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  */
  33
  34 #include "opt_ntp.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/sysproto.h>
  39 #include <sys/kernel.h>
  40 #include <sys/proc.h>
  41 #include <sys/priv.h>
  42 #include <sys/time.h>
  43 #include <sys/timex.h>
  44 #include <sys/timepps.h>
  45 #include <sys/sysctl.h>
  46
  47 #include <sys/thread2.h>
  48 #include <sys/mplock2.h>
  49
  50 /*
  51  * Single-precision macros for 64-bit machines
  52  */
  53 typedef long long l_fp;
  54 #define L_ADD(v, u)     ((v) += (u))
  55 #define L_SUB(v, u)     ((v) -= (u))
  56 #define L_ADDHI(v, a)   ((v) += (long long)(a) << 32)
  57 #define L_NEG(v)        ((v) = -(v))
  58 #define L_RSHIFT(v, n) \
  59         do { \
  60                 if ((v) < 0) \
  61                         (v) = -(-(v) >> (n)); \
  62                 else \
  63                         (v) = (v) >> (n); \
  64         } while (0)
  65 #define L_MPY(v, a)     ((v) *= (a))
  66 #define L_CLR(v)        ((v) = 0)
  67 #define L_ISNEG(v)      ((v) < 0)
  68 #define L_LINT(v, a)    ((v) = (long long)(a) << 32)
  69 #define L_GINT(v)       ((v) < 0 ? -(-(v) >> 32) : (v) >> 32)
  70
  71 /*
  72  * Generic NTP kernel interface
  73  *
  74  * These routines constitute the Network Time Protocol (NTP) interfaces
  75  * for user and daemon application programs. The ntp_gettime() routine
  76  * provides the time, maximum error (synch distance) and estimated error
  77  * (dispersion) to client user application programs. The ntp_adjtime()
  78  * routine is used by the NTP daemon to adjust the system clock to an
  79  * externally derived time. The time offset and related variables set by
  80  * this routine are used by other routines in this module to adjust the
  81  * phase and frequency of the clock discipline loop which controls the
  82  * system clock.
  83  *
  84  * When the kernel time is reckoned directly in nanoseconds (NTP_NANO
  85  * defined), the time at each tick interrupt is derived directly from
  86  * the kernel time variable. When the kernel time is reckoned in
  87  * microseconds, (NTP_NANO undefined), the time is derived from the
  88  * kernel time variable together with a variable representing the
  89  * leftover nanoseconds at the last tick interrupt. In either case, the
  90  * current nanosecond time is reckoned from these values plus an
  91  * interpolated value derived by the clock routines in another
  92  * architecture-specific module. The interpolation can use either a
  93  * dedicated counter or a processor cycle counter (PCC) implemented in
  94  * some architectures.
  95  *
  96  * Note that all routines must run at priority splclock or higher.
  97  */
  98 /*
  99  * Phase/frequency-lock loop (PLL/FLL) definitions
 100  *
 101  * The nanosecond clock discipline uses two variable types, time
 102  * variables and frequency variables. Both types are represented as 64-
 103  * bit fixed-point quantities with the decimal point between two 32-bit
 104  * halves. On a 32-bit machine, each half is represented as a single
 105  * word and mathematical operations are done using multiple-precision
 106  * arithmetic. On a 64-bit machine, ordinary computer arithmetic is
 107  * used.
 108  *
 109  * A time variable is a signed 64-bit fixed-point number in ns and
 110  * fraction. It represents the remaining time offset to be amortized
 111  * over succeeding tick interrupts. The maximum time offset is about
 112  * 0.5 s and the resolution is about 2.3e-10 ns.
 113  *
 114  *                      1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
 115  *  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
 116  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 117  * |s s s|                       ns                                |
 118  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 119  * |                        fraction                               |
 120  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 121  *
 122  * A frequency variable is a signed 64-bit fixed-point number in ns/s
 123  * and fraction. It represents the ns and fraction to be added to the
 124  * kernel time variable at each second. The maximum frequency offset is
 125  * about +-500000 ns/s and the resolution is about 2.3e-10 ns/s.
 126  *
 127  *                      1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
 128  *  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
 129  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 130  * |s s s s s s s s s s s s s|            ns/s                     |
 131  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 132  * |                        fraction                               |
 133  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 134  */
 135 /*
 136  * The following variables establish the state of the PLL/FLL and the
 137  * residual time and frequency offset of the local clock.
 138  */
 139 #define SHIFT_PLL       4               /* PLL loop gain (shift) */
 140 #define SHIFT_FLL       2               /* FLL loop gain (shift) */
 141
 142 static int time_state = TIME_OK;        /* clock state */
 143 static int time_status = STA_UNSYNC;    /* clock status bits */
 144 static long time_tai;                   /* TAI offset (s) */
 145 static long time_monitor;               /* last time offset scaled (ns) */
 146 static long time_constant;              /* poll interval (shift) (s) */
 147 static long time_precision = 1;         /* clock precision (ns) */
 148 static long time_maxerror = MAXPHASE / 1000; /* maximum error (us) */
 149 static long time_esterror = MAXPHASE / 1000; /* estimated error (us) */
 150 static time_t time_reftime;             /* time at last adjustment (s) */
 151 static long time_tick;                  /* nanoseconds per tick (ns) */
 152 static l_fp time_offset;                /* time offset (ns) */
 153 static l_fp time_freq;                  /* frequency offset (ns/s) */
 154 static l_fp time_adj;                   /* tick adjust (ns/s) */
 155
 156 #ifdef PPS_SYNC
 157 /*
 158  * The following variables are used when a pulse-per-second (PPS) signal
 159  * is available and connected via a modem control lead. They establish
 160  * the engineering parameters of the clock discipline loop when
 161  * controlled by the PPS signal.
 162  */
 163 #define PPS_FAVG        2               /* min freq avg interval (s) (shift) */
 164 #define PPS_FAVGDEF     8               /* default freq avg int (s) (shift) */
 165 #define PPS_FAVGMAX     15              /* max freq avg interval (s) (shift) */
 166 #define PPS_PAVG        4               /* phase avg interval (s) (shift) */
 167 #define PPS_VALID       120             /* PPS signal watchdog max (s) */
 168 #define PPS_MAXWANDER   100000          /* max PPS wander (ns/s) */
 169 #define PPS_POPCORN     2               /* popcorn spike threshold (shift) */
 170
 171 static struct timespec pps_tf[3];       /* phase median filter */
 172 static l_fp pps_freq;                   /* scaled frequency offset (ns/s) */
 173 static long pps_fcount;                 /* frequency accumulator */
 174 static long pps_jitter;                 /* nominal jitter (ns) */
 175 static long pps_stabil;                 /* nominal stability (scaled ns/s) */
 176 static long pps_lastsec;                /* time at last calibration (s) */
 177 static int pps_valid;                   /* signal watchdog counter */
 178 static int pps_shift = PPS_FAVG;        /* interval duration (s) (shift) */
 179 static int pps_shiftmax = PPS_FAVGDEF;  /* max interval duration (s) (shift) */
 180 static int pps_intcnt;                  /* wander counter */
 181
 182 /*
 183  * PPS signal quality monitors
 184  */
 185 static long pps_calcnt;                 /* calibration intervals */
 186 static long pps_jitcnt;                 /* jitter limit exceeded */
 187 static long pps_stbcnt;                 /* stability limit exceeded */
 188 static long pps_errcnt;                 /* calibration errors */
 189 #endif /* PPS_SYNC */
 190 /*
 191  * End of phase/frequency-lock loop (PLL/FLL) definitions
 192  */
 193
 194 static void ntp_init(void);
 195 static void hardupdate(long offset);
 196
 197 /*
 198  * ntp_gettime() - NTP user application interface
 199  *
 200  * See the timex.h header file for synopsis and API description. Note
 201  * that the TAI offset is returned in the ntvtimeval.tai structure
 202  * member.
 203  */
 204 static int
 205 ntp_sysctl(SYSCTL_HANDLER_ARGS)
 206 {
 207         struct ntptimeval ntv;  /* temporary structure */
 208         struct timespec atv;    /* nanosecond time */
 209
 210         nanotime(&atv);
 211         ntv.time.tv_sec = atv.tv_sec;
 212         ntv.time.tv_nsec = atv.tv_nsec;
 213         ntv.maxerror = time_maxerror;
 214         ntv.esterror = time_esterror;
 215         ntv.tai = time_tai;
 216         ntv.time_state = time_state;
 217
 218         /*
 219          * Status word error decode. If any of these conditions occur,
 220          * an error is returned, instead of the status word. Most
 221          * applications will care only about the fact the system clock
 222          * may not be trusted, not about the details.
 223          *
 224          * Hardware or software error
 225          */
 226         if ((time_status & (STA_UNSYNC | STA_CLOCKERR)) ||
 227
 228         /*
 229          * PPS signal lost when either time or frequency synchronization
 230          * requested
 231          */
 232             (time_status & (STA_PPSFREQ | STA_PPSTIME) &&
 233             !(time_status & STA_PPSSIGNAL)) ||
 234
 235         /*
 236          * PPS jitter exceeded when time synchronization requested
 237          */
 238             (time_status & STA_PPSTIME &&
 239             time_status & STA_PPSJITTER) ||
 240
 241         /*
 242          * PPS wander exceeded or calibration error when frequency
 243          * synchronization requested
 244          */
 245             (time_status & STA_PPSFREQ &&
 246             time_status & (STA_PPSWANDER | STA_PPSERROR)))
 247                 ntv.time_state = TIME_ERROR;
 248         return (sysctl_handle_opaque(oidp, &ntv, sizeof ntv, req));
 249 }
 250
 251 SYSCTL_NODE(_kern, OID_AUTO, ntp_pll, CTLFLAG_RW, 0, "");
 252 SYSCTL_PROC(_kern_ntp_pll, OID_AUTO, gettime, CTLTYPE_OPAQUE|CTLFLAG_RD,
 253         0, sizeof(struct ntptimeval) , ntp_sysctl, "S,ntptimeval", "");
 254
 255 #ifdef PPS_SYNC
 256 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, pps_shiftmax, CTLFLAG_RW, &pps_shiftmax, 0, "");
 257 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, pps_shift, CTLFLAG_RW, &pps_shift, 0, "");
 258 SYSCTL_INT(_kern_ntp_pll, OID_AUTO, time_monitor, CTLFLAG_RD, &time_monitor, 0, "");
 259
 260 SYSCTL_OPAQUE(_kern_ntp_pll, OID_AUTO, pps_freq, CTLFLAG_RD, &pps_freq, sizeof(pps_freq), "I", "");
 261 SYSCTL_OPAQUE(_kern_ntp_pll, OID_AUTO, time_freq, CTLFLAG_RD, &time_freq, sizeof(time_freq), "I", "");
 262 #endif
 263 /*
 264  * ntp_adjtime() - NTP daemon application interface
 265  *
 266  * See the timex.h header file for synopsis and API description. Note
 267  * that the timex.constant structure member has a dual purpose to set
 268  * the time constant and to set the TAI offset.
 269  *
 270  * MPALMOSTSAFE
 271  */
 272 int
 273 sys_ntp_adjtime(struct ntp_adjtime_args *uap)
 274 {
 275         struct thread *td = curthread;
 276         struct timex ntv;       /* temporary structure */
 277         long freq;              /* frequency ns/s) */
 278         int modes;              /* mode bits from structure */
 279         int error;
 280
 281         error = copyin((caddr_t)uap->tp, (caddr_t)&ntv, sizeof(ntv));
 282         if (error)
 283                 return(error);
 284
 285         /*
 286          * Update selected clock variables - only the superuser can
 287          * change anything. Note that there is no error checking here on
 288          * the assumption the superuser should know what it is doing.
 289          * Note that either the time constant or TAI offset are loaded
 290          * from the ntv.constant member, depending on the mode bits. If
 291          * the STA_PLL bit in the status word is cleared, the state and
 292          * status words are reset to the initial values at boot.
 293          */
 294         modes = ntv.modes;
 295         if (modes)
 296                 error = priv_check(td, PRIV_NTP_ADJTIME);
 297         if (error)
 298                 return (error);
 299
 300         get_mplock();
 301         crit_enter();
 302         if (modes & MOD_MAXERROR)
 303                 time_maxerror = ntv.maxerror;
 304         if (modes & MOD_ESTERROR)
 305                 time_esterror = ntv.esterror;
 306         if (modes & MOD_STATUS) {
 307                 if (time_status & STA_PLL && !(ntv.status & STA_PLL)) {
 308                         time_state = TIME_OK;
 309                         time_status = STA_UNSYNC;
 310 #ifdef PPS_SYNC
 311                         pps_shift = PPS_FAVG;
 312 #endif /* PPS_SYNC */
 313                 }
 314                 time_status &= STA_RONLY;
 315                 time_status |= ntv.status & ~STA_RONLY;
 316         }
 317         if (modes & MOD_TIMECONST) {
 318                 if (ntv.constant < 0)
 319                         time_constant = 0;
 320                 else if (ntv.constant > MAXTC)
 321                         time_constant = MAXTC;
 322                 else
 323                         time_constant = ntv.constant;
 324         }
 325         if (modes & MOD_TAI) {
 326                 if (ntv.constant > 0) /* XXX zero & negative numbers ? */
 327                         time_tai = ntv.constant;
 328         }
 329 #ifdef PPS_SYNC
 330         if (modes & MOD_PPSMAX) {
 331                 if (ntv.shift < PPS_FAVG)
 332                         pps_shiftmax = PPS_FAVG;
 333                 else if (ntv.shift > PPS_FAVGMAX)
 334                         pps_shiftmax = PPS_FAVGMAX;
 335                 else
 336                         pps_shiftmax = ntv.shift;
 337         }
 338 #endif /* PPS_SYNC */
 339         if (modes & MOD_NANO)
 340                 time_status |= STA_NANO;
 341         if (modes & MOD_MICRO)
 342                 time_status &= ~STA_NANO;
 343         if (modes & MOD_CLKB)
 344                 time_status |= STA_CLK;
 345         if (modes & MOD_CLKA)
 346                 time_status &= ~STA_CLK;
 347         if (modes & MOD_OFFSET) {
 348                 if (time_status & STA_NANO)
 349                         hardupdate(ntv.offset);
 350                 else
 351                         hardupdate(ntv.offset * 1000);
 352         }
 353         /*
 354          * Note: the userland specified frequency is in seconds per second
 355          * times 65536e+6.  Multiply by a thousand and divide by 65336 to
 356          * get nanoseconds.
 357          */
 358         if (modes & MOD_FREQUENCY) {
 359                 freq = (ntv.freq * 1000LL) >> 16;
 360                 if (freq > MAXFREQ)
 361                         L_LINT(time_freq, MAXFREQ);
 362                 else if (freq < -MAXFREQ)
 363                         L_LINT(time_freq, -MAXFREQ);
 364                 else
 365                         L_LINT(time_freq, freq);
 366 #ifdef PPS_SYNC
 367                 pps_freq = time_freq;
 368 #endif /* PPS_SYNC */
 369         }
 370
 371         /*
 372          * Retrieve all clock variables. Note that the TAI offset is
 373          * returned only by ntp_gettime();
 374          */
 375         if (time_status & STA_NANO)
 376                 ntv.offset = time_monitor;
 377         else
 378                 ntv.offset = time_monitor / 1000; /* XXX rounding ? */
 379         ntv.freq = L_GINT((time_freq / 1000LL) << 16);
 380         ntv.maxerror = time_maxerror;
 381         ntv.esterror = time_esterror;
 382         ntv.status = time_status;
 383         ntv.constant = time_constant;
 384         if (time_status & STA_NANO)
 385                 ntv.precision = time_precision;
 386         else
 387                 ntv.precision = time_precision / 1000;
 388         ntv.tolerance = MAXFREQ * SCALE_PPM;
 389 #ifdef PPS_SYNC
 390         ntv.shift = pps_shift;
 391         ntv.ppsfreq = L_GINT((pps_freq / 1000LL) << 16);
 392         if (time_status & STA_NANO)
 393                 ntv.jitter = pps_jitter;
 394         else
 395                 ntv.jitter = pps_jitter / 1000;
 396         ntv.stabil = pps_stabil;
 397         ntv.calcnt = pps_calcnt;
 398         ntv.errcnt = pps_errcnt;
 399         ntv.jitcnt = pps_jitcnt;
 400         ntv.stbcnt = pps_stbcnt;
 401 #endif /* PPS_SYNC */
 402         crit_exit();
 403         rel_mplock();
 404
 405         error = copyout((caddr_t)&ntv, (caddr_t)uap->tp, sizeof(ntv));
 406         if (error)
 407                 return (error);
 408
 409         /*
 410          * Status word error decode. See comments in
 411          * ntp_gettime() routine.
 412          */
 413         if ((time_status & (STA_UNSYNC | STA_CLOCKERR)) ||
 414             (time_status & (STA_PPSFREQ | STA_PPSTIME) &&
 415             !(time_status & STA_PPSSIGNAL)) ||
 416             (time_status & STA_PPSTIME &&
 417             time_status & STA_PPSJITTER) ||
 418             (time_status & STA_PPSFREQ &&
 419             time_status & (STA_PPSWANDER | STA_PPSERROR))) {
 420                 uap->sysmsg_result = TIME_ERROR;
 421         } else {
 422                 uap->sysmsg_result = time_state;
 423         }
 424         return (error);
 425 }
 426
 427 /*
 428  * second_overflow() - called after ntp_tick_adjust()
 429  *
 430  * This routine is ordinarily called from hardclock() whenever the seconds
 431  * hand rolls over.  It returns leap seconds to add or drop, and sets nsec_adj
 432  * to the total adjustment to make over the next second in (ns << 32).
 433  *
 434  * This routine is only called by cpu #0.
 435  */
 436 int
 437 ntp_update_second(time_t newsec, int64_t *nsec_adj)
 438 {
 439         l_fp ftemp;             /* 32/64-bit temporary */
 440         int  adjsec = 0;
 441
 442         /*
 443          * On rollover of the second both the nanosecond and microsecond
 444          * clocks are updated and the state machine cranked as
 445          * necessary. The phase adjustment to be used for the next
 446          * second is calculated and the maximum error is increased by
 447          * the tolerance.
 448          */
 449         time_maxerror += MAXFREQ / 1000;
 450
 451         /*
 452          * Leap second processing. If in leap-insert state at
 453          * the end of the day, the system clock is set back one
 454          * second; if in leap-delete state, the system clock is
 455          * set ahead one second. The nano_time() routine or
 456          * external clock driver will insure that reported time
 457          * is always monotonic.
 458          */
 459         switch (time_state) {
 460
 461                 /*
 462                  * No warning.
 463                  */
 464                 case TIME_OK:
 465                 if (time_status & STA_INS)
 466                         time_state = TIME_INS;
 467                 else if (time_status & STA_DEL)
 468                         time_state = TIME_DEL;
 469                 break;
 470
 471                 /*
 472                  * Insert second 23:59:60 following second
 473                  * 23:59:59.
 474                  */
 475                 case TIME_INS:
 476                 if (!(time_status & STA_INS))
 477                         time_state = TIME_OK;
 478                 else if ((newsec) % 86400 == 0) {
 479                         --adjsec;
 480                         time_state = TIME_OOP;
 481                 }
 482                 break;
 483
 484                 /*
 485                  * Delete second 23:59:59.
 486                  */
 487                 case TIME_DEL:
 488                 if (!(time_status & STA_DEL))
 489                         time_state = TIME_OK;
 490                 else if (((newsec) + 1) % 86400 == 0) {
 491                         ++adjsec;
 492                         time_tai--;
 493                         time_state = TIME_WAIT;
 494                 }
 495                 break;
 496
 497                 /*
 498                  * Insert second in progress.
 499                  */
 500                 case TIME_OOP:
 501                         time_tai++;
 502                         time_state = TIME_WAIT;
 503                 break;
 504
 505                 /*
 506                  * Wait for status bits to clear.
 507                  */
 508                 case TIME_WAIT:
 509                 if (!(time_status & (STA_INS | STA_DEL)))
 510                         time_state = TIME_OK;
 511         }
 512
 513         /*
 514          * time_offset represents the total time adjustment we wish to
 515          * make (over no particular period of time).  time_freq represents
 516          * the frequency compensation we wish to apply.
 517          *
 518          * time_adj represents the total adjustment we wish to make over
 519          * one full second.  hardclock usually applies this adjustment in
 520          * time_adj / hz jumps, hz times a second.
 521          */
 522         ftemp = time_offset;
 523 #ifdef PPS_SYNC
 524         /* XXX even if PPS signal dies we should finish adjustment ? */
 525         if ((time_status & STA_PPSTIME) && (time_status & STA_PPSSIGNAL))
 526                 L_RSHIFT(ftemp, pps_shift);
 527         else
 528                 L_RSHIFT(ftemp, SHIFT_PLL + time_constant);
 529 #else
 530                 L_RSHIFT(ftemp, SHIFT_PLL + time_constant);
 531 #endif /* PPS_SYNC */
 532         time_adj = ftemp;               /* adjustment for part of the offset */
 533         L_SUB(time_offset, ftemp);
 534         L_ADD(time_adj, time_freq);     /* add frequency correction */
 535         *nsec_adj = time_adj;
 536 #ifdef PPS_SYNC
 537         if (pps_valid > 0)
 538                 pps_valid--;
 539         else
 540                 time_status &= ~STA_PPSSIGNAL;
 541 #endif /* PPS_SYNC */
 542         return(adjsec);
 543 }
 544
 545 /*
 546  * ntp_init() - initialize variables and structures
 547  *
 548  * This routine must be called after the kernel variables hz and tick
 549  * are set or changed and before the next tick interrupt. In this
 550  * particular implementation, these values are assumed set elsewhere in
 551  * the kernel. The design allows the clock frequency and tick interval
 552  * to be changed while the system is running. So, this routine should
 553  * probably be integrated with the code that does that.
 554  */
 555 static void
 556 ntp_init(void)
 557 {
 558
 559         /*
 560          * The following variable must be initialized any time the
 561          * kernel variable hz is changed.
 562          */
 563         time_tick = NANOSECOND / hz;
 564
 565         /*
 566          * The following variables are initialized only at startup. Only
 567          * those structures not cleared by the compiler need to be
 568          * initialized, and these only in the simulator. In the actual
 569          * kernel, any nonzero values here will quickly evaporate.
 570          */
 571         L_CLR(time_offset);
 572         L_CLR(time_freq);
 573 #ifdef PPS_SYNC
 574         pps_tf[0].tv_sec = pps_tf[0].tv_nsec = 0;
 575         pps_tf[1].tv_sec = pps_tf[1].tv_nsec = 0;
 576         pps_tf[2].tv_sec = pps_tf[2].tv_nsec = 0;
 577         pps_fcount = 0;
 578         L_CLR(pps_freq);
 579 #endif /* PPS_SYNC */
 580 }
 581
 582 SYSINIT(ntpclocks, SI_BOOT2_CLOCKS, SI_ORDER_FIRST, ntp_init, NULL);
 583
 584 /*
 585  * hardupdate() - local clock update
 586  *
 587  * This routine is called by ntp_adjtime() to update the local clock
 588  * phase and frequency. The implementation is of an adaptive-parameter,
 589  * hybrid phase/frequency-lock loop (PLL/FLL). The routine computes new
 590  * time and frequency offset estimates for each call. If the kernel PPS
 591  * discipline code is configured (PPS_SYNC), the PPS signal itself
 592  * determines the new time offset, instead of the calling argument.
 593  * Presumably, calls to ntp_adjtime() occur only when the caller
 594  * believes the local clock is valid within some bound (+-128 ms with
 595  * NTP). If the caller's time is far different than the PPS time, an
 596  * argument will ensue, and it's not clear who will lose.
 597  *
 598  * For uncompensated quartz crystal oscillators and nominal update
 599  * intervals less than 256 s, operation should be in phase-lock mode,
 600  * where the loop is disciplined to phase. For update intervals greater
 601  * than 1024 s, operation should be in frequency-lock mode, where the
 602  * loop is disciplined to frequency. Between 256 s and 1024 s, the mode
 603  * is selected by the STA_MODE status bit.
 604  */
 605 static void
 606 hardupdate(long offset)
 607 {
 608         long mtemp;
 609         l_fp ftemp;
 610
 611         /*
 612          * Select how the phase is to be controlled and from which
 613          * source. If the PPS signal is present and enabled to
 614          * discipline the time, the PPS offset is used; otherwise, the
 615          * argument offset is used.
 616          */
 617         if (!(time_status & STA_PLL))
 618                 return;
 619         if (!((time_status & STA_PPSTIME) && (time_status & STA_PPSSIGNAL))) {
 620                 if (offset > MAXPHASE)
 621                         time_monitor = MAXPHASE;
 622                 else if (offset < -MAXPHASE)
 623                         time_monitor = -MAXPHASE;
 624                 else
 625                         time_monitor = offset;
 626                 L_LINT(time_offset, time_monitor);
 627         }
 628
 629         /*
 630          * Select how the frequency is to be controlled and in which
 631          * mode (PLL or FLL). If the PPS signal is present and enabled
 632          * to discipline the frequency, the PPS frequency is used;
 633          * otherwise, the argument offset is used to compute it.
 634          */
 635         if ((time_status & STA_PPSFREQ) && time_status & STA_PPSSIGNAL) {
 636                 time_reftime = time_uptime;
 637                 return;
 638         }
 639         if ((time_status & STA_FREQHOLD) || time_reftime == 0)
 640                 time_reftime = time_uptime;
 641         mtemp = time_uptime - 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_uptime;
 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 */