When more than one task is concurrently trying to access
the same i2c bus and a timeout occurs on bus lock the
transfer would just continue and blow up the pios_i2c
driver. This has been fixed.
Actuator did not provide an option to completely shutdown selected channels
(set PWM pulse = 0). It is useful for camera stabilization boot delay (we
want few seconds of gimbal servo inactivity to calibrate gyros). It also
might be useful for failsafe on some channels. This option is now available.
It is used to disable camera outputs during fixed 7s delay after boot
instead of setting them to minimum position.
As a side effect, few bugs are fixed (ticks should be multiplied by
portTICK_RATE_MS, not divided, to get time in ms). And few floating point
operations were optimized out as well.
ActuatorCommand.UpdateTime was promoted to uint16 because it is not unusual
to have it 20000ms during flash updates (was seen on the CC after UAV
settings import). So it should be 16bit as well.
This is a partial rework of Cossacs' camera gimbal software.
This patch adds LPF to airframe attitude used for camera stabilisation
and feed forward for camera actuators. Either of options can be
disabled on the compilation level to save flash and RAM if not
required.
Original Cossacs' code was optimized and code flow shortcuts were
added where applicable.
This is a partial rework of Cossacs' camera gimbal software.
This patch adds LPF to some of manual control inputs. Mostly
useful are accessory channels (for camera gimbal control) and
yaw channel (for smooth filming). The code may be used for
stand-alone CC[3D]-based gimbal software, but also should work
for complete FC+camera system.
The uavo_handles section is not part of the .data segment so
it doesn't get automatically zeroed during the startup code.
This led to random data being stored in the table which resulted
in bus errors at runtime.
This change ensures that the table is zeroed before we start to
use it.
There shouldn't be any reason to need 8-byte alignment on the
F1 platform. This allows better packing of all malloc'd data.
Reducing this below 4-byte alignment is not recommended and will
likely result in misaligned pointers being passed to peripherals.
RAM savings is another 300 bytes.
Telemetry module was iterating over all UAVOs including
meta UAVOs and creating a periodic event item for each
object. These items cost us about 32 bytes for each list
item.
This is wasteful for two main reasons. First, meta UAVOs can't
meaningfully have periodic updates so excluding them entirely
makes sense. That halves the number of objects in this list since
there is one meta object for every data object. This is worth
about 500 bytes of RAM on CC.
Second, about half of the remaining UAVOs are not periodic by
default so they're wasting memory unless someone happens to want
to make them periodic at runtime. This is worth another 450 bytes
of RAM on CC.
So, objects that are configured as periodic during board init will
support all of the periodic config at runtime. Objects that are
*not* periodic during board init can only be made periodic on the
next boot.
Each object that you make periodic during init will cost you an
extra 32 bytes of RAM.
With erased settings, free RAM comparison for CC is:
Before: 2736 bytes free
After: 4048 bytes free
Total RAM savings with this update is 1312 bytes!
When reading the jedec device id the code only transfered one byte via spi leaving
the expected input buffer uninitialized. This may lead to the problem that flash
initialization fails because the expected input may be whatever the stack was set
when entering the function. The impact of the bug is somewhat limited tough as the
initialization usually takes place before starting up the rtos and thus is pretty
deterministic. So if the code passed init while testing it should pass init in
production as well.
