No code changes, just file, variable and define names are changed.
First, it better describes the serial protocol used by DSMx satellite
receivers. Second, many people using Spektrum radio, assume Spektrum
protocol. This is the attempt to address those inaccuracies.
- both CC serial ports are now disabled by default (no telemetry);
- serial ports now have DSM2, DSMX (10bit) and DSMX (11bit) options;
- ReceiverGroups now have DSM (MainPort) and DSM (FlexiPort) options.
For DSM2 protocol there is an explicit resolution bit in the stream, so
the DSM2 should be selected. For DSMX there is no such bit, and user
should choose the resolution from the list configuring the spektrum port.
ReceiverGroups have single DSM option which is handled by the same driver.
Downside: this implementation saves received frame first, unrolls by the
end of frame. This should be ok, but may be improved by unrolling channels
on the fly in the rx callback.
Another minor difference is that a ChannelGroup is now bound to port:
DSM (MainPort) or DSM (FlexiPort). This was considered as acceptable
solution in order to not have 6 DSM options for each ChannelGroup and
even more in case of new DSM protocol variations.
Known problem: it is not possible to choose same protocols like
DSM2/DSM2 for two ports. It can be enabled by adding an exception to
common rule, though.
The DSMX throttle channel misbehavior (zero value) is not treated
specially yet. It should trigger the failsafe being out of bounds.
More info and data dumps are required to handle this properly.
PPM. This saves resources. Good suggestion Os. In this configuration we
could allow 12 channels of output but for now I'll leave it capped at 10 to
lessen resources on the mixer table.
With spektrum and camera stab enabled there was
1632 bytes heap remaining
180 bytes irq stack remaining
In the previous version the decoder could in rare cases get synced from
the middle of data stream in case of data byte equal to the S.Bus start
of frame (SOF) byte (wrong data will be rejected but it was not perfect).
Now it waits for the real start of frame and then checks the SOF byte.
- does not glitch when used in 2-frame mode (DM9, 9503, etc)
- does NOT provides yet DSMX stream decoding - do NOT merge
- uses a bit more time in the interrupt, but frees 16 bytes of RAM.
This is done to help decoding the weird DSMX stream which does not
contain explicit resolution/frame/lost frames info and needs special
processing (to be done yet).
TelemetrySettings object removed (saved 200+ bytes of RAM). Telemetry
port speed moved to the HwSettings object. Added GPS port speed setting.
GCS code updated to reflect changes and support both fields.
Besides of knob PID tuning it is now possible to use throttle channel
to ramp-shape PID coefficients. This can be used to lower some PIDs on
VTOL while sinking to prevent wobble.
To use the feature select throttle as control input, choose throttle
range max and min values, assign the instance to particular PID
coefficient and define a range for it. When throttle is lower than
defined throttle range min value (or higher than max), then min and max
PID values will be used accordingly. Changing throttle from throttle
min to max will linearly scale PID value.
Note that it is possible to set MinPID > MaxPID. In that case increasing
control input value will decrease the PID coefficient.
Up to 3 independent instances can be configured. The number can be
increased changing the UAVO definition, but at the cost of extra RAM.
This module will periodically update values of stabilization PID settings
depending on configured input control channels. New values of stabilization
settings are not saved to flash, but updated in RAM. It is expected that the
module will be enabled only for tuning. When desired values are found, they
can be read via GCS and saved permanently. Then this module should be
disabled again.
TIM5-8. Also TIM1 was not handled probably (for CC either) as the name of the
IRQ is TIM1_CC and TIM1_UP for the capture compare versus update. I haven't
checked the downstream code that the registers it uses to map from a context to
event is invariant under timer channel.
TIM5-8. Also TIM1 was not handled probably (for CC either) as the name of the
IRQ is TIM1_CC and TIM1_UP for the capture compare versus update. I haven't
checked the downstream code that the registers it uses to map from a context to
event is invariant under timer channel.
them symbolic constants.
- A timeout is 0
- A missing driver is 65534
- An invalid channel is 65535
ManualControl: Make it deal with the values explicitly. A timed out value
should not be treated like a minimum duration signal. Instead it does not
updated the scaled value but marks the data window as invalid to trigger the
failsafe.
Move the configuration files for INS from AHRS* to INS*. Strip out unused
fields in settings and merge calibration and settings since settings has
basically no information.
Allocate per-instance data for drivers from the heap
rather than as static variables from the .data segment.
This converts > 800 bytes of RAM from being always consumed
as static data into being allocated from the heap only when
a particular feature is enabled in the hwsettings object.
A minimal config (no receivers, flexi port disabled, main port
disabled) leaves 2448 bytes of free heap. That's our new baseline.
Approximate RAM (heap) costs of enabling various features:
+ 632 Serial Telemetry (includes 400 bytes of Rx/Tx buffers)
+ 108 PWM Rcvr
+ 152 PPM Rcvr
+ 112 Spektrum Rcvr
+ 24 S.Bus (Should be closer to 68 since driver is still using
static memory)
There are still some drivers that pre-allocate all of their memory
as static data. It'll take some work to convert those over to
dynamically allocating their instance data.
PWM and PPM can now coexist in the same load and be
selected at boot time via the hwsettings UAVObject.
This is basically a complete restructuring of the
way the drivers interact with the TIM peripheral in
the STM32.
As a side effect, the PWM and PPM drivers are now
ready to support multiple instances of each.
This also provides the first step toward being able
to reassign some of the PWM input pins to be servo
output pins. Still more work required, but this is
a good start.
transfers from IRQ. Also catch the double 0x70084 event which was locking up
the FSM with -Os enabled. I did this in a cheating way (filtering the event
based on state) but it's the cleanest I can see. Hopefully a DMA version of
I2C will fix this.
can catch it and return -1. Do NOT panic when they are null as this is
something that should be caugth at run time in some cases.
Ideally as a compromise in a task start it could have a set of
PIOS_Assert(ObjectNameHandle()) to make sure the minimal set of objects it
needs are there.
CRC. This wasn't the case on F1. With CRC the last byte of the buffer passed
to PIOS_SPI_TransferBlock is NOT USED. This is the case on both F1 and F2.
Also need to DeInit DMA before enabling or it doesn't enable successfully.
Finally added a timeout which sets a fail on the pios spi transfer in the case
that either of the dma channels fails to enable.
needed by users because if too much changes I change the FS magic and trigger a
wipe.
Possibly the erase should require a particular "magic" object id value to
execute? This would make it harder to do manually through UAVOs though.
This update saves 448 bytes of RAM with the current
crop of UAVObjects. It reduces a 19 byte struct to
8 bytes.
Note: This also introduces a limit of 65.534s for the
update periods.
This allows the GCS to emulate a receiver device via the
telemetry link.
Select "GCS" as your input type in the manualcontrol config
screen and calibrate it as normal.
Note: The expected values for the channels are in microseconds
just like a PWM or PPM input device. The channel values
are validated against minimum/maximum pulse lengths just
like normal receivers.
The small bootloaders (CC and PipX) are out of flash space
so their stopwatch implementation has been swapped out for
one based on the DELAY clock that takes about 500 bytes less
of code space.
Identical functionality is preserved.
This allows the spektrum and sbus receiver drivers to bind
directly to the usart layer using a properly exported API
rather than overriding the interrupt handler.
Bytes are now pushed directly from the usart layer into the
com layer without any buffering. The com layer performs all
of the buffering.
A further benefit from this approach is that we can put all
blocking/non-blocking behaviour into the COM layer and not
in the underlying drivers.
Misc related changes:
- Remove obsolete .handler field from irq configs
- Adapt all users of PIOS_COM_* functions to new API
- Fixup callers of PIOS_USB_HID_Init()
Update to use 32-bit microsecond values.
Remove PIOS_DELAY_DiffuS per consensus (caller can do it easily themselves).
Update the core delay logic per Stac's suggestion to a version that is
resistant to various overflows.
Address the following review feedback items:
- use stdint types
- avoid the use of magic numbers (define CYCCNTENA)
- remove expository comment about sneakiness and corresponding code, replace with something simpler based on the API
- remove commented/#if 0 code
Each channel was previously tracking a separate driver.
Now, channels are grouped within a channel group to save
RAM used for tracking and to better reflect how channels
are actually mapped.
Working spektrum bind routine, depending your TX try BIND_PULSES 3,5,7,9 (5 works with DX7)
Boot process takes too long on MB so bind command misses the window (20-140ms).
the throttle is < 0. This will make things like axis-lock work while armed
without throttle. HOWEVER don't hold your stick in the arming position for a
long time or you can wind up the integrals now.
zero and making the attitude get NaN. Wrote recovery code for that condition
(should never occur) and also force minimum dT to 1 ms (also shouldn't occur)
Also reduce heap has it does not fit in SRAM anymore (not with current compiler).
(that's ok since if there is more space available, it will be reclaimed).
Merge branch 'master' into OP-423_Mathieu_Change_Init_To_Reduce_Memory_Footprint
Conflicts:
flight/CopterControl/System/inc/pios_config.h
flight/Modules/ManualControl/manualcontrol.c
This is a port of a work-in-progress by Sambas onto
the new driver infrastructure needed for boot-time
configuration.
PPM and PWM still don't coexist in a build but this
is closer.
The initial baud rates of each interface are now forced in the
board init code.
Any modules using USARTs should have fields added to
their settings object to allow the user to change the
baud rate from the default by using the COM layer APIs.
Developers requiring custom baud rates before the settings
objects are in place should locally edit the cfg structs
to specify the desired baud rates.
Tested this heap2 at runtime with CC and new compiler since old one (current) triggers strict alliasing error.
That's ok since strict aliasing is disabled on OP, and CC only use heap1.
This should mark an end to the compile-time selection of HW
configurations.
Minor changes in board initialization for all platforms:
- Most config structs are marked static to prevent badly written
drivers from directly referring to config data.
- Adapt to changes in .irq fields in config data.
- Adapt to changes in USART IRQ handling.
Major changes in board initialization for CC:
- Use HwSettings UAVObj to decide which drivers to attach to
the "main" port and the flexi port, and select the appropriate
device configuration data.
- HwSettings allows choosing between Disabled, Telemetry, SBUS,
Spektrum,GPS, and I2C for each of the two ports.
- Use ManualControlSettings.InputMode to init/configure the
appropriate receiver module, and register its available rx channels
with the PIOS_RCVR layer. Can choose between PWM, Spektrum and PPM
at board init time. PPM driver is broken, and SBUS will work once
it is added to this UAVObj as an option.
- CC build now includes code for SBUS, Spektrum and PWM receivers in
every firmware image.
PIOS_USART driver:
- Now handles its own low-level IRQs internally
- If NULL upper-level IRQ handler is bound in at board init time
then rx/tx is satisfied by internal PIOS_USART buffered IO routines
which are (typically) attached to the COM layer.
- If an alternate upper-level IRQ handler is bound in at board init
then that handler is called and expected to clear down the USART
IRQ sources. This is used by Spektrum and SBUS drivers.
PIOS_SBUS and PIOS_SPEKTRUM drivers:
- Improved data/API hiding
- No longer assume they know where their config data is stored which
allows for boot-time alternate configurations for the driver.
- Now registers an upper-level IRQ handlerwith the USART layer to
decouple the driver from which USART it is actually attached to.
This separates the RTC device and interrupt handling
from the devices that rely on the tick notifications.
Drivers can now register tick notification functions
that will be called on each RTC tick event.
All receivers now fall under the same driver API provided
by pios_rcvr.c.
This is part of a larger sequence of commits that will
switch the receiver selection over to boot time dynamic
configuration via UAVObjects.
The FreeRTOS IDLE task was using 512 bytes of stack.
The UAVObject Event task was also using 512 bytes of stack.
Both have been reduced, recovering 400+ bytes of heap.
high RateKp terms. However it might be sensitive to gyro noise (vibrations).
In addition it is mathematically similar to lead shapign so probably only use
one or the other.
stabilization output a bit more resilient to the high frequency noise from
gyros. However this value shouldn't be too high as it will increase the phase
delay of the feedback loop and decrease stability. Default is 5 ms.
Note: this resests the stabilizationsettings object. Sorry guys.
Also implement some ordering (quite ugly still) in the module init and task creation order so we can decide which module to start/init first
and which module to start/init last.
This will be replaced/adapter with the uavobject list later (once it's implemented).
reserving some space for module init and task create parameters to customize module/task creation (this will be usefull once we get the list and customization from customer).
Changes have been made for OP and CC. Tested comped with CC,OP, sim_posix.
Only ran on bench with CC for couple of minutes (code increase expected but no dropping of stack which is good).
This gives task creation at the time wherethe all heap is available.
For CopterControl the following make options are available:
USE_TELEMETRY=[YES|NO|1|3] (default is YES, USART1)
USE_GPS=[YES|NO|1|3] (default is NO, USART3)
USE_SPEKTRUM=[YES|NO|1|3] (default is NO, USART3)
USE_SBUS=[YES|NO|1] (default is NO, USART1 only)
heap reamining is low (about 500) but stacks can be ajusted (specially the 200 bytes from system) to give the level close to 1Ko if needed.
Merge branch 'master' into OP-423_Mathieu_Change_Init_To_Reduce_Memory_Footprint
Conflicts:
flight/CopterControl/System/inc/FreeRTOSConfig.h
flight/CopterControl/System/inc/pios_config.h
- create linker section for those <module>Initialize()
- later this list will incorporate parameters as well. (this probably will be more a OP feature to swap/remove/delete module on the fly.
- this is not done at compile time anymore by Makefile.
- this will allow us to have control on the module start at run-time (not implemented but build the ground for it).
- this simplify the startup (Part of code re-org).
- this change does not affect sim_posix and win32 (since they don't need that)
- ensure it's compiling for PiOS.posix
- port to PiOS.win32 but not tested (not compiled)
- tested on CC
- compile on OP.
- this free ~200 bytes.
- current avalable bytes (is we keep the same remaining bytes on the stack than before) is easily passed the 1.2Ko mark on CC with new gcc (4.5.2)
- this does not include init-reorg for each module (I still think more can be freed)
It was tested being merged with OP-472_CorvusCorax_CopterControl-Guidance_v3
branch, Spektrum on USART3 and GPS on USART1 and seems to work.
Currently defaults mimic original behavior, that is, if USE_SPEKTRUM
is not defined - define USE_PWM and USE_GPS. Thsi should be refactored
later to make it configurable from the Makefile.
Also it was not ported to the OP MB: it currently does not support the
S.Bus hardware and still has original behavior with the patch. But this
is one more step to dynamic configuration of ports.
TODO: This should be dynamic in the future.
But for now define any compatile combination of:
USE_I2C (shared with USART3)
USE_TELEMETRY
USE_GPS
USE_SPEKTRUM
USE_SBUS (USART1 only, it needs an invertor)
and optionally define PIOS_PORT_* to USART port numbers
Defaults are:
#define PIOS_PORT_TELEMETRY 1
#define PIOS_PORT_GPS 3
#define PIOS_PORT_SPEKTRUM 3
#define PIOS_PORT_SBUS 1
#define USE_TELEMETRY
#define USE_GPS
Telemetry, GPS and PWM input are enabled by default.
CAREFULL: the heap section need to be the last section in RAM to avoid overwritting data...
Tested with GCC 4.5.2 this gives 1K of free bytes usable in heap right away (including the 200 bytes saved just by using the new gcc).
This does not include any code re-org yet!
I managed to test CC with heap2 changes and the init stack claimed back to heap once scheduler starts.
the changes of this commit are OP related (just cleanup on CC side):
Arch specific stuff (in reset vector) to hide this from portable code:
- switch back to MSP stack before starting the scheduler so that the sheduler can use the IRQ stack (when/if needed).
- call the C portable function in heap2 to claim some stack back (the number to claim is taken from linker file).
- start the scheduler from reset vector (I move this here from main because it make sense to not go back to C (so that I don't need to copy the rolled stack in case the sheduler returns). This make it more clean.
- Also I have added the call to the mem manager if sheduler return. that way, we don't reset indefinitely if memory runs out. We will go to this handler and figure things out (right now, it's just looping but at least not rebooting. Probably trap NMI would be better (later improvement).
- switch back to MSP stack before starting the scheduler so that the sheduler can use the IRQ stack (when/if needed).
- call the C portable function in heap1 to claim some stack back (the number to claim is taken from linker file).
- start the scheduler from reset vector (I move this here from main because it make sense to not go back to C (so that I don't need to copy the rolled stack in case the sheduler returns). This make it more clean.
- Also I have added the call to the mem manager if sheduler return. that way, we don't reset indefinitely if memory runs out. We will go to this handler and figure things out (right now, it's just looping but at least not rebooting. Probably trap NMI would be better (later improvement).
The part missing for this part is the weak attribute for the function in heap1.c so that we don't have to update everything with empty stub.
I think the weak atrribute for C function called in assembly is arch dependent so I am not sure if this is possible (will look into it, maybe somebody outthere nows).
Right now, it's heap1 dependent and won't work with heap2. I will clean that up the next couple of days.
I did some test and it looks good.
this is without init code re-organization so we don't free as much as we will be it's good starts.
This compile with sim_posix (since it does not affect portable code) so this is really clean.
I only tested this with CC. I will port it for OP when I will work on heap2.
- use IRQStack for ISRs (at begening of SRAM) (let's call it the irq stack)
- use end of heap for stack needed during initialization (let's call it the init stack).
- the systemStats in GCS indicate the remaining bytes in the IRQ stack (this is realy usefull to monitor our (nested) IRQs.
This is the base ground to provide as much memory as possible available at task creation time.
Next step is to re-organize the initialization in order to move all the init out of the thread's stacks onto the init stack.
This will provide as much memory as possible available at task creation time.
Basically the stack during initialization will be destroyed once the scheduler starts and dynamic alloc are made (since the init stack is at the end of the heap). We will need to make sure we don't clobber the heap during initialization otherwise this will lead to stack corruption.
