From: Chris Metcalf on
We were using the same 5-sec minsec for the clocksource and sched_clock
that we were using for the clock_event_device. For the clock_event_device
that's exactly right since it has a short maximum countdown time.
But for sched_clock we want to avoid wraparound when converting from
ticks to nsec over a much longer window, so we force a shift of 10.
And for clocksource it seems dodgy to use a 5-sec minsec as well, so we
copy some other platforms and force a shift of 22.

Signed-off-by: Chris Metcalf <cmetcalf(a)tilera.com>
---
arch/tile/kernel/time.c | 33 +++++++++++++++++++--------------
1 files changed, 19 insertions(+), 14 deletions(-)

diff --git a/arch/tile/kernel/time.c b/arch/tile/kernel/time.c
index b9ab25a..6bed820 100644
--- a/arch/tile/kernel/time.c
+++ b/arch/tile/kernel/time.c
@@ -36,16 +36,6 @@
/* How many cycles per second we are running at. */
static cycles_t cycles_per_sec __write_once;

-/*
- * We set up shift and multiply values with a minsec of five seconds,
- * since our timer counter counts down 31 bits at a frequency of
- * no less than 500 MHz. See @minsec for clocks_calc_mult_shift().
- * We could use a different value for the 64-bit free-running
- * cycle counter, but we use the same one for consistency, and since
- * we will be reasonably precise with this value anyway.
- */
-#define TILE_MINSEC 5
-
cycles_t get_clock_rate(void)
{
return cycles_per_sec;
@@ -68,6 +58,14 @@ cycles_t get_cycles(void)
}
#endif

+/*
+ * We use a relatively small shift value so that sched_clock()
+ * won't wrap around very often.
+ */
+#define SCHED_CLOCK_SHIFT 10
+
+static unsigned long sched_clock_mult __write_once;
+
static cycles_t clocksource_get_cycles(struct clocksource *cs)
{
return get_cycles();
@@ -78,6 +76,7 @@ static struct clocksource cycle_counter_cs = {
.rating = 300,
.read = clocksource_get_cycles,
.mask = CLOCKSOURCE_MASK(64),
+ .shift = 22, /* typical value, e.g. x86 tsc uses this */
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
};

@@ -88,8 +87,10 @@ static struct clocksource cycle_counter_cs = {
void __init setup_clock(void)
{
cycles_per_sec = hv_sysconf(HV_SYSCONF_CPU_SPEED);
- clocksource_calc_mult_shift(&cycle_counter_cs, cycles_per_sec,
- TILE_MINSEC);
+ sched_clock_mult =
+ clocksource_hz2mult(cycles_per_sec, SCHED_CLOCK_SHIFT);
+ cycle_counter_cs.mult =
+ clocksource_hz2mult(cycles_per_sec, cycle_counter_cs.shift);
}

void __init calibrate_delay(void)
@@ -117,9 +118,14 @@ void __init time_init(void)
* counter, plus bit 31, which signifies that the counter has wrapped
* from zero to (2**31) - 1. The INT_TILE_TIMER interrupt will be
* raised as long as bit 31 is set.
+ *
+ * The TILE_MINSEC value represents the largest range of real-time
+ * we can possibly cover with the timer, based on MAX_TICK combined
+ * with the slowest reasonable clock rate we might run at.
*/

#define MAX_TICK 0x7fffffff /* we have 31 bits of countdown timer */
+#define TILE_MINSEC 5 /* timer covers no more than 5 seconds */

static int tile_timer_set_next_event(unsigned long ticks,
struct clock_event_device *evt)
@@ -211,8 +217,7 @@ void do_timer_interrupt(struct pt_regs *regs, int fault_num)
unsigned long long sched_clock(void)
{
return clocksource_cyc2ns(get_cycles(),
- cycle_counter_cs.mult,
- cycle_counter_cs.shift);
+ sched_clock_mult, SCHED_CLOCK_SHIFT);
}

int setup_profiling_timer(unsigned int multiplier)
--
1.6.5.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/