Re: [PATCH] fix and optimize clock source update

From: Roman Zippel
Date: Thu Jun 22 2006 - 07:18:27 EST


Hi,

On Wed, 21 Jun 2006, Andrew Morton wrote:

> > The alternative is duplicating the code and an inline function which takes
> > the address of these variables would likely generate worse code.
>
> Can you verify that please? It is pretty sick.

I'm pleasantly surprised. I already expected to be no difference with
gcc-4.0, but gcc-3.x can deal with it as well.
Below is a new patch which moves the macro into the bigadjust function and
includes John's changes, while at it I also added a few more comments.

bye, Roman




This fixes the clock source updates in update_wall_time() to correctly
track the time coming in via current_tick_length(). Optimize the fast
paths to be as short as possible to keep the overhead low.

Signed-off-by: Roman Zippel <zippel@xxxxxxxxxxxxxx>

---

Andrew, please apply this patch, it's essential so that we can finally
get the clock source stuff merged. It keeps the overhead added by all
this in a not really good but acceptable range.
Tested with 2.6.17-rc6-mm2, applies and compiles with 2.6.17-mm1

arch/powerpc/kernel/time.c | 4 -
include/linux/clocksource.h | 113 ++------------------------------
include/linux/timex.h | 4 -
kernel/timer.c | 151 +++++++++++++++++++++++++++++++-------------
4 files changed, 123 insertions(+), 149 deletions(-)

Index: linux-2.6-mm/arch/powerpc/kernel/time.c
===================================================================
--- linux-2.6-mm.orig/arch/powerpc/kernel/time.c 2006-06-22 11:56:06.000000000 +0200
+++ linux-2.6-mm/arch/powerpc/kernel/time.c 2006-06-22 11:56:24.000000000 +0200
@@ -102,7 +102,7 @@ EXPORT_SYMBOL(tb_ticks_per_sec); /* for
u64 tb_to_xs;
unsigned tb_to_us;

-#define TICKLEN_SCALE (SHIFT_SCALE - 10)
+#define TICKLEN_SCALE TICK_LENGTH_SHIFT
u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */
u64 ticklen_to_xs; /* 0.64 fraction */

@@ -534,7 +534,7 @@ static __inline__ void timer_recalc_offs

if (__USE_RTC())
return;
- tlen = current_tick_length(SHIFT_SCALE - 10);
+ tlen = current_tick_length();
offset = cur_tb - do_gtod.varp->tb_orig_stamp;
if (tlen == last_tick_len && offset < 0x80000000u)
return;
Index: linux-2.6-mm/include/linux/clocksource.h
===================================================================
--- linux-2.6-mm.orig/include/linux/clocksource.h 2006-06-22 11:56:06.000000000 +0200
+++ linux-2.6-mm/include/linux/clocksource.h 2006-06-22 11:56:24.000000000 +0200
@@ -46,8 +46,8 @@ typedef u64 cycle_t;
* @shift: cycle to nanosecond divisor (power of two)
* @update_callback: called when safe to alter clocksource values
* @is_continuous: defines if clocksource is free-running.
- * @interval_cycles: Used internally by timekeeping core, please ignore.
- * @interval_snsecs: Used internally by timekeeping core, please ignore.
+ * @cycle_interval: Used internally by timekeeping core, please ignore.
+ * @xtime_interval: Used internally by timekeeping core, please ignore.
*/
struct clocksource {
char *name;
@@ -61,8 +61,9 @@ struct clocksource {
int is_continuous;

/* timekeeping specific data, ignore */
- cycle_t interval_cycles;
- u64 interval_snsecs;
+ cycle_t cycle_last, cycle_interval;
+ u64 xtime_nsec, xtime_interval;
+ s64 error;
};

/* simplify initialization of mask field */
@@ -168,107 +169,11 @@ static inline void clocksource_calculate
tmp += c->mult/2;
do_div(tmp, c->mult);

- c->interval_cycles = (cycle_t)tmp;
- if(c->interval_cycles == 0)
- c->interval_cycles = 1;
+ c->cycle_interval = (cycle_t)tmp;
+ if (c->cycle_interval == 0)
+ c->cycle_interval = 1;

- c->interval_snsecs = (u64)c->interval_cycles * c->mult;
-}
-
-
-/**
- * error_aproximation - calculates an error adjustment for a given error
- *
- * @error: Error value (unsigned)
- * @unit: Adjustment unit
- *
- * For a given error value, this function takes the adjustment unit
- * and uses binary approximation to return a power of two adjustment value.
- *
- * This function is only for use by the the make_ntp_adj() function
- * and you must hold a write on the xtime_lock when calling.
- */
-static inline int error_aproximation(u64 error, u64 unit)
-{
- static int saved_adj = 0;
- u64 adjusted_unit = unit << saved_adj;
-
- if (error > (adjusted_unit * 2)) {
- /* large error, so increment the adjustment factor */
- saved_adj++;
- } else if (error > adjusted_unit) {
- /* just right, don't touch it */
- } else if (saved_adj) {
- /* small error, so drop the adjustment factor */
- saved_adj--;
- return 0;
- }
-
- return saved_adj;
-}
-
-
-/**
- * make_ntp_adj - Adjusts the specified clocksource for a given error
- *
- * @clock: Pointer to clock to be adjusted
- * @cycles_delta: Current unacounted cycle delta
- * @error: Pointer to current error value
- *
- * Returns clock shifted nanosecond adjustment to be applied against
- * the accumulated time value (ie: xtime).
- *
- * If the error value is large enough, this function calulates the
- * (power of two) adjustment value, and adjusts the clock's mult and
- * interval_snsecs values accordingly.
- *
- * However, since there may be some unaccumulated cycles, to avoid
- * time inconsistencies we must adjust the accumulation value
- * accordingly.
- *
- * This is not very intuitive, so the following proof should help:
- * The basic timeofday algorithm: base + cycle * mult
- * Thus:
- * new_base + cycle * new_mult = old_base + cycle * old_mult
- * new_base = old_base + cycle * old_mult - cycle * new_mult
- * new_base = old_base + cycle * (old_mult - new_mult)
- * new_base - old_base = cycle * (old_mult - new_mult)
- * base_delta = cycle * (old_mult - new_mult)
- * base_delta = cycle * (mult_delta)
- *
- * Where mult_delta is the adjustment value made to mult
- *
- */
-static inline s64 make_ntp_adj(struct clocksource *clock,
- cycles_t cycles_delta, s64* error)
-{
- s64 ret = 0;
- if (*error > ((s64)clock->interval_cycles+1)/2) {
- /* calculate adjustment value */
- int adjustment = error_aproximation(*error,
- clock->interval_cycles);
- /* adjust clock */
- clock->mult += 1 << adjustment;
- clock->interval_snsecs += clock->interval_cycles << adjustment;
-
- /* adjust the base and error for the adjustment */
- ret = -(cycles_delta << adjustment);
- *error -= clock->interval_cycles << adjustment;
- /* XXX adj error for cycle_delta offset? */
- } else if ((-(*error)) > ((s64)clock->interval_cycles+1)/2) {
- /* calculate adjustment value */
- int adjustment = error_aproximation(-(*error),
- clock->interval_cycles);
- /* adjust clock */
- clock->mult -= 1 << adjustment;
- clock->interval_snsecs -= clock->interval_cycles << adjustment;
-
- /* adjust the base and error for the adjustment */
- ret = cycles_delta << adjustment;
- *error += clock->interval_cycles << adjustment;
- /* XXX adj error for cycle_delta offset? */
- }
- return ret;
+ c->xtime_interval = (u64)c->cycle_interval * c->mult;
}


Index: linux-2.6-mm/include/linux/timex.h
===================================================================
--- linux-2.6-mm.orig/include/linux/timex.h 2006-06-22 11:56:06.000000000 +0200
+++ linux-2.6-mm/include/linux/timex.h 2006-06-22 11:56:24.000000000 +0200
@@ -303,8 +303,10 @@ time_interpolator_reset(void)

#endif /* !CONFIG_TIME_INTERPOLATION */

+#define TICK_LENGTH_SHIFT 32
+
/* Returns how long ticks are at present, in ns / 2^(SHIFT_SCALE-10). */
-extern u64 current_tick_length(long);
+extern u64 current_tick_length(void);

extern int do_adjtimex(struct timex *);

Index: linux-2.6-mm/kernel/timer.c
===================================================================
--- linux-2.6-mm.orig/kernel/timer.c 2006-06-22 11:56:07.000000000 +0200
+++ linux-2.6-mm/kernel/timer.c 2006-06-22 12:29:37.000000000 +0200
@@ -771,7 +771,7 @@ static void update_ntp_one_tick(void)
* specified number of bits to the right of the binary point.
* This function has no side-effects.
*/
-u64 current_tick_length(long shift)
+u64 current_tick_length(void)
{
long delta_nsec;
u64 ret;
@@ -780,14 +780,8 @@ u64 current_tick_length(long shift)
* ie: nanosecond value shifted by (SHIFT_SCALE - 10)
*/
delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
- ret = ((u64) delta_nsec << (SHIFT_SCALE - 10)) + time_adj;
-
- /* convert from (SHIFT_SCALE - 10) to specified shift scale: */
- shift = shift - (SHIFT_SCALE - 10);
- if (shift < 0)
- ret >>= -shift;
- else
- ret <<= shift;
+ ret = (u64)delta_nsec << TICK_LENGTH_SHIFT;
+ ret += (s64)time_adj << (TICK_LENGTH_SHIFT - (SHIFT_SCALE - 10));

return ret;
}
@@ -795,7 +789,6 @@ u64 current_tick_length(long shift)
/* XXX - all of this timekeeping code should be later moved to time.c */
#include <linux/clocksource.h>
static struct clocksource *clock; /* pointer to current clocksource */
-static cycle_t last_clock_cycle; /* cycle value at last update_wall_time */

#ifdef CONFIG_GENERIC_TIME
/**
@@ -814,7 +807,7 @@ static inline s64 __get_nsec_offset(void
cycle_now = clocksource_read(clock);

/* calculate the delta since the last update_wall_time: */
- cycle_delta = (cycle_now - last_clock_cycle) & clock->mask;
+ cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;

/* convert to nanoseconds: */
ns_offset = cyc2ns(clock, cycle_delta);
@@ -928,7 +921,7 @@ static int change_clocksource(void)
timespec_add_ns(&xtime, nsec);

clock = new;
- last_clock_cycle = now;
+ clock->cycle_last = now;
printk(KERN_INFO "Time: %s clocksource has been installed.\n",
clock->name);
return 1;
@@ -969,7 +962,7 @@ void __init timekeeping_init(void)
write_seqlock_irqsave(&xtime_lock, flags);
clock = clocksource_get_next();
clocksource_calculate_interval(clock, tick_nsec);
- last_clock_cycle = clocksource_read(clock);
+ clock->cycle_last = clocksource_read(clock);
ntp_clear();
write_sequnlock_irqrestore(&xtime_lock, flags);
}
@@ -989,7 +982,7 @@ static int timekeeping_resume(struct sys

write_seqlock_irqsave(&xtime_lock, flags);
/* restart the last cycle value */
- last_clock_cycle = clocksource_read(clock);
+ clock->cycle_last = clocksource_read(clock);
write_sequnlock_irqrestore(&xtime_lock, flags);
return 0;
}
@@ -1016,60 +1009,134 @@ static int __init timekeeping_init_devic
device_initcall(timekeeping_init_device);

/*
+ * If the error is already larger, we look ahead another tick,
+ * to compensate for late or lost adjustments.
+ */
+static __always_inline int clocksource_bigadjust(int sign, s64 error, s64 *interval, s64 *offset)
+{
+ int adj;
+
+ /*
+ * As soon as the machine is synchronized to the external time
+ * source this should be the common case.
+ */
+ error >>= 2;
+ if (likely(sign > 0 ? error <= *interval : error >= *interval))
+ return sign;
+
+ /*
+ * An extra look ahead dampens the effect of the current error,
+ * which can grow quite large with continously late updates, as
+ * it would dominate the adjustment value and can lead to
+ * oscillation.
+ */
+ error += current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
+ error -= clock->xtime_interval >> 1;
+
+ adj = 0;
+ while (1) {
+ error >>= 1;
+ if (sign > 0 ? error <= *interval : error >= *interval)
+ break;
+ adj++;
+ }
+
+ /*
+ * Add the current adjustments to the error and take the offset
+ * into account, the latter can cause the error to be hardly
+ * reduced at the next tick. Check the error again if there's
+ * room for another adjustment, thus further reducing the error
+ * which otherwise had to be corrected at the next update.
+ */
+ error = (error << 1) - *interval + *offset;
+ if (sign > 0 ? error > *interval : error < *interval)
+ adj++;
+
+ *interval <<= adj;
+ *offset <<= adj;
+ return sign << adj;
+}
+
+/*
+ * Adjust the multiplier to reduce the error value,
+ * this is optimized for the most common adjustments of -1,0,1,
+ * for other values we can do a bit more work.
+ */
+static void clocksource_adjust(struct clocksource *clock, s64 offset)
+{
+ s64 error, interval = clock->cycle_interval;
+ int adj;
+
+ error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
+ if (error > interval) {
+ adj = clocksource_bigadjust(1, error, &interval, &offset);
+ } else if (error < -interval) {
+ interval = -interval;
+ offset = -offset;
+ adj = clocksource_bigadjust(-1, error, &interval, &offset);
+ } else
+ return;
+
+ clock->mult += adj;
+ clock->xtime_interval += interval;
+ clock->xtime_nsec -= offset;
+ clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
+}
+
+/*
* update_wall_time - Uses the current clocksource to increment the wall time
*
* Called from the timer interrupt, must hold a write on xtime_lock.
*/
static void update_wall_time(void)
{
- static s64 remainder_snsecs, error;
- s64 snsecs_per_sec;
- cycle_t now, offset;
+ cycle_t offset;

- snsecs_per_sec = (s64)NSEC_PER_SEC << clock->shift;
- remainder_snsecs += (s64)xtime.tv_nsec << clock->shift;
+ clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;

- now = clocksource_read(clock);
- offset = (now - last_clock_cycle)&clock->mask;
+#ifdef CONFIG_GENERIC_TIME
+ offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
+#else
+ offset = clock->cycle_interval;
+#endif

/* normally this loop will run just once, however in the
* case of lost or late ticks, it will accumulate correctly.
*/
- while (offset > clock->interval_cycles) {
- /* get the ntp interval in clock shifted nanoseconds */
- s64 ntp_snsecs = current_tick_length(clock->shift);
-
+ while (offset >= clock->cycle_interval) {
/* accumulate one interval */
- remainder_snsecs += clock->interval_snsecs;
- last_clock_cycle += clock->interval_cycles;
- offset -= clock->interval_cycles;
+ clock->xtime_nsec += clock->xtime_interval;
+ clock->cycle_last += clock->cycle_interval;
+ offset -= clock->cycle_interval;
+
+ if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
+ clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
+ xtime.tv_sec++;
+ second_overflow();
+ }

/* interpolator bits */
- time_interpolator_update(clock->interval_snsecs
+ time_interpolator_update(clock->xtime_interval
>> clock->shift);
/* increment the NTP state machine */
update_ntp_one_tick();

/* accumulate error between NTP and clock interval */
- error += (ntp_snsecs - (s64)clock->interval_snsecs);
+ clock->error += current_tick_length();
+ clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
+ }

- /* correct the clock when NTP error is too big */
- remainder_snsecs += make_ntp_adj(clock, offset, &error);
+ /* correct the clock when NTP error is too big */
+ clocksource_adjust(clock, offset);

- if (remainder_snsecs >= snsecs_per_sec) {
- remainder_snsecs -= snsecs_per_sec;
- xtime.tv_sec++;
- second_overflow();
- }
- }
/* store full nanoseconds into xtime */
- xtime.tv_nsec = remainder_snsecs >> clock->shift;
- remainder_snsecs -= (s64)xtime.tv_nsec << clock->shift;
+ xtime.tv_nsec = clock->xtime_nsec >> clock->shift;
+ clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;

/* check to see if there is a new clocksource to use */
if (change_clocksource()) {
- error = 0;
- remainder_snsecs = 0;
+ clock->error = 0;
+ clock->xtime_nsec = 0;
clocksource_calculate_interval(clock, tick_nsec);
}
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/