virtio_net: use correct accessors for scatterlists
[safe/jmp/linux-2.6] / kernel / time / ntp.c
index 3fc8106..f5f793d 100644 (file)
 
 #include <linux/mm.h>
 #include <linux/time.h>
-#include <linux/timer.h>
 #include <linux/timex.h>
 #include <linux/jiffies.h>
 #include <linux/hrtimer.h>
 #include <linux/capability.h>
 #include <linux/math64.h>
+#include <linux/clocksource.h>
+#include <linux/workqueue.h>
 #include <asm/timex.h>
 
 /*
  */
 unsigned long tick_usec = TICK_USEC;           /* USER_HZ period (usec) */
 unsigned long tick_nsec;                       /* ACTHZ period (nsec) */
-static u64 tick_length, tick_length_base;
+u64 tick_length;
+static u64 tick_length_base;
+
+static struct hrtimer leap_timer;
 
 #define MAX_TICKADJ            500             /* microsecs */
 #define MAX_TICKADJ_SCALED     (((u64)(MAX_TICKADJ * NSEC_PER_USEC) << \
-                                 TICK_LENGTH_SHIFT) / NTP_INTERVAL_FREQ)
+                                 NTP_SCALE_SHIFT) / NTP_INTERVAL_FREQ)
 
 /*
  * phase-lock loop variables
@@ -35,11 +39,12 @@ static u64 tick_length, tick_length_base;
 /* TIME_ERROR prevents overwriting the CMOS clock */
 static int time_state = TIME_OK;       /* clock synchronization status */
 int time_status = STA_UNSYNC;          /* clock status bits            */
+static long time_tai;                  /* TAI offset (s)               */
 static s64 time_offset;                        /* time adjustment (ns)         */
 static long time_constant = 2;         /* pll time constant            */
 long time_maxerror = NTP_PHASE_LIMIT;  /* maximum error (us)           */
 long time_esterror = NTP_PHASE_LIMIT;  /* estimated error (us)         */
-long time_freq;                                /* frequency offset (scaled ppm)*/
+static s64 time_freq;                  /* frequency offset (scaled ns/s)*/
 static long time_reftime;              /* time at last adjustment (s)  */
 long time_adjust;
 static long ntp_tick_adj;
@@ -47,13 +52,13 @@ static long ntp_tick_adj;
 static void ntp_update_frequency(void)
 {
        u64 second_length = (u64)(tick_usec * NSEC_PER_USEC * USER_HZ)
-                               << TICK_LENGTH_SHIFT;
-       second_length += (s64)ntp_tick_adj << TICK_LENGTH_SHIFT;
-       second_length += (s64)time_freq << (TICK_LENGTH_SHIFT - SHIFT_NSEC);
+                               << NTP_SCALE_SHIFT;
+       second_length += (s64)ntp_tick_adj << NTP_SCALE_SHIFT;
+       second_length += time_freq;
 
        tick_length_base = second_length;
 
-       tick_nsec = div_u64(second_length, HZ) >> TICK_LENGTH_SHIFT;
+       tick_nsec = div_u64(second_length, HZ) >> NTP_SCALE_SHIFT;
        tick_length_base = div_u64(tick_length_base, NTP_INTERVAL_FREQ);
 }
 
@@ -65,16 +70,15 @@ static void ntp_update_offset(long offset)
        if (!(time_status & STA_PLL))
                return;
 
-       time_offset = offset;
        if (!(time_status & STA_NANO))
-               time_offset *= NSEC_PER_USEC;
+               offset *= NSEC_PER_USEC;
 
        /*
         * Scale the phase adjustment and
         * clamp to the operating range.
         */
-       time_offset = min(time_offset, (s64)MAXPHASE * NSEC_PER_USEC);
-       time_offset = max(time_offset, (s64)-MAXPHASE * NSEC_PER_USEC);
+       offset = min(offset, MAXPHASE);
+       offset = max(offset, -MAXPHASE);
 
        /*
         * Select how the frequency is to be controlled
@@ -85,19 +89,19 @@ static void ntp_update_offset(long offset)
        mtemp = xtime.tv_sec - time_reftime;
        time_reftime = xtime.tv_sec;
 
-       freq_adj = time_offset * mtemp;
-       freq_adj = shift_right(freq_adj, time_constant * 2 +
-                          (SHIFT_PLL + 2) * 2 - SHIFT_NSEC);
+       freq_adj = (s64)offset * mtemp;
+       freq_adj <<= NTP_SCALE_SHIFT - 2 * (SHIFT_PLL + 2 + time_constant);
        time_status &= ~STA_MODE;
        if (mtemp >= MINSEC && (time_status & STA_FLL || mtemp > MAXSEC)) {
-               freq_adj += div_s64(time_offset << (SHIFT_NSEC - SHIFT_FLL), mtemp);
+               freq_adj += div_s64((s64)offset << (NTP_SCALE_SHIFT - SHIFT_FLL),
+                                   mtemp);
                time_status |= STA_MODE;
        }
        freq_adj += time_freq;
-       freq_adj = min(freq_adj, (s64)MAXFREQ_NSEC);
-       time_freq = max(freq_adj, (s64)-MAXFREQ_NSEC);
-       time_offset = div_s64(time_offset, NTP_INTERVAL_FREQ);
-       time_offset <<= SHIFT_UPDATE;
+       freq_adj = min(freq_adj, MAXFREQ_SCALED);
+       time_freq = max(freq_adj, -MAXFREQ_SCALED);
+
+       time_offset = div_s64((s64)offset << NTP_SCALE_SHIFT, NTP_INTERVAL_FREQ);
 }
 
 /**
@@ -119,62 +123,69 @@ void ntp_clear(void)
 }
 
 /*
- * this routine handles the overflow of the microsecond field
- *
- * The tricky bits of code to handle the accurate clock support
- * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
- * They were originally developed for SUN and DEC kernels.
- * All the kudos should go to Dave for this stuff.
+ * Leap second processing. If in leap-insert state at the end of the
+ * day, the system clock is set back one second; if in leap-delete
+ * state, the system clock is set ahead one second.
  */
-void second_overflow(void)
+static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
 {
-       long time_adj;
+       enum hrtimer_restart res = HRTIMER_NORESTART;
 
-       /* Bump the maxerror field */
-       time_maxerror += MAXFREQ >> SHIFT_USEC;
-       if (time_maxerror > NTP_PHASE_LIMIT) {
-               time_maxerror = NTP_PHASE_LIMIT;
-               time_status |= STA_UNSYNC;
-       }
+       write_seqlock(&xtime_lock);
 
-       /*
-        * Leap second processing. If in leap-insert state at the end of the
-        * day, the system clock is set back one second; if in leap-delete
-        * state, the system clock is set ahead one second. The microtime()
-        * routine or external clock driver will insure that reported time is
-        * always monotonic. The ugly divides should be replaced.
-        */
        switch (time_state) {
        case TIME_OK:
-               if (time_status & STA_INS)
-                       time_state = TIME_INS;
-               else if (time_status & STA_DEL)
-                       time_state = TIME_DEL;
                break;
        case TIME_INS:
-               if (xtime.tv_sec % 86400 == 0) {
-                       xtime.tv_sec--;
-                       wall_to_monotonic.tv_sec++;
-                       time_state = TIME_OOP;
-                       printk(KERN_NOTICE "Clock: inserting leap second "
-                                       "23:59:60 UTC\n");
-               }
+               xtime.tv_sec--;
+               wall_to_monotonic.tv_sec++;
+               time_state = TIME_OOP;
+               printk(KERN_NOTICE "Clock: "
+                      "inserting leap second 23:59:60 UTC\n");
+               hrtimer_add_expires_ns(&leap_timer, NSEC_PER_SEC);
+               res = HRTIMER_RESTART;
                break;
        case TIME_DEL:
-               if ((xtime.tv_sec + 1) % 86400 == 0) {
-                       xtime.tv_sec++;
-                       wall_to_monotonic.tv_sec--;
-                       time_state = TIME_WAIT;
-                       printk(KERN_NOTICE "Clock: deleting leap second "
-                                       "23:59:59 UTC\n");
-               }
+               xtime.tv_sec++;
+               time_tai--;
+               wall_to_monotonic.tv_sec--;
+               time_state = TIME_WAIT;
+               printk(KERN_NOTICE "Clock: "
+                      "deleting leap second 23:59:59 UTC\n");
                break;
        case TIME_OOP:
+               time_tai++;
                time_state = TIME_WAIT;
-               break;
+               /* fall through */
        case TIME_WAIT:
                if (!(time_status & (STA_INS | STA_DEL)))
                        time_state = TIME_OK;
+               break;
+       }
+       update_vsyscall(&xtime, clock);
+
+       write_sequnlock(&xtime_lock);
+
+       return res;
+}
+
+/*
+ * this routine handles the overflow of the microsecond field
+ *
+ * The tricky bits of code to handle the accurate clock support
+ * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
+ * They were originally developed for SUN and DEC kernels.
+ * All the kudos should go to Dave for this stuff.
+ */
+void second_overflow(void)
+{
+       s64 time_adj;
+
+       /* Bump the maxerror field */
+       time_maxerror += MAXFREQ / NSEC_PER_USEC;
+       if (time_maxerror > NTP_PHASE_LIMIT) {
+               time_maxerror = NTP_PHASE_LIMIT;
+               time_status |= STA_UNSYNC;
        }
 
        /*
@@ -184,7 +195,7 @@ void second_overflow(void)
        tick_length = tick_length_base;
        time_adj = shift_right(time_offset, SHIFT_PLL + time_constant);
        time_offset -= time_adj;
-       tick_length += (s64)time_adj << (TICK_LENGTH_SHIFT - SHIFT_UPDATE);
+       tick_length += time_adj;
 
        if (unlikely(time_adjust)) {
                if (time_adjust > MAX_TICKADJ) {
@@ -195,35 +206,22 @@ void second_overflow(void)
                        tick_length -= MAX_TICKADJ_SCALED;
                } else {
                        tick_length += (s64)(time_adjust * NSEC_PER_USEC /
-                                       NTP_INTERVAL_FREQ) << TICK_LENGTH_SHIFT;
+                                       NTP_INTERVAL_FREQ) << NTP_SCALE_SHIFT;
                        time_adjust = 0;
                }
        }
 }
 
-/*
- * Return how long ticks are at the moment, that is, how much time
- * update_wall_time_one_tick will add to xtime next time we call it
- * (assuming no calls to do_adjtimex in the meantime).
- * The return value is in fixed-point nanoseconds shifted by the
- * specified number of bits to the right of the binary point.
- * This function has no side-effects.
- */
-u64 current_tick_length(void)
-{
-       return tick_length;
-}
-
 #ifdef CONFIG_GENERIC_CMOS_UPDATE
 
 /* Disable the cmos update - used by virtualization and embedded */
 int no_sync_cmos_clock  __read_mostly;
 
-static void sync_cmos_clock(unsigned long dummy);
+static void sync_cmos_clock(struct work_struct *work);
 
-static DEFINE_TIMER(sync_cmos_timer, sync_cmos_clock, 0, 0);
+static DECLARE_DELAYED_WORK(sync_cmos_work, sync_cmos_clock);
 
-static void sync_cmos_clock(unsigned long dummy)
+static void sync_cmos_clock(struct work_struct *work)
 {
        struct timespec now, next;
        int fail = 1;
@@ -246,7 +244,7 @@ static void sync_cmos_clock(unsigned long dummy)
        if (abs(now.tv_nsec - (NSEC_PER_SEC / 2)) <= tick_nsec / 2)
                fail = update_persistent_clock(now);
 
-       next.tv_nsec = (NSEC_PER_SEC / 2) - now.tv_nsec;
+       next.tv_nsec = (NSEC_PER_SEC / 2) - now.tv_nsec - (TICK_NSEC / 2);
        if (next.tv_nsec <= 0)
                next.tv_nsec += NSEC_PER_SEC;
 
@@ -259,13 +257,13 @@ static void sync_cmos_clock(unsigned long dummy)
                next.tv_sec++;
                next.tv_nsec -= NSEC_PER_SEC;
        }
-       mod_timer(&sync_cmos_timer, jiffies + timespec_to_jiffies(&next));
+       schedule_delayed_work(&sync_cmos_work, timespec_to_jiffies(&next));
 }
 
 static void notify_cmos_timer(void)
 {
        if (!no_sync_cmos_clock)
-               mod_timer(&sync_cmos_timer, jiffies + 1);
+               schedule_delayed_work(&sync_cmos_work, 0);
 }
 
 #else
@@ -278,34 +276,50 @@ static inline void notify_cmos_timer(void) { }
 int do_adjtimex(struct timex *txc)
 {
        struct timespec ts;
-       long save_adjust;
        int result;
 
-       /* In order to modify anything, you gotta be super-user! */
-       if (txc->modes && !capable(CAP_SYS_TIME))
-               return -EPERM;
-
-       /* Now we validate the data before disabling interrupts */
-
-       if ((txc->modes & ADJ_OFFSET_SINGLESHOT) == ADJ_OFFSET_SINGLESHOT) {
+       /* Validate the data before disabling interrupts */
+       if (txc->modes & ADJ_ADJTIME) {
                /* singleshot must not be used with any other mode bits */
-               if (txc->modes & ~ADJ_OFFSET_SS_READ)
+               if (!(txc->modes & ADJ_OFFSET_SINGLESHOT))
                        return -EINVAL;
+               if (!(txc->modes & ADJ_OFFSET_READONLY) &&
+                   !capable(CAP_SYS_TIME))
+                       return -EPERM;
+       } else {
+               /* In order to modify anything, you gotta be super-user! */
+                if (txc->modes && !capable(CAP_SYS_TIME))
+                       return -EPERM;
+
+               /* if the quartz is off by more than 10% something is VERY wrong! */
+               if (txc->modes & ADJ_TICK &&
+                   (txc->tick <  900000/USER_HZ ||
+                    txc->tick > 1100000/USER_HZ))
+                               return -EINVAL;
+
+               if (txc->modes & ADJ_STATUS && time_state != TIME_OK)
+                       hrtimer_cancel(&leap_timer);
        }
 
-       /* if the quartz is off by more than 10% something is VERY wrong ! */
-       if (txc->modes & ADJ_TICK)
-               if (txc->tick <  900000/USER_HZ ||
-                   txc->tick > 1100000/USER_HZ)
-                       return -EINVAL;
+       getnstimeofday(&ts);
 
        write_seqlock_irq(&xtime_lock);
 
-       /* Save for later - semantics of adjtime is to return old value */
-       save_adjust = time_adjust;
-
        /* If there are input parameters, then process them */
+       if (txc->modes & ADJ_ADJTIME) {
+               long save_adjust = time_adjust;
+
+               if (!(txc->modes & ADJ_OFFSET_READONLY)) {
+                       /* adjtime() is independent from ntp_adjtime() */
+                       time_adjust = txc->offset;
+                       ntp_update_frequency();
+               }
+               txc->offset = save_adjust;
+               goto adj_done;
+       }
        if (txc->modes) {
+               long sec;
+
                if (txc->modes & ADJ_STATUS) {
                        if ((time_status & STA_PLL) &&
                            !(txc->status & STA_PLL)) {
@@ -315,6 +329,34 @@ int do_adjtimex(struct timex *txc)
                        /* only set allowed bits */
                        time_status &= STA_RONLY;
                        time_status |= txc->status & ~STA_RONLY;
+
+                       switch (time_state) {
+                       case TIME_OK:
+                       start_timer:
+                               sec = ts.tv_sec;
+                               if (time_status & STA_INS) {
+                                       time_state = TIME_INS;
+                                       sec += 86400 - sec % 86400;
+                                       hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
+                               } else if (time_status & STA_DEL) {
+                                       time_state = TIME_DEL;
+                                       sec += 86400 - (sec + 1) % 86400;
+                                       hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
+                               }
+                               break;
+                       case TIME_INS:
+                       case TIME_DEL:
+                               time_state = TIME_OK;
+                               goto start_timer;
+                               break;
+                       case TIME_WAIT:
+                               if (!(time_status & (STA_INS | STA_DEL)))
+                                       time_state = TIME_OK;
+                               break;
+                       case TIME_OOP:
+                               hrtimer_restart(&leap_timer);
+                               break;
+                       }
                }
 
                if (txc->modes & ADJ_NANO)
@@ -323,10 +365,9 @@ int do_adjtimex(struct timex *txc)
                        time_status &= ~STA_NANO;
 
                if (txc->modes & ADJ_FREQUENCY) {
-                       time_freq = min(txc->freq, MAXFREQ);
-                       time_freq = min(time_freq, -MAXFREQ);
-                       time_freq = ((s64)time_freq * NSEC_PER_USEC)
-                                       >> (SHIFT_USEC - SHIFT_NSEC);
+                       time_freq = (s64)txc->freq * PPM_SCALE;
+                       time_freq = min(time_freq, MAXFREQ_SCALED);
+                       time_freq = max(time_freq, -MAXFREQ_SCALED);
                }
 
                if (txc->modes & ADJ_MAXERROR)
@@ -342,13 +383,11 @@ int do_adjtimex(struct timex *txc)
                        time_constant = max(time_constant, 0l);
                }
 
-               if (txc->modes & ADJ_OFFSET) {
-                       if (txc->modes == ADJ_OFFSET_SINGLESHOT)
-                               /* adjtime() is independent from ntp_adjtime() */
-                               time_adjust = txc->offset;
-                       else
-                               ntp_update_offset(txc->offset);
-               }
+               if (txc->modes & ADJ_TAI && txc->constant > 0)
+                       time_tai = txc->constant;
+
+               if (txc->modes & ADJ_OFFSET)
+                       ntp_update_offset(txc->offset);
                if (txc->modes & ADJ_TICK)
                        tick_usec = txc->tick;
 
@@ -356,28 +395,26 @@ int do_adjtimex(struct timex *txc)
                        ntp_update_frequency();
        }
 
+       txc->offset = shift_right(time_offset * NTP_INTERVAL_FREQ,
+                                 NTP_SCALE_SHIFT);
+       if (!(time_status & STA_NANO))
+               txc->offset /= NSEC_PER_USEC;
+
+adj_done:
        result = time_state;    /* mostly `TIME_OK' */
        if (time_status & (STA_UNSYNC|STA_CLOCKERR))
                result = TIME_ERROR;
 
-       if ((txc->modes == ADJ_OFFSET_SINGLESHOT) ||
-           (txc->modes == ADJ_OFFSET_SS_READ))
-               txc->offset = save_adjust;
-       else {
-               txc->offset = ((long)shift_right(time_offset, SHIFT_UPDATE)) *
-                               NTP_INTERVAL_FREQ;
-               if (!(time_status & STA_NANO))
-                       txc->offset /= NSEC_PER_USEC;
-       }
-       txc->freq          = (time_freq / NSEC_PER_USEC) <<
-                               (SHIFT_USEC - SHIFT_NSEC);
+       txc->freq          = shift_right((time_freq >> PPM_SCALE_INV_SHIFT) *
+                                        (s64)PPM_SCALE_INV, NTP_SCALE_SHIFT);
        txc->maxerror      = time_maxerror;
        txc->esterror      = time_esterror;
        txc->status        = time_status;
        txc->constant      = time_constant;
        txc->precision     = 1;
-       txc->tolerance     = MAXFREQ;
+       txc->tolerance     = MAXFREQ_SCALED / PPM_SCALE;
        txc->tick          = tick_usec;
+       txc->tai           = time_tai;
 
        /* PPS is not implemented, so these are zero */
        txc->ppsfreq       = 0;
@@ -390,7 +427,6 @@ int do_adjtimex(struct timex *txc)
        txc->stbcnt        = 0;
        write_sequnlock_irq(&xtime_lock);
 
-       getnstimeofday(&ts);
        txc->time.tv_sec = ts.tv_sec;
        txc->time.tv_usec = ts.tv_nsec;
        if (!(time_status & STA_NANO))
@@ -408,3 +444,10 @@ static int __init ntp_tick_adj_setup(char *str)
 }
 
 __setup("ntp_tick_adj=", ntp_tick_adj_setup);
+
+void __init ntp_init(void)
+{
+       ntp_clear();
+       hrtimer_init(&leap_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
+       leap_timer.function = ntp_leap_second;
+}