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