To prepare for STM32F2/F4 DMA support, introduce a new libmaple DMA
API, and move some code around to make priority level and interrupt
handling more generic.
The new API is based on a new set of types (dma_tube, struct
dma_tube_reg_map, enum dma_request_src, enum dma_cfg_flags, and struct
dma_tube_config).
The central abstraction is the dma_tube type. STM32F2/F4 use DMA
streams to control dataflow, and STM32F1 uses channels. dma_tube
stands for whichever is appropriate for the current target. Dealing
with tubes allows for configuring and using DMA with opaque tube
values in the same source, instead of (as with ST's firmware)
requiring two separate codebases.
The new API is also more user-friendly, as it doesn't require knowing
which DMA address registers to set and which configuration register
flags go along with them. It now suffices to specify the source and
destination for the DMA transfer, along with their sizes. This avoids
confusion (e.g. for memory-to-memory transfers, data flows from the
peripheral address register to the memory register, which might be
surprising on F2, which has two memory address registers).
The old API (based on enum dma_mode_flags and dma_setup_transfer()) is
still available on F1, but deprecate it.
Signed-off-by: Marti Bolivar <mbolivar@leaflabs.com>
Breaking change set up: struct dma_handler_config is no longer part of
the public API in <libmaple/dma.h>. User code which was touching
these was always mistaken; it should be using dma_attach_interrupt()
or dma_detach_interrupt() instead.
Other than that, just move the nonportable bits in <libmaple/dma.h>
and libmaple/dma.c to the appropriate places under
libmaple/stm32f1/. (Ouch. This is almost everything.) Patch the
(new) STM32F1 <series/dma.h> here and there to make everything
compile; this is mostly limited to forward-declaring struct dma_dev
and providing a hack _dma_dev_regs() declaration so inline functions
in the series header can still access a device's registers.
Signed-off-by: Marti Bolivar <mbolivar@leaflabs.com>
Move libmaple/*.h to (new) libmaple/include/libmaple/. The new
accepted way to include a libmaple header foo.h is with:
#include <libmaple/foo.h>
This is more polite in terms of the include namespace. It also allows
us to e.g. implement the Arduino SPI library at all (which has header
SPI.h; providing it was previously impossible on case-insensitive
filesystems due to libmaple's spi.h).
Similarly for Wirish.
The old include style (#include "header.h") is now deprecated.
libmaple/*.h:
- Change include guard #defines from _FOO_H_ to _LIBMAPLE_FOO_H_.
- Add license headers where they're missing
- Add conditional extern "C" { ... } blocks where they're missing
(they aren't always necessary, but we might was well do it against
the future, while we're at it.).
- Change includes from #include "foo.h" to #include <libmaple/foo.h>.
- Move includes after extern "C".
- Remove extra trailing newlines
Note that this doesn't include the headers under libmaple/usb/ or
libmaple/usb/usb_lib. These will get fixed later.
libmaple/*.c:
- Change includes from #include "foo.h" to #include <libmaple/foo.h>.
Makefile:
- Add I$(LIBMAPLE_PATH)/include/libmaple to GLOBAL_FLAGS. This allows
for users (including Wirish) to migrate their code, but should go
away ASAP, since it slows down compilation.
Wirish:
- Move wirish/**/*.h to (new) wirish/include/wirish/. This ignores
the USB headers, which, as usual, are getting handled after
everything else.
- Similarly generify wirish/boards/ structure. For each supported
board "foo", move wirish/boards/foo.h and wirish/boards/foo.cpp to
wirish/boards/foo/include/board/board.h and
wirish/boards/foo/board.cpp, respectively. Also remove the #ifdef
hacks around the .cpp files.
- wirish/rules.mk: put wirish/boards/foo/include in the include path
(and add wirish/boards/foo/board.cpp to the list of sources to be
compiled). This allows saying:
#include <board/board.h>
instead of the hack currently in place. We can allow the user to
override this setting later to make adding custom board definitions
easier.
- Disable -Werror in libmaple/rules.mk, as the current USB warnings
don't let the olimex_stm32_h103 board compile. We can re-enable
-Werror once we've moved the board-specific bits out of libmaple
proper.
libraries, examples:
- Update includes accordingly.
- Miscellaneous cosmetic fixups.
Signed-off-by: Marti Bolivar <mbolivar@leaflabs.com>
libmaple/dma.c defines DMA interrupts __irq_dma_channel[1-7],
consistent with what is specified by support/ld/names.inc. However,
names.inc is inconsistent with what support/ld/libcs3_stm32_src/
expects. Specifically, it contradicts the files
- support/ld/libcs3_stm32_src/stm32_isrs.S
- support/ld/libcs3_stm32_src/stm32_vector_table.S
Which use the names __irq_dma1_channel[1-7].
Change names.inc and dma.c to use the correct IRQ names.
The original names.inc/libcs3_stm32_src inconsistency was introduced
in 43d6921658, but dma.c had the correct
names until ec3cf2903f, where they were
renamed for consistency with names.inc. At that point, DMA interrupts
stopped working. (This was documented in the commit message).
Thanks to forum user robodude666 for tracking this down.
Go through overlong source code lines and convert as many of them as
appropriate to be 80-column clean. This mostly affects license
headers. Overlong lines are determined by running following from the
libmaple base directory:
$ ack-grep --nocolor --nogroup --cpp --cc --ignore-dir=usb -- '.{80}'
Note that this excludes libmaple's usb subdirectory, which is still
full of ST code that doesn't follow the libmaple source code
guidelines.
Contents of ~/.ackrc (these won't matter, but are included for
completeness):
--ignore-dir=docs
--ignore-dir=build
--type-set
ld=.ld
--type-set
rst=.rst
--type-set
txt=.txt
--type-set
mk=.mk
Simple USART receiver to SRAM buffer demo partially working.
Interrupting when buffer is full fails mysteriously. GDB thinks
we ended up in an STM32 reserved exception.
renamed SysTick_Handler back to SysTickHandler since all of our linker magic/lanchon-stm32 depends on that name. added backup register support in order to test independent watchdog support; it seems to work. next major test target is DMA support.
