime.tv_sec; 327 } 328 timedelta -= tickdelta; 329 rel_mplock(); 330 } 331 332 /* 333 * Apply per-tick compensation. ticks_adj adjusts for both 334 * offset and frequency, and could be negative. 335 */ 336 if (nsec_adj != 0 && try_mplock()) { 337 nsec_acc += nsec_adj; 338 if (nsec_acc >= 0x100000000LL) { 339 basetime.tv_nsec += nsec_acc >> 32; 340 nsec_acc = (nsec_acc & 0xFFFFFFFFLL); 341 } else if (nsec_acc <= -0x100000000LL) { 342 basetime.tv_nsec -= -nsec_acc >> 32; 343 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL); 344 } 345 if (basetime.tv_nsec >= 1000000000) { 346 basetime.tv_nsec -= 1000000000; 347 ++basetime.tv_sec; 348 } else if (basetime.tv_nsec < 0) { 349 basetime.tv_nsec += 1000000000; 350 --basetime.tv_sec; 351 } 352 rel_mplock(); 353 } 354 355 /* 356 * If the realtime-adjusted seconds hand rolls over then tell 357 * ntp_update_second() what we did in the last second so it can 358 * calculate what to do in the next second. It may also add 359 * or subtract a leap second. 360 */ 361 getnanotime(&nts); 362 if (time_second != nts.tv_sec) { 363 leap = ntp_update_second(time_second, &nsec_adj); 364 basetime.tv_sec += leap; 365 time_second = nts.tv_sec + leap; 366 nsec_adj /= hz; 367 } 368 } 369 370 /* 371 * softticks are handled for all cpus 372 */ 373 hardclock_softtick(gd); 374 375 /* 376 * ITimer handling is per-tick, per-cpu. I don't think psignal() 377 * is mpsafe on curproc, so XXX get the mplock. 378 */ 379 if ((p = curproc) != NULL && try_mplock()) { 380 pstats = p->p_stats; 381 if (frame && CLKF_USERMODE(frame) && 382 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && 383 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) 384 psignal(p, SIGVTALRM); 385 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && 386 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) 387 psignal(p, SIGPROF); 388 rel_mplock(); 389 } 390 setdelayed(); 391 } 392 393 /* 394 * The statistics clock typically runs at a 125Hz rate, and is intended 395 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu. 396 * 397 * NOTE! systimer! the MP lock might not be held here. We can only safely 398 * manipulate objects owned by the current cpu. 399 * 400 * The stats clock is responsible for grabbing a profiling sample. 401 * Most of the statistics are only used by user-level statistics programs. 402 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and 403 * p->p_estcpu. 404 * 405 * Like the other clocks, the stat clock is called from what is effectively 406 * a fast interrupt, so the context should be the thread/process that got 407 * interrupted. 408 */ 409 static void 410 statclock(systimer_t info, struct intrframe *frame) 411 { 412 #ifdef GPROF 413 struct gmonparam *g; 414 int i; 415 #endif 416 thread_t td; 417 struct proc *p; 418 int bump; 419 struct timeval tv; 420 struct timeval *stv; 421 422 /* 423 * How big was our timeslice relative to the last time? 424 */ 425 microuptime(&tv); /* mpsafe */ 426 stv = &mycpu->gd_stattv; 427 if (stv->tv_sec == 0) { 428 bump = 1; 429 } else { 430 bump = tv.tv_usec - stv->tv_usec + 431 (tv.tv_sec - stv->tv_sec) * 1000000; 432 if (bump < 0) 433 bump = 0; 434 if (bump > 1000000) 435 bump = 1000000; 436 } 437 *stv = tv; 438 439 td = curthread; 440 p = td->td_proc; 441 442 if (frame && CLKF_USERMODE(frame)) { 443 /* 444 * Came from userland, handle user time and deal with 445 * possible process. 446 */ 447 if (p && (p->p_flag & P_PROFIL)) 448 addupc_intr(p, CLKF_PC(frame), 1); 449 td->td_uticks += bump; 450 451 /* 452 * Charge the time as appropriate 453 */ 454 if (p && p->p_nice > NZERO) 455 cp_time[CP_NICE] += bump; 456 else 457 cp_time[CP_USER] += bump; 458 } else { 459 #ifdef GPROF 460 /* 461 * Kernel statistics are just like addupc_intr, only easier. 462 */ 463 g = &_gmonparam; 464 if (g->state == GMON_PROF_ON && frame) { 465 i = CLKF_PC(frame) - g->lowpc; 466 if (i < g->textsize) { 467 i /= HISTFRACTION * sizeof(*g->kcount); 468 g->kcount[i]++; 469 } 470 } 471 #endif 472 /* 473 * Came from kernel mode, so we were: 474 * - handling an interrupt, 475 * - doing syscall or trap work on behalf of the current 476 * user process, or 477 * - spinning in the idle loop. 478 * Whichever it is, charge the time as appropriate. 479 * Note that we charge interrupts to the current process, 480 * regardless of whether they are ``for'' that process, 481 * so that we know how much of its real time was spent 482 * in ``non-process'' (i.e., interrupt) work. 483 * 484 * XXX assume system if frame is NULL. A NULL frame 485 * can occur if ipi processing is done from an splx(). 486 */ 487 if (frame && CLKF_INTR(frame)) 488 td->td_iticks += bump; 489 else 490 td->td_sticks += bump; 491 492 if (frame && CLKF_INTR(frame)) { 493 cp_time[CP_INTR] += bump; 494 } else { 495 if (td == &mycpu->gd_idlethread) 496 cp_time[CP_IDLE] += bump; 497 else 498 cp_time[CP_SYS] += bump; 499 } 500 } 501 } 502 503 /* 504 * The scheduler clock typically runs at a 20Hz rate. NOTE! systimer, 505 * the MP lock might not be held. We can safely manipulate parts of curproc 506 * but that's about it. 507 */ 508 static void 509 schedclock(systimer_t info, struct intrframe *frame) 510 { 511 struct proc *p; 512 struct pstats *pstats; 513 struct rusage *ru; 514 struct vmspace *vm; 515 long rss; 516 517 schedulerclock(NULL); /* mpsafe */ 518 if ((p = curproc) != NULL) { 519 /* Update resource usage integrals and maximums. */ 520 if ((pstats = p->p_stats) != NULL && 521 (ru = &pstats->p_ru) != NULL && 522 (vm = p->p_vmspace) != NULL) { 523 ru->ru_ixrss += pgtok(vm->vm_tsize); 524 ru->ru_idrss += pgtok(vm->vm_dsize); 525 ru->ru_isrss += pgtok(vm->vm_ssize); 526 rss = pgtok(vmspace_resident_count(vm)); 527 if (ru->ru_maxrss < rss) 528 ru->ru_maxrss = rss; 529 } 530 } 531 } 532 533 /* 534 * Compute number of ticks for the specified amount of time. The 535 * return value is intended to be used in a clock interrupt timed 536 * operation and guarenteed to meet or exceed the requested time. 537 * If the representation overflows, return INT_MAX. The minimum return 538 * value is 1 ticks and the function will average the calculation up. 539 * If any value greater then 0 microseconds is supplied, a value 540 * of at least 2 will be returned to ensure that a near-term clock 541 * interrupt does not cause the timeout to occur (degenerately) early. 542 * 543 * Note that limit checks must take into account microseconds, which is 544 * done simply by using the smaller signed long maximum instead of 545 * the unsigned long maximum. 546 * 547 * If ints have 32 bits, then the maximum value for any timeout in 548 * 10ms ticks is 248 days. 549 */ 550 int 551 tvtohz_high(struct timeval *tv) 552 { 553 int ticks; 554 long sec, usec; 555 556 sec = tv->tv_sec; 557 usec = tv->tv_usec; 558 if (usec < 0) { 559 sec--; 560 usec += 1000000; 561 } 562 if (sec < 0) { 563 #ifdef DIAGNOSTIC 564 if (usec > 0) { 565 sec++; 566 usec -= 1000000; 567 } 568 printf("tvotohz: negative time difference %ld sec %ld usec\n", 569 sec, usec); 570 #endif 571 ticks = 1; 572 } else if (sec <= INT_MAX / hz) { 573 ticks = (int)(sec * hz + 574 ((u_long)usec + (tick - 1)) / tick) + 1; 575 } else { 576 ticks = INT_MAX; 577 } 578 return (ticks); 579 } 580 581 /* 582 * Compute number of ticks for the specified amount of time, erroring on 583 * the side of it being too low to ensure that sleeping the returned number 584 * of ticks will not result in a late return. 585 * 586 * The supplied timeval may not be negative and should be normalized. A 587 * return value of 0 is possible if the timeval converts to less then 588 * 1 tick. 589 * 590 * If ints have 32 bits, then the maximum value for any timeout in 591 * 10ms ticks is 248 days. 592 */ 593 int 594 tvtohz_low(struct timeval *tv) 595 { 596 int ticks; 597 long sec; 598 599 sec = tv->tv_sec; 600 if (sec <= INT_MAX / hz) 601 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick); 602 else 603 ticks = INT_MAX; 604 return (ticks); 605 } 606 607 608 /* 609 * Start profiling on a process. 610 * 611 * Kernel profiling passes proc0 which never exits and hence 612 * keeps the profile clock running constantly. 613 */ 614 void 615 startprofclock(struct proc *p) 616 { 617 if ((p->p_flag & P_PROFIL) == 0) { 618 p->p_flag |= P_PROFIL; 619 #if 0 /* XXX */ 620 if (++profprocs == 1 && stathz != 0) { 621 s = splstatclock(); 622 psdiv = psratio; 623 setstatclockrate(profhz); 624 splx(s); 625 } 626 #endif 627 } 628 } 629 630 /* 631 * Stop profiling on a process. 632 */ 633 void 634 stopprofclock(struct proc *p) 635 { 636 if (p->p_flag & P_PROFIL) { 637 p->p_flag &= ~P_PROFIL; 638 #if 0 /* XXX */ 639 if (--profprocs == 0 && stathz != 0) { 640 s = splstatclock(); 641 psdiv = 1; 642 setstatclockrate(stathz); 643 splx(s); 644 } 645 #endif 646 } 647 } 648 649 /* 650 * Return information about system clocks. 651 */ 652 static int 653 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) 654 { 655 struct clockinfo clkinfo; 656 /* 657 * Construct clockinfo structure. 658 */ 659 clkinfo.hz = hz; 660 clkinfo.tick = tick; 661 clkinfo.tickadj = tickadj; 662 clkinfo.profhz = profhz; 663 clkinfo.stathz = stathz ? stathz : hz; 664 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); 665 } 666 667 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 668 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); 669 670 /* 671 * We have eight functions for looking at the clock, four for 672 * microseconds and four for nanoseconds. For each there is fast 673 * but less precise version "get{nano|micro}[up]time" which will 674 * return a time which is up to 1/HZ previous to the call, whereas 675 * the raw version "{nano|micro}[up]time" will return a timestamp 676 * which is as precise as possible. The "up" variants return the 677 * time relative to system boot, these are well suited for time 678 * interval measurements. 679 * 680 * Each cpu independantly maintains the current time of day, so all 681 * we need to do to protect ourselves from changes is to do a loop 682 * check on the seconds field changing out from under us. 683 * 684 * The system timer maintains a 32 bit count and due to various issues 685 * it is possible for the calculated delta to occassionally exceed 686 * cputimer_freq. If this occurs the cputimer_freq64_nsec multiplication 687 * can easily overflow, so we deal with the case. For uniformity we deal 688 * with the case in the usec case too. 689 */ 690 void 691 getmicrouptime(struct timeval *tvp) 692 { 693 struct globaldata *gd = mycpu; 694 sysclock_t delta; 695 696 do { 697 tvp->tv_sec = gd->gd_time_seconds; 698 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; 699 } while (tvp->tv_sec != gd->gd_time_seconds); 700 701 if (delta >= cputimer_freq) { 702 tvp->tv_sec += delta / cputimer_freq; 703 delta %= cputimer_freq; 704 } 705 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; 706 if (tvp->tv_usec >= 1000000) { 707 tvp->tv_usec -= 1000000; 708 ++tvp->tv_sec; 709 } 710 } 711 712 void 713 getnanouptime(struct timespec *tsp) 714 { 715 struct globaldata *gd = mycpu; 716 sysclock_t delta; 717 718 do { 719 tsp->tv_sec = gd->gd_time_seconds; 720 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; 721 } while (tsp->tv_sec != gd->gd_time_seconds); 722 723 if (delta >= cputimer_freq) { 724 tsp->tv_sec += delta / cputimer_freq; 725 delta %= cputimer_freq; 726 } 727 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; 728 } 729 730 void 731 microuptime(struct timeval *tvp) 732 { 733 struct globaldata *gd = mycpu; 734 sysclock_t delta; 735 736 do { 737 tvp->tv_sec = gd->gd_time_seconds; 738 delta = cputimer_count() - gd->gd_cpuclock_base; 739 } while (tvp->tv_sec != gd->gd_time_seconds); 740 741 if (delta >= cputimer_freq) { 742 tvp->tv_sec += delta / cputimer_freq; 743 delta %= cputimer_freq; 744 } 745 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; 746 } 747 748 void 749 nanouptime(struct timespec *tsp) 750 { 751 struct globaldata *gd = mycpu; 752 sysclock_t delta; 753 754 do { 755 tsp->tv_sec = gd->gd_time_seconds; 756 delta = cputimer_count() - gd->gd_cpuclock_base; 757 } while (tsp->tv_sec != gd->gd_time_seconds); 758 759 if (delta >= cputimer_freq) { 760 tsp->tv_sec += delta / cputimer_freq; 761 delta %= cputimer_freq; 762 } 763 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; 764 } 765 766 /* 767 * realtime routines 768 */ 769 770 void 771 getmicrotime(struct timeval *tvp) 772 { 773 struct globaldata *gd = mycpu; 774 sysclock_t delta; 775 776 do { 777 tvp->tv_sec = gd->gd_time_seconds; 778 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; 779 } while (tvp->tv_sec != gd->gd_time_seconds); 780 781 if (delta >= cputimer_freq) { 782 tvp->tv_sec += delta / cputimer_freq; 783 delta %= cputimer_freq; 784 } 785 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; 786 787 tvp->tv_sec += basetime.tv_sec; 788 tvp->tv_usec += basetime.tv_nsec / 1000; 789 while (tvp->tv_usec >= 1000000) { 790 tvp->tv_usec -= 1000000; 791 ++tvp->tv_sec; 792 } 793 } 794 795 void 796 getnanotime(struct timespec *tsp) 797 { 798 struct globaldata *gd = mycpu; 799 sysclock_t delta; 800 801 do { 802 tsp->tv_sec = gd->gd_time_seconds; 803 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base; 804 } while (tsp->tv_sec != gd->gd_time_seconds); 805 806 if (delta >= cputimer_freq) { 807 tsp->tv_sec += delta / cputimer_freq; 808 delta %= cputimer_freq; 809 } 810 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; 811 812 tsp->tv_sec += basetime.tv_sec; 813 tsp->tv_nsec += basetime.tv_nsec; 814 while (tsp->tv_nsec >= 1000000000) { 815 tsp->tv_nsec -= 1000000000; 816 ++tsp->tv_sec; 817 } 818 } 819 820 void 821 microtime(struct timeval *tvp) 822 { 823 struct globaldata *gd = mycpu; 824 sysclock_t delta; 825 826 do { 827 tvp->tv_sec = gd->gd_time_seconds; 828 delta = cputimer_count() - gd->gd_cpuclock_base; 829 } while (tvp->tv_sec != gd->gd_time_seconds); 830 831 if (delta >= cputimer_freq) { 832 tvp->tv_sec += delta / cputimer_freq; 833 delta %= cputimer_freq; 834 } 835 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32; 836 837 tvp->tv_sec += basetime.tv_sec; 838 tvp->tv_usec += basetime.tv_nsec / 1000; 839 while (tvp->tv_usec >= 1000000) { 840 tvp->tv_usec -= 1000000; 841 ++tvp->tv_sec; 842 } 843 } 844 845 void 846 nanotime(struct timespec *tsp) 847 { 848 struct globaldata *gd = mycpu; 849 sysclock_t delta; 850 851 do { 852 tsp->tv_sec = gd->gd_time_seconds; 853 delta = cputimer_count() - gd->gd_cpuclock_base; 854 } while (tsp->tv_sec != gd->gd_time_seconds); 855 856 if (delta >= cputimer_freq) { 857 tsp->tv_sec += delta / cputimer_freq; 858 delta %= cputimer_freq; 859 } 860 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32; 861 862 tsp->tv_sec += basetime.tv_sec; 863 tsp->tv_nsec += basetime.tv_nsec; 864 while (tsp->tv_nsec >= 1000000000) { 865 tsp->tv_nsec -= 1000000000; 866 ++tsp->tv_sec; 867 } 868 } 869 870 int 871 pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps) 872 { 873 pps_params_t *app; 874 struct pps_fetch_args *fapi; 875 #ifdef PPS_SYNC 876 struct pps_kcbind_args *kapi; 877 #endif 878 879 switch (cmd) { 880 case PPS_IOC_CREATE: 881 return (0); 882 case PPS_IOC_DESTROY: 883 return (0); 884 case PPS_IOC_SETPARAMS: 885 app = (pps_params_t *)data; 886 if (app->mode & ~pps->ppscap) 887 return (EINVAL); 888 pps->ppsparam = *app; 889 return (0); 890 case PPS_IOC_GETPARAMS: 891 app = (pps_params_t *)data; 892 *app = pps->ppsparam; 893 app->api_version = PPS_API_VERS_1; 894 return (0); 895 case PPS_IOC_GETCAP: 896 *(int*)data = pps->ppscap; 897 return (0); 898 case PPS_IOC_FETCH: 899 fapi = (struct pps_fetch_args *)data; 900 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC) 901 return (EINVAL); 902 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec) 903 return (EOPNOTSUPP); 904 pps->ppsinfo.current_mode = pps->ppsparam.mode; 905 fapi->pps_info_buf = pps->ppsinfo; 906 return (0); 907 case PPS_IOC_KCBIND: 908 #ifdef PPS_SYNC 909 kapi = (struct pps_kcbind_args *)data; 910 /* XXX Only root should be able to do this */ 911 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC) 912 return (EINVAL); 913 if (kapi->kernel_consumer != PPS_KC_HARDPPS) 914 return (EINVAL); 915 if (kapi->edge & ~pps->ppscap) 916 return (EINVAL); 917 pps->kcmode = kapi->edge; 918 return (0); 919 #else 920 return (EOPNOTSUPP); 921 #endif 922 default: 923 return (ENOTTY); 924 } 925 } 926 927 void 928 pps_init(struct pps_state *pps) 929 { 930 pps->ppscap |= PPS_TSFMT_TSPEC; 931 if (pps->ppscap & PPS_CAPTUREASSERT) 932 pps->ppscap |= PPS_OFFSETASSERT; 933 if (pps->ppscap & PPS_CAPTURECLEAR) 934 pps->ppscap |= PPS_OFFSETCLEAR; 935 } 936 937 void 938 pps_event(struct pps_state *pps, sysclock_t count, int event) 939 { 940 struct globaldata *gd; 941 struct timespec *tsp; 942 struct timespec *osp; 943 struct timespec ts; 944 sysclock_t *pcount; 945 #ifdef PPS_SYNC 946 sysclock_t tcount; 947 #endif 948 sysclock_t delta; 949 pps_seq_t *pseq; 950 int foff; 951 int fhard; 952 953 gd = mycpu; 954 955 /* Things would be easier with arrays... */ 956 if (event == PPS_CAPTUREASSERT) { 957 tsp = &pps->ppsinfo.assert_timestamp; 958 osp = &pps->ppsparam.assert_offset; 959 foff = pps->ppsparam.mode & PPS_OFFSETASSERT; 960 fhard = pps->kcmode & PPS_CAPTUREASSERT; 961 pcount = &pps->ppscount[0]; 962 pseq = &pps->ppsinfo.assert_sequence; 963 } else { 964 tsp = &pps->ppsinfo.clear_timestamp; 965 osp = &pps->ppsparam.clear_offset; 966 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR; 967 fhard = pps->kcmode & PPS_CAPTURECLEAR; 968 pcount = &pps->ppscount[1]; 969 pseq = &pps->ppsinfo.clear_sequence; 970 } 971 972 /* Nothing really happened */ 973 if (*pcount == count) 974 return; 975 976 *pcount = count; 977 978 do { 979 ts.tv_sec = gd->gd_time_seconds; 980 delta = count - gd->gd_cpuclock_base; 981 } while (ts.tv_sec != gd->gd_time_seconds); 982 983 if (delta >= cputimer_freq) { 984 ts.tv_sec += delta / cputimer_freq; 985 delta %= cputimer_freq; 986 } 987 ts.tv_nsec = (cputimer_freq64_nsec * delta) >> 32; 988 ts.tv_sec += basetime.tv_sec; 989 ts.tv_nsec += basetime.tv_nsec; 990 while (ts.tv_nsec >= 1000000000) { 991 ts.tv_nsec -= 1000000000; 992 ++ts.tv_sec; 993 } 994 995 (*pseq)++; 996 *tsp = ts; 997 998 if (foff) { 999 timespecadd(tsp, osp); 1000 if (tsp->tv_nsec < 0) { 1001 tsp->tv_nsec += 1000000000; 1002 tsp->tv_sec -= 1; 1003 } 1004 } 1005 #ifdef PPS_SYNC 1006 if (fhard) { 1007 /* magic, at its best... */ 1008 tcount = count - pps->ppscount[2]; 1009 pps->ppscount[2] = count; 1010 if (tcount >= cputimer_freq) { 1011 delta = 1000000000 * (tcount / cputimer_freq) + 1012 (cputimer_freq64_nsec * 1013 (tcount % cputimer_freq)) >> 32; 1014 } else { 1015 delta = (cputimer_freq64_nsec * tcount) >> 32; 1016 } 1017 hardpps(tsp, delta); 1018 } 1019 #endif 1020 } 1021 1022