motion sense: Calculate loop time based on sensor needs

Currently the motion sense loop bases its sleep time based on the
fastest active sensor. This method has several flaws:

1. It does not take into account any task switching overhead
2. With a mix of interrupt driven and forced sensors the sleep time gets
   recalculated every time there is an interrupt causing the loop to
   oversleep
3. If multiple sensors do not have rates that are in sync the timing of
   the slower sensor will be off. For example if there was a sensor running
   at 50 Hz and one running at 20 Hz the slower sensor would end up being
   sampled at about 16 Hz instead of 20 Hz

This change calculates an ideal read time for every forced mode sensor
and calculates the sleep time based on the nearest read time. Every time
a sensor is read the next read time is calculated based on the ideal read
time not the actual read time so that reading does not drift because of
system load or other overhead.

BUG=b:129159505
TEST=Ran sensor CTS tests on arcada, without this change the
     magnetometer was failing 50 Hz tests at about 38 Hz with 30% jitter
     with this change in place 50 Hz was spot on with about 10% jitter
BRANCH=none

Change-Id: Ia4fccb083713b490518d45e7398eb3be3b957eae
Signed-off-by: Mathew King <mathewk@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1574786
Reviewed-by: Jett Rink <jettrink@chromium.org>

1 files changed, 0 insertions, 1 deletions

diff --git a/test/motion_common.h b/test/motion_common.h
index 3adb9f24e5..45d856d9ef 100644
--- a/test/motion_common.h
+++ b/test/motion_common.h
@@ -24,7 +24,6 @@
 #define TEST_LID_SAMPLE_SIZE (2 * 3)
 
 extern enum chipset_state_mask sensor_active;
-extern unsigned int motion_interval;
 
 extern struct motion_sensor_t motion_sensors[];
 extern const unsigned int motion_sensor_count;