sys/kern/kern_ntptime.c

   1 /***********************************************************************
   2  *                                                                     *
   3  * Copyright (c) David L. Mills 1993-2001                              *
   4  *                                                                     *
   5  * Permission to use, copy, modify, and distribute this software and   *
   6  * its documentation for any purpose and without fee is hereby         *
   7  * granted, provided that the above copyright notice appears in all    *
   8  * copies and that both the copyright notice and this permission       *
   9  * notice appear in supporting documentation, and that the name        *
  10  * University of Delaware not be used in advertising or publicity      *
  11  * pertaining to distribution of the software without specific,        *
  12  * written prior permission. The University of Delaware makes no       *
  13  * representations about the suitability this software for any         *
  14  * purpose. It is provided "as is" without express or implied          *
  15  * warranty.                                                           *
  16  *                                                                     *
  17  **********************************************************************/
  18
  19 /*
  20  * Adapted from the original sources for FreeBSD and timecounters by:
  21  * Poul-Henning Kamp <phk@FreeBSD.org>.
  22  *
  23  * The 32bit version of the "LP" macros seems a bit past its "sell by"
  24  * date so I have retained only the 64bit version and included it directly
  25  * in this file.
  26  *
  27  * Only minor changes done to interface with the timecounters over in
  28  * sys/kern/kern_clock.c.   Some of the comments below may be (even more)
  29  * confusing and/or plain wrong in that context.
  30  *
  31  * $FreeBSD: src/sys/kern/kern_ntptime.c,v 1.32.2.2 2001/04/22 11:19:46 jhay Exp $
  32  * $DragonFly: src/sys/kern/kern_ntptime.c,v 1.13 2007/04/30 07:18:53 dillon Exp $
  33  */
  34
  35 #include "opt_ntp.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/sysproto.h>
  40 #include <sys/kernel.h>
  41 #include <sys/proc.h>
  42 #include <sys/priv.h>
  43 #include <sys/time.h>
  44 #include <sys/timex.h>
  45 #include <sys/timepps.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/thread2.h>
  48
  49 /*
  50  * Single-precision macros for 64-bit machines
  51  */
  52 typedef long long l_fp;
  53 #define L_ADD(v, u)     ((v) += (u))
  54 #define L_SUB(v, u)     ((v) -= (u))
  55 #define L_ADDHI(v, a)   ((v) += (long long)(a) << 32)
  56 #define L_NEG(v)        ((v) = -(v))
  57 #define L_RSHIFT(v, n) \
  58         do { \
  59                 if ((v) < 0) \
  60                         (v) = -(-(v) >> (n)); \
  61                 else \
  62                         (v) = (v) >> (n); \
  63         } while (0)
  64 #define L_MPY(v, a)     ((v) *= (a))
  65 #define L_CLR(v)        ((v) = 0)
  66 #define L_ISNEG(v)      ((v) < 0)
  67 #define L_LINT(v, a)    ((v) = (long long)(a) << 32)
  68 #define L_GINT(v)       ((v) < 0 ? -(-(v) >> 32) : (v) >> 32)
  69
  70 /*
  71  * Generic NTP kernel interface
  72  *
  73  * These routines constitute the Network Time Protocol (NTP) interfaces
  74  * for user and daemon application programs. The ntp_gettime() routine
  75  * provides the time, maximum error (synch distance) and estimated error
  76  * (dispersion) to client user application programs. The ntp_adjtime()
  77  * routine is used by the NTP daemon to adjust the system clock to an
  78  * externally derived time. The time offset and related variables set by
  79  * this routine are used by other routines in this module to adjust the
  80  * phase and frequency of the clock discipline loop which controls the
  81  * system clock.
  82  *
  83  * When the kernel time is reckoned directly in nanoseconds (NTP_NANO
  84  * defined), the time at each tick interrupt is derived directly from
  85  * the kernel time variable. When the kernel time is reckoned in
  86  * microseconds, (NTP_NANO undefined), the time is derived from the
  87  * kernel time variable together with a variable representing the
  88  * leftover nanoseconds at the last tick interrupt. In either case, the
  89  * current nanosecond time is reckoned from these values plus an
  90  * interpolated value derived by the clock routines in another
  91  * architecture-specific module. The interpolation can use either a
  92  * dedicated counter or a processor cycle counter (PCC) implemented in
  93  * some architectures.
  94  *
  95  * Note that all routines must run at priority splclock or higher.
  96  */
  97 /*
  98  * Phase/frequency-lock loop (PLL/FLL) definitions
  99  *
 100  * The nanosecond clock discipline uses two variable types, time
 101  * variables and frequency variables. Both types are represented as 64-
 102  * bit fixed-point quantities with the decimal point between two 32-bit
 103  * halves. On a 32-bit machine, each half is represented as a single
 104  * word and mathematical operations are done using multiple-precision
 105  * arithmetic. On a 64-bit machine, ordinary computer arithmetic is
 106  * used.
 107  *
 108  * A time variable is a signed 64-bit fixed-point number in ns and
 109  * fraction. It represents the remaining time offset to be amortized
 110  * over succeeding tick interrupts. The maximum time offset is about
 111  * 0.5 s and the resolution is about 2.3e-10 ns.
 112  *
 113  *                      1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
 114  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
 115  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 116  * |s s s|                       ns                                |
 117  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 118  * |                        fraction                               |
 119  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 120  *
 121  * A frequency variable is a signed 64-bit fixed-point number in ns/s
 122  * and fraction. It represents the ns and fraction to be added to the
 123  * kernel time variable at each second. The maximum frequency offset is
 124  * about +-500000 ns/s and the resolution is about 2.3e-10 ns/s.
 125  *
 126  *                      1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
 127  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
 128  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 129  * |s s s s s s s s s s s s s|            ns/s                     |
 130  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 131  * |                        fraction                               |
 132  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 133  */
 134 /*
 135  * The following variables establish the state of the PLL/FLL and the
 136  * residual time and frequency offset of the local clock.
 137  */
 138 #define SHIFT_PLL       4               /* PLL loop gain (shift) */
 139 #define SHIFT_FLL       2               /* FLL loop gain (shift) */
 140
 141 static int time_state = TIME_OK;        /* clock state */
 142 static int time_status = STA_UNSYNC;    /* clock status bits */
 143 static long time_tai;                   /* TAI offset (s) */
 144 static long time_monitor;               /* last time offset scaled (ns) */
 145 static long time_constant;              /* poll interval (shift) (s) */
 146 static long time_precision = 1;         /* clock precision (ns) */
 147 static long time_maxerror = MAXPHASE / 1000; /* maximum error (us) */
 148 static long time_esterror = MAXPHASE / 1000; /* estimated error (us) */
 149 static long time_reftime;               /* time at last adjustment (s) */
 150 static long time_tick;                  /* nanoseconds per tick (ns) */
 151 static l_fp time_offset;                /* time offset (ns) */
 152 static l_fp time_freq;                  /* frequency offset (ns/s) */
 153 static l_fp time_adj;                   /* tick adjust (ns/s) */
 154
 155 #ifdef PPS_SYNC
 156 /*
 157  * The following variables are used when a pulse-per-second (PPS) signal
 158  * is available and connected via a modem control lead. They establish
 159  * the engineering parameters of the clock discipline loop when
 160  * controlled by the PPS signal.
 161  */
 162 #define PPS_FAVG        2               /* min freq avg interval (s) (shift) */
 163 #define PPS_FAVGDEF     8               /* default freq avg int (s) (shift) */
 164 #define PPS_FAVGMAX     15              /* max freq avg interval (s) (shift) */
 165 #define PPS_PAVG        4               /* phase avg interval (s) (shift) */
 166 #define PPS_VALID       120             /* PPS signal watchdog max (s) */
 167 #define PPS_MAXWANDER   100000          /* max PPS wander (ns/s) */
 168 #define PPS_POPCORN     2               /* popcorn spike threshold (shift) */
 169
 170 static struct timespec pps_tf[3];       /* phase median filter */
 171 static l_fp pps_freq;                   /* scaled frequency offset (ns/s) */
 172 static long pps_fcount;                 /* frequency accumulator */
 173 static long pps_jitter;                 /* nominal jitter (ns) */
 174 static long pps_stabil;                 /* nominal stability (scaled ns/s) */
 175 static long pps_lastsec;                /* time at last calibration (s) */
 176 static int pps_valid;                   /* signal watchdog counter */
 177 static int pps_shift = PPS_FAVG;        /* interval duration (s) (shift) */
 178 static int pps_shiftmax = PPS_FAVGDEF;  /* max interval duration (s) (shift) */
 179 static int pps_intcnt;                  /* wander counter */
 180
 181 /*
 182  * PPS signal quality monitors
 183  */
 184 static long pps_calcnt;                 /* calibration intervals */
 185 static long pps_jitcnt;                 /* jitter limit exceeded */
 186 static long pps_stbcnt;                 /* stability limit exceeded */
 187 static long pps_errcnt;                 /* calibration errors */
 188 #endif /* PPS_SYNC */
 189 /*
 190  * End of phase/frequency-lock loop (PLL/FLL) definitions
 191  */
 192
 193 static void ntp_init(void);
 194 static void hardupdate(long offset);
 195
 196 /*
 197  * ntp_gettime() - NTP user application interface
 198  *
 199  * See the timex.h header file for synopsis and API description. Note
 200  * that the TAI offset is returned in the ntvtimeval.tai structure
 201  * member.
 202  */
 203 static int
 204 ntp_sysctl(SYSCTL_HANDLER_ARGS)
 205 {
 206         struct ntptimeval ntv;  /* temporary structure */
 207         struct timespec atv;    /* nanosecond time */
 208
 209         nanotime(&atv);
 210         ntv.time.tv_sec = atv.tv_sec;
 211         ntv.time.tv_nsec = atv.tv_nsec;
 212         ntv.maxerror = time_maxerror;
 213         ntv.esterror = time_esterror;
 214         ntv.tai = time_tai;
 215         ntv.time_state = time_state;
 216
 217         /*
 218          * Status word error decode. If any of these conditions occur,
 219          * an error is returned, instead of the status word. Most
 220          * applications will care only about the fact the system clock
 221          * may not be trusted, not about the details.
 222          *
 223          * Hardware or software error
 224          */
 225         if ((time_status & (STA_UNSYNC | STA_CLOCKERR)) ||
 226
 227         /*
 228          * PPS signal lost when either time or frequency synchronization
 229          * requested
 230          */
 231             (time_status & (STA_PPSFREQ | STA_PPSTIME) &&
 232             !(time_status & STA_PPSSIGNAL)) ||
 233
 234         /*
 235          * PPS jitter exceeded when time synchronization requested
 236          */
 237             (time_status & STA_PPSTIME &&
 238             time_status & STA_PPSJITTER) ||
 239
 240         /*
 241          * PPS wander exceeded or calibration error when frequency
 242          * synchronization requested
 243          */
 244             (time_status & STA_PPSFREQ &&
 245             time_status & (STA_PPSWANDER | STA_PPSERROR)))
 246                 ntv.time_state = TIME_ERROR;
 247         return (sysctl_handle_opaque(oidp, &ntv, sizeof ntv, req));
 248 }
 249
 250 SYSCTL_NODE(_kern, OID_AUTO, ntp_pll, CTLFLAG_RW, 0, "");
 251 SYSCTL_PROC(_kern_ntp_pll, OID_AUTO, gettime, CTLTYPE_OPAQUE|CTLFLAG_RD,
 252         0, sizeof(struct ntptimeval) , ntp_sysctl, "S,ntptimeval", "");
 253
 254 #ifdef PPS_SYNC
 255 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, pps_shiftmax, CTLFLAG_RW, &pps_shiftmax, 0, "");
 256 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, pps_shift, CTLFLAG_RW, &pps_shift, 0, "");
 257 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, time_monitor, CTLFLAG_RD, &time_monitor, 0, "");
 258
 259 SYSCTL_OPAQUE(_kern_ntp_pll, OID_AUTO, pps_freq, CTLFLAG_RD, &pps_freq, sizeof(pps_freq), "I", "");
 260 SYSCTL_OPAQUE(_kern_ntp_pll, OID_AUTO, time_freq, CTLFLAG_RD, &time_freq, sizeof(time_freq), "I", "");
 261 #endif
 262 /*
 263  * ntp_adjtime() - NTP daemon application interface
 264  *
 265  * See the timex.h header file for synopsis and API description. Note
 266  * that the timex.constant structure member has a dual purpose to set
 267  * the time constant and to set the TAI offset.
 268  */
 269 int
 270 sys_ntp_adjtime(struct ntp_adjtime_args *uap)
 271 {
 272         struct thread *td = curthread;
 273         struct timex ntv;       /* temporary structure */
 274         long freq;              /* frequency ns/s) */
 275         int modes;              /* mode bits from structure */
 276         int error;
 277
 278         error = copyin((caddr_t)uap->tp, (caddr_t)&ntv, sizeof(ntv));
 279         if (error)
 280                 return(error);
 281
 282         /*
 283          * Update selected clock variables - only the superuser can
 284          * change anything. Note that there is no error checking here on
 285          * the assumption the superuser should know what it is doing.
 286          * Note that either the time constant or TAI offset are loaded
 287          * from the ntv.constant member, depending on the mode bits. If
 288          * the STA_PLL bit in the status word is cleared, the state and
 289          * status words are reset to the initial values at boot.
 290          */
 291         modes = ntv.modes;
 292         if (modes)
 293                 error = priv_check(td, PRIV_NTP_ADJTIME);
 294         if (error)
 295                 return (error);
 296         crit_enter();
 297         if (modes & MOD_MAXERROR)
 298                 time_maxerror = ntv.maxerror;
 299         if (modes & MOD_ESTERROR)
 300                 time_esterror = ntv.esterror;
 301         if (modes & MOD_STATUS) {
 302                 if (time_status & STA_PLL && !(ntv.status & STA_PLL)) {
 303                         time_state = TIME_OK;
 304                         time_status = STA_UNSYNC;
 305 #ifdef PPS_SYNC
 306                         pps_shift = PPS_FAVG;
 307 #endif /* PPS_SYNC */
 308                 }
 309                 time_status &= STA_RONLY;
 310                 time_status |= ntv.status & ~STA_RONLY;
 311         }
 312         if (modes & MOD_TIMECONST) {
 313                 if (ntv.constant < 0)
 314                         time_constant = 0;
 315                 else if (ntv.constant > MAXTC)
 316                         time_constant = MAXTC;
 317                 else
 318                         time_constant = ntv.constant;
 319         }
 320         if (modes & MOD_TAI) {
 321                 if (ntv.constant > 0) /* XXX zero & negative numbers ? */
 322                         time_tai = ntv.constant;
 323         }
 324 #ifdef PPS_SYNC
 325         if (modes & MOD_PPSMAX) {
 326                 if (ntv.shift < PPS_FAVG)
 327                         pps_shiftmax = PPS_FAVG;
 328                 else if (ntv.shift > PPS_FAVGMAX)
 329                         pps_shiftmax = PPS_FAVGMAX;
 330                 else
 331                         pps_shiftmax = ntv.shift;
 332         }
 333 #endif /* PPS_SYNC */
 334         if (modes & MOD_NANO)
 335                 time_status |= STA_NANO;
 336         if (modes & MOD_MICRO)
 337                 time_status &= ~STA_NANO;
 338         if (modes & MOD_CLKB)
 339                 time_status |= STA_CLK;
 340         if (modes & MOD_CLKA)
 341                 time_status &= ~STA_CLK;
 342         if (modes & MOD_OFFSET) {
 343                 if (time_status & STA_NANO)
 344                         hardupdate(ntv.offset);
 345                 else
 346                         hardupdate(ntv.offset * 1000);
 347         }
 348         /*
 349          * Note: the userland specified frequency is in seconds per second
 350          * times 65536e+6.  Multiply by a thousand and divide by 65336 to
 351          * get nanoseconds.
 352          */
 353         if (modes & MOD_FREQUENCY) {
 354                 freq = (ntv.freq * 1000LL) >> 16;
 355                 if (freq > MAXFREQ)
 356                         L_LINT(time_freq, MAXFREQ);
 357                 else if (freq < -MAXFREQ)
 358                         L_LINT(time_freq, -MAXFREQ);
 359                 else
 360                         L_LINT(time_freq, freq);
 361 #ifdef PPS_SYNC
 362                 pps_freq = time_freq;
 363 #endif /* PPS_SYNC */
 364         }
 365
 366         /*
 367          * Retrieve all clock variables. Note that the TAI offset is
 368          * returned only by ntp_gettime();
 369          */
 370         if (time_status & STA_NANO)
 371                 ntv.offset = time_monitor;
 372         else
 373                 ntv.offset = time_monitor / 1000; /* XXX rounding ? */
 374         ntv.freq = L_GINT((time_freq / 1000LL) << 16);
 375         ntv.maxerror = time_maxerror;
 376         ntv.esterror = time_esterror;
 377         ntv.status = time_status;
 378         ntv.constant = time_constant;
 379         if (time_status & STA_NANO)
 380                 ntv.precision = time_precision;
 381         else
 382                 ntv.precision = time_precision / 1000;
 383         ntv.tolerance = MAXFREQ * SCALE_PPM;
 384 #ifdef PPS_SYNC
 385         ntv.shift = pps_shift;
 386         ntv.ppsfreq = L_GINT((pps_freq / 1000LL) << 16);
 387         if (time_status & STA_NANO)
 388                 ntv.jitter = pps_jitter;
 389         else
 390                 ntv.jitter = pps_jitter / 1000;
 391         ntv.stabil = pps_stabil;
 392         ntv.calcnt = pps_calcnt;
 393         ntv.errcnt = pps_errcnt;
 394         ntv.jitcnt = pps_jitcnt;
 395         ntv.stbcnt = pps_stbcnt;
 396 #endif /* PPS_SYNC */
 397         crit_exit();
 398
 399         error = copyout((caddr_t)&ntv, (caddr_t)uap->tp, sizeof(ntv));
 400         if (error)
 401                 return (error);
 402
 403         /*
 404          * Status word error decode. See comments in
 405          * ntp_gettime() routine.
 406          */
 407         if ((time_status & (STA_UNSYNC | STA_CLOCKERR)) ||
 408             (time_status & (STA_PPSFREQ | STA_PPSTIME) &&
 409             !(time_status & STA_PPSSIGNAL)) ||
 410             (time_status & STA_PPSTIME &&
 411             time_status & STA_PPSJITTER) ||
 412             (time_status & STA_PPSFREQ &&
 413             time_status & (STA_PPSWANDER | STA_PPSERROR))) {
 414                 uap->sysmsg_result = TIME_ERROR;
 415         } else {
 416                 uap->sysmsg_result = time_state;
 417         }
 418         return (error);
 419 }
 420
 421 /*
 422  * second_overflow() - called after ntp_tick_adjust()
 423  *
 424  * This routine is ordinarily called from hardclock() whenever the seconds
 425  * hand rolls over.  It returns leap seconds to add or drop, and sets nsec_adj
 426  * to the total adjustment to make over the next second in (ns << 32).
 427  *
 428  * This routine is only called by cpu #0.
 429  */
 430 int
 431 ntp_update_second(time_t newsec, int64_t *nsec_adj)
 432 {
 433         l_fp ftemp;             /* 32/64-bit temporary */
 434         int  adjsec = 0;
 435
 436         /*
 437          * On rollover of the second both the nanosecond and microsecond
 438          * clocks are updated and the state machine cranked as
 439          * necessary. The phase adjustment to be used for the next
 440          * second is calculated and the maximum error is increased by
 441          * the tolerance.
 442          */
 443         time_maxerror += MAXFREQ / 1000;
 444
 445         /*
 446          * Leap second processing. If in leap-insert state at
 447          * the end of the day, the system clock is set back one
 448          * second; if in leap-delete state, the system clock is
 449          * set ahead one second. The nano_time() routine or
 450          * external clock driver will insure that reported time
 451          * is always monotonic.
 452          */
 453         switch (time_state) {
 454
 455                 /*
 456                  * No warning.
 457                  */
 458                 case TIME_OK:
 459                 if (time_status & STA_INS)
 460                         time_state = TIME_INS;
 461                 else if (time_status & STA_DEL)
 462                         time_state = TIME_DEL;
 463                 break;
 464
 465                 /*
 466                  * Insert second 23:59:60 following second
 467                  * 23:59:59.
 468                  */
 469                 case TIME_INS:
 470                 if (!(time_status & STA_INS))
 471                         time_state = TIME_OK;
 472                 else if ((newsec) % 86400 == 0) {
 473                         --adjsec;
 474                         time_state = TIME_OOP;
 475                 }
 476                 break;
 477
 478                 /*
 479                  * Delete second 23:59:59.
 480                  */
 481                 case TIME_DEL:
 482                 if (!(time_status & STA_DEL))
 483                         time_state = TIME_OK;
 484                 else if (((newsec) + 1) % 86400 == 0) {
 485                         ++adjsec;
 486                         time_tai--;
 487                         time_state = TIME_WAIT;
 488                 }
 489                 break;
 490
 491                 /*
 492                  * Insert second in progress.
 493                  */
 494                 case TIME_OOP:
 495                         time_tai++;
 496                         time_state = TIME_WAIT;
 497                 break;
 498
 499                 /*
 500                  * Wait for status bits to clear.
 501                  */
 502                 case TIME_WAIT:
 503                 if (!(time_status & (STA_INS | STA_DEL)))
 504                         time_state = TIME_OK;
 505         }
 506
 507         /*
 508          * time_offset represents the total time adjustment we wish to
 509          * make (over no particular period of time).  time_freq represents
 510          * the frequency compensation we wish to apply.
 511          *
 512          * time_adj represents the total adjustment we wish to make over
 513          * one full second.  hardclock usually applies this adjustment in
 514          * time_adj / hz jumps, hz times a second.
 515          */
 516         ftemp = time_offset;
 517 #ifdef PPS_SYNC
 518         /* XXX even if PPS signal dies we should finish adjustment ? */
 519         if ((time_status & STA_PPSTIME) && (time_status & STA_PPSSIGNAL))
 520                 L_RSHIFT(ftemp, pps_shift);
 521         else
 522                 L_RSHIFT(ftemp, SHIFT_PLL + time_constant);
 523 #else
 524                 L_RSHIFT(ftemp, SHIFT_PLL + time_constant);
 525 #endif /* PPS_SYNC */
 526         time_adj = ftemp;               /* adjustment for part of the offset */
 527         L_SUB(time_offset, ftemp);
 528         L_ADD(time_adj, time_freq);     /* add frequency correction */
 529         *nsec_adj = time_adj;
 530 #ifdef PPS_SYNC
 531         if (pps_valid > 0)
 532                 pps_valid--;
 533         else
 534                 time_status &= ~STA_PPSSIGNAL;
 535 #endif /* PPS_SYNC */
 536         return(adjsec);
 537 }
 538
 539 /*
 540  * ntp_init() - initialize variables and structures
 541  *
 542  * This routine must be called after the kernel variables hz and tick
 543  * are set or changed and before the next tick interrupt. In this
 544  * particular implementation, these values are assumed set elsewhere in
 545  * the kernel. The design allows the clock frequency and tick interval
 546  * to be changed while the system is running. So, this routine should
 547  * probably be integrated with the code that does that.
 548  */
 549 static void
 550 ntp_init(void)
 551 {
 552
 553         /*
 554          * The following variable must be initialized any time the
 555          * kernel variable hz is changed.
 556          */
 557         time_tick = NANOSECOND / hz;
 558
 559         /*
 560          * The following variables are initialized only at startup. Only
 561          * those structures not cleared by the compiler need to be
 562          * initialized, and these only in the simulator. In the actual
 563          * kernel, any nonzero values here will quickly evaporate.
 564          */
 565         L_CLR(time_offset);
 566         L_CLR(time_freq);
 567 #ifdef PPS_SYNC
 568         pps_tf[0].tv_sec = pps_tf[0].tv_nsec = 0;
 569         pps_tf[1].tv_sec = pps_tf[1].tv_nsec = 0;
 570         pps_tf[2].tv_sec = pps_tf[2].tv_nsec = 0;
 571         pps_fcount = 0;
 572         L_CLR(pps_freq);
 573 #endif /* PPS_SYNC */
 574 }
 575
 576 SYSINIT(ntpclocks, SI_BOOT2_CLOCKS, SI_ORDER_FIRST, ntp_init, NULL)
 577
 578 /*
 579  * hardupdate() - local clock update
 580  *
 581  * This routine is called by ntp_adjtime() to update the local clock
 582  * phase and frequency. The implementation is of an adaptive-parameter,
 583  * hybrid phase/frequency-lock loop (PLL/FLL). The routine computes new
 584  * time and frequency offset estimates for each call. If the kernel PPS
 585  * discipline code is configured (PPS_SYNC), the PPS signal itself
 586  * determines the new time offset, instead of the calling argument.
 587  * Presumably, calls to ntp_adjtime() occur only when the caller
 588  * believes the local clock is valid within some bound (+-128 ms with
 589  * NTP). If the caller's time is far different than the PPS time, an
 590  * argument will ensue, and it's not clear who will lose.
 591  *
 592  * For uncompensated quartz crystal oscillators and nominal update
 593  * intervals less than 256 s, operation should be in phase-lock mode,
 594  * where the loop is disciplined to phase. For update intervals greater
 595  * than 1024 s, operation should be in frequency-lock mode, where the
 596  * loop is disciplined to frequency. Between 256 s and 1024 s, the mode
 597  * is selected by the STA_MODE status bit.
 598  */
 599 static void
 600 hardupdate(long offset)
 601 {
 602         long mtemp;
 603         l_fp ftemp;
 604         globaldata_t gd;
 605
 606         gd = mycpu;
 607
 608         /*
 609          * Select how the phase is to be controlled and from which
 610          * source. If the PPS signal is present and enabled to
 611          * discipline the time, the PPS offset is used; otherwise, the
 612          * argument offset is used.
 613          */
 614         if (!(time_status & STA_PLL))
 615                 return;
 616         if (!((time_status & STA_PPSTIME) && (time_status & STA_PPSSIGNAL))) {
 617                 if (offset > MAXPHASE)
 618                         time_monitor = MAXPHASE;
 619                 else if (offset < -MAXPHASE)
 620                         time_monitor = -MAXPHASE;
 621                 else
 622                         time_monitor = offset;
 623                 L_LINT(time_offset, time_monitor);
 624         }
 625
 626         /*
 627          * Select how the frequency is to be controlled and in which
 628          * mode (PLL or FLL). If the PPS signal is present and enabled
 629          * to discipline the frequency, the PPS frequency is used;
 630          * otherwise, the argument offset is used to compute it.
 631          *
 632          * gd_time_seconds is basically an uncompensated uptime.  We use
 633          * this for consistency.
 634          */
 635         if (time_status & STA_PPSFREQ && time_status & STA_PPSSIGNAL) {
 636                 time_reftime = time_second;
 637                 return;
 638         }
 639         if (time_status & STA_FREQHOLD || time_reftime == 0)
 640                 time_reftime = time_second;
 641         mtemp = time_second - time_reftime;
 642         L_LINT(ftemp, time_monitor);
 643         L_RSHIFT(ftemp, (SHIFT_PLL + 2 + time_constant) << 1);
 644         L_MPY(ftemp, mtemp);
 645         L_ADD(time_freq, ftemp);
 646         time_status &= ~STA_MODE;
 647         if (mtemp >= MINSEC && (time_status & STA_FLL || mtemp > MAXSEC)) {
 648                 L_LINT(ftemp, (time_monitor << 4) / mtemp);
 649                 L_RSHIFT(ftemp, SHIFT_FLL + 4);
 650                 L_ADD(time_freq, ftemp);
 651                 time_status |= STA_MODE;
 652         }
 653         time_reftime = time_second;
 654         if (L_GINT(time_freq) > MAXFREQ)
 655                 L_LINT(time_freq, MAXFREQ);
 656         else if (L_GINT(time_freq) < -MAXFREQ)
 657                 L_LINT(time_freq, -MAXFREQ);
 658 }
 659
 660 #ifdef PPS_SYNC
 661 /*
 662  * hardpps() - discipline CPU clock oscillator to external PPS signal
 663  *
 664  * This routine is called at each PPS interrupt in order to discipline
 665  * the CPU clock oscillator to the PPS signal. There are two independent
 666  * first-order feedback loops, one for the phase, the other for the
 667  * frequency. The phase loop measures and grooms the PPS phase offset
 668  * and leaves it in a handy spot for the seconds overflow routine. The
 669  * frequency loop averages successive PPS phase differences and
 670  * calculates the PPS frequency offset, which is also processed by the
 671  * seconds overflow routine. The code requires the caller to capture the
 672  * time and architecture-dependent hardware counter values in
 673  * nanoseconds at the on-time PPS signal transition.
 674  *
 675  * Note that, on some Unix systems this routine runs at an interrupt
 676  * priority level higher than the timer interrupt routine hardclock().
 677  * Therefore, the variables used are distinct from the hardclock()
 678  * variables, except for the actual time and frequency variables, which
 679  * are determined by this routine and updated atomically.
 680  */
 681 void
 682 hardpps(struct timespec *tsp, long nsec)
 683 {
 684         long u_sec, u_nsec, v_nsec; /* temps */
 685         l_fp ftemp;
 686
 687         /*
 688          * The signal is first processed by a range gate and frequency
 689          * discriminator. The range gate rejects noise spikes outside
 690          * the range +-500 us. The frequency discriminator rejects input
 691          * signals with apparent frequency outside the range 1 +-500
 692          * PPM. If two hits occur in the same second, we ignore the
 693          * later hit; if not and a hit occurs outside the range gate,
 694          * keep the later hit for later comparison, but do not process
 695          * it.
 696          */
 697         time_status |= STA_PPSSIGNAL | STA_PPSJITTER;
 698         time_status &= ~(STA_PPSWANDER | STA_PPSERROR);
 699         pps_valid = PPS_VALID;
 700         u_sec = tsp->tv_sec;
 701         u_nsec = tsp->tv_nsec;
 702         if (u_nsec >= (NANOSECOND >> 1)) {
 703                 u_nsec -= NANOSECOND;
 704                 u_sec++;
 705         }
 706         v_nsec = u_nsec - pps_tf[0].tv_nsec;
 707         if (u_sec == pps_tf[0].tv_sec && v_nsec < NANOSECOND -
 708             MAXFREQ)
 709                 return;
 710         pps_tf[2] = pps_tf[1];
 711         pps_tf[1] = pps_tf[0];
 712         pps_tf[0].tv_sec = u_sec;
 713         pps_tf[0].tv_nsec = u_nsec;
 714
 715         /*
 716          * Compute the difference between the current and previous
 717          * counter values. If the difference exceeds 0.5 s, assume it
 718          * has wrapped around, so correct 1.0 s. If the result exceeds
 719          * the tick interval, the sample point has crossed a tick
 720          * boundary during the last second, so correct the tick. Very
 721          * intricate.
 722          */
 723         u_nsec = nsec;
 724         if (u_nsec > (NANOSECOND >> 1))
 725                 u_nsec -= NANOSECOND;
 726         else if (u_nsec < -(NANOSECOND >> 1))
 727                 u_nsec += NANOSECOND;
 728         pps_fcount += u_nsec;
 729         if (v_nsec > MAXFREQ || v_nsec < -MAXFREQ)
 730                 return;
 731         time_status &= ~STA_PPSJITTER;
 732
 733         /*
 734          * A three-stage median filter is used to help denoise the PPS
 735          * time. The median sample becomes the time offset estimate; the
 736          * difference between the other two samples becomes the time
 737          * dispersion (jitter) estimate.
 738          */
 739         if (pps_tf[0].tv_nsec > pps_tf[1].tv_nsec) {
 740                 if (pps_tf[1].tv_nsec > pps_tf[2].tv_nsec) {
 741                         v_nsec = pps_tf[1].tv_nsec;     /* 0 1 2 */
 742                         u_nsec = pps_tf[0].tv_nsec - pps_tf[2].tv_nsec;
 743                 } else if (pps_tf[2].tv_nsec > pps_tf[0].tv_nsec) {
 744                         v_nsec = pps_tf[0].tv_nsec;     /* 2 0 1 */
 745                         u_nsec = pps_tf[2].tv_nsec - pps_tf[1].tv_nsec;
 746                 } else {
 747                         v_nsec = pps_tf[2].tv_nsec;     /* 0 2 1 */
 748                         u_nsec = pps_tf[0].tv_nsec - pps_tf[1].tv_nsec;
 749                 }
 750         } else {
 751                 if (pps_tf[1].tv_nsec < pps_tf[2].tv_nsec) {
 752                         v_nsec = pps_tf[1].tv_nsec;     /* 2 1 0 */
 753                         u_nsec = pps_tf[2].tv_nsec - pps_tf[0].tv_nsec;
 754                 } else if (pps_tf[2].tv_nsec < pps_tf[0].tv_nsec) {
 755                         v_nsec = pps_tf[0].tv_nsec;     /* 1 0 2 */
 756                         u_nsec = pps_tf[1].tv_nsec - pps_tf[2].tv_nsec;
 757                 } else {
 758                         v_nsec = pps_tf[2].tv_nsec;     /* 1 2 0 */
 759                         u_nsec = pps_tf[1].tv_nsec - pps_tf[0].tv_nsec;
 760                 }
 761         }
 762
 763         /*
 764          * Nominal jitter is due to PPS signal noise and interrupt
 765          * latency. If it exceeds the popcorn threshold, the sample is
 766          * discarded. otherwise, if so enabled, the time offset is
 767          * updated. We can tolerate a modest loss of data here without
 768          * much degrading time accuracy.
 769          */
 770         if (u_nsec > (pps_jitter << PPS_POPCORN)) {
 771                 time_status |= STA_PPSJITTER;
 772                 pps_jitcnt++;
 773         } else if (time_status & STA_PPSTIME) {
 774                 time_monitor = -v_nsec;
 775                 L_LINT(time_offset, time_monitor);
 776         }
 777         pps_jitter += (u_nsec - pps_jitter) >> PPS_FAVG;
 778         u_sec = pps_tf[0].tv_sec - pps_lastsec;
 779         if (u_sec < (1 << pps_shift))
 780                 return;
 781
 782         /*
 783          * At the end of the calibration interval the difference between
 784          * the first and last counter values becomes the scaled
 785          * frequency. It will later be divided by the length of the
 786          * interval to determine the frequency update. If the frequency
 787          * exceeds a sanity threshold, or if the actual calibration
 788          * interval is not equal to the expected length, the data are
 789          * discarded. We can tolerate a modest loss of data here without
 790          * much degrading frequency accuracy.
 791          */
 792         pps_calcnt++;
 793         v_nsec = -pps_fcount;
 794         pps_lastsec = pps_tf[0].tv_sec;
 795         pps_fcount = 0;
 796         u_nsec = MAXFREQ << pps_shift;
 797         if (v_nsec > u_nsec || v_nsec < -u_nsec || u_sec != (1 <<
 798             pps_shift)) {
 799                 time_status |= STA_PPSERROR;
 800                 pps_errcnt++;
 801                 return;
 802         }
 803
 804         /*
 805          * Here the raw frequency offset and wander (stability) is
 806          * calculated. If the wander is less than the wander threshold
 807          * for four consecutive averaging intervals, the interval is
 808          * doubled; if it is greater than the threshold for four
 809          * consecutive intervals, the interval is halved. The scaled
 810          * frequency offset is converted to frequency offset. The
 811          * stability metric is calculated as the average of recent
 812          * frequency changes, but is used only for performance
 813          * monitoring.
 814          */
 815         L_LINT(ftemp, v_nsec);
 816         L_RSHIFT(ftemp, pps_shift);
 817         L_SUB(ftemp, pps_freq);
 818         u_nsec = L_GINT(ftemp);
 819         if (u_nsec > PPS_MAXWANDER) {
 820                 L_LINT(ftemp, PPS_MAXWANDER);
 821                 pps_intcnt--;
 822                 time_status |= STA_PPSWANDER;
 823                 pps_stbcnt++;
 824         } else if (u_nsec < -PPS_MAXWANDER) {
 825                 L_LINT(ftemp, -PPS_MAXWANDER);
 826                 pps_intcnt--;
 827                 time_status |= STA_PPSWANDER;
 828                 pps_stbcnt++;
 829         } else {
 830                 pps_intcnt++;
 831         }
 832         if (pps_intcnt >= 4) {
 833                 pps_intcnt = 4;
 834                 if (pps_shift < pps_shiftmax) {
 835                         pps_shift++;
 836                         pps_intcnt = 0;
 837                 }
 838         } else if (pps_intcnt <= -4 || pps_shift > pps_shiftmax) {
 839                 pps_intcnt = -4;
 840                 if (pps_shift > PPS_FAVG) {
 841                         pps_shift--;
 842                         pps_intcnt = 0;
 843                 }
 844         }
 845         if (u_nsec < 0)
 846                 u_nsec = -u_nsec;
 847         pps_stabil += (u_nsec * SCALE_PPM - pps_stabil) >> PPS_FAVG;
 848
 849         /*
 850          * The PPS frequency is recalculated and clamped to the maximum
 851          * MAXFREQ. If enabled, the system clock frequency is updated as
 852          * well.
 853          */
 854         L_ADD(pps_freq, ftemp);
 855         u_nsec = L_GINT(pps_freq);
 856         if (u_nsec > MAXFREQ)
 857                 L_LINT(pps_freq, MAXFREQ);
 858         else if (u_nsec < -MAXFREQ)
 859                 L_LINT(pps_freq, -MAXFREQ);
 860         if (time_status & STA_PPSFREQ)
 861                 time_freq = pps_freq;
 862 }
 863 #endif /* PPS_SYNC */