This fixes a bug introduced in this PR:
https://github.com/ArduPilot/ardupilot/pull/25900
the bug is that when we don't go via the bouncebuffer on H7 we were
also skipping the cache invalidate/flush ops. This caused data
corruption for filesystem operations, particularly noticible as lua
scripts failing to load with parse errors or log corruption
`__main_thread_stack_base__` and `__main_thread_stack_end__` are
variables whose address is defined to be the corresponding part of the
stack. These are declared as `extern stkalign_t` in ChibiOS code, and
being declared as `extern uint32_t` in ArduPilot code creates a warning
at link time when using LTO. Correct the declaration to eliminate this
warning.
Also update `__main_stack_base__` and `__main_stack_end__` which don't
currently trigger this warning but serve similar purposes and so might
in the future.
The hardware expects an alignment of `stkalign_t` (which is 8 bytes) and
the linker script defines the variable values with this alignment as
well, so this is safe.
No code size or functional change.
this fixes DMA failures affecting IMUs on H743. The reason for the
failure is not yet clear, but this reverts back to our old stragegy of
always using the bouncebuffer on H7 as a quick workaround
this allows for hwdef.dat lines like this:
PA8 MCO1 RCC
PC9 MCO2 RCC
you can control the scaling by defining the following:
- STM32_MCO1SEL
- STM32_MCO1PRE_VALUE
- STM32_MCO2SEL
- STM32_MCO2PRE_VALUE
add support for undef of STM32 lines and DMA_NOMAP
add support for F103 running at 24Mhz
add f103 variants of iofirmware
reduce memory usage in iomcu for new model
correct pwm output reporting
split iofirmware into config + mcu to allow f103 definition
don't enable non-existant timers on F103
this adds a chibios_flags.h which contains the flags we pass to the
ChibiOS make, which allows us to have a make level dependency on the
flags
this fixes (among other things) the build of copter followed by
bootloader for SPRacingH7
this implements FATFS memory allocation in the HAL_ChibiOS malloc.c
which fixes mkdir() after a format of the microSD
The strategy is to deny large allocations and try for DMA allocations
if possible. This both fixes mkdir() and makes filesystem operations
faster as we have less copying of memory via the bouncebuffer
these are causing some boards to crash on startup. Tested with a
QiotekZealotH743 which doesn't get out of setup_usb_strings()
once we have debugged this we can re-add the functionality
Andrew Tridgell
Thomas Watson
thu5cob
Andy Piper
Tom Pittenger
bugobliterator
Martin Luessi