We already had the hooks for ccmram and ram1 and ram2 and ram3 and xsram
for "other" ram sections, but there was no method, out of the box, for
placing a function in "normal" ram. The "easy" method of labelling a
function as ".data.someramfunc" causes warnings because code isn't meant
to have a name like ".data*" so we must explicitly add a new code
section.
Signed-off-by: Karl Palsson <karlp@tweak.net.au>
Add support for a section(s) ".noinit*" that will be allocated in ram,
but not cleared or initialized. This can be used for passing variables
between a bootloader and an app for instance, or even just between
restarts of your application.
Without any assigned usages of the section, there is zero change in ram
usage. The align does nothing when the prior section was already
aligned.
We were never going to be capable of supporting every single part
variant with their own files, so stop trying. We've been supporting the
linker script generator for a long time now, so move on from these old
static files.
Breaking: if you were using one of these, and don't wish to use the
linker script generator, you should simply declare a stub linker script,
and include the generic section definitions file, and provide this in
your own application.
Old:
LDSCRIPT = $(OPENCM3_DIR)/lib/stm32/l1/stm32l15xxb.ld
New (linker script generator):
DEVICE=stm32l151cb
New (manual):
* Add file mymemorymap.ld, with contents similar too:
```
MEMORY
{
rom (rx) : ORIGIN = 0x08000000, LENGTH = 256K
ram (rwx) : ORIGIN = 0x20000000, LENGTH = 48K
}
/* Include the common ld script. */
INCLUDE cortex-m-generic.ld
```
LDSCRIPT=mymemorymap.ld
See ld/README for more information on using the linker script generator
Replaced & with &&
Wait for all three bits to be set - SB, MSL, and BUSY.
Old code worked by chance, use booleans to correctly convey intent.
Reviewed-by: Karl Palsson <karlp@tweak.net.au>
Fixes: https://github.com/libopencm3/libopencm3/issues/234
uses the new mechanisms introduced to address a similar problem on F7.
Tested on a medium density part (0x429) that returns the same ids as
before, tested on a high density part that now _doesnt_, but that's now
correct :)
Reduced the sea of enums, and renamed config parameters to match other
implementations, cribbing off of the G0 config, as it is closer to the level
of complexity. Updated initialization code to utilize the new values.
Added flash and LDO configuration from RCC init to be more consistent with
STM32 platform initialization.
PLL configuration on the H7 is pretty involved, and takes a number of
configurations to make it work. In order to make peripheral drivers a bit
easier to implement, working with a soft clock tree in the rcc module which
stores the clock settings for each clock as they are setup such that users
can request the clock value from the RCC module for configuration. Added
getter for the clock which allows the user to pass the base address of the
peripheral, and get the peripheral clock value for convenience.
Clock configuration is still missing values for setting up all of the kernel
clocks for the peripherals, but this is in work, and there is a framework to
do so.
Have tested to 400MHz without issue. Peripherals that are explicitly supported
are working and the clock tree values appear to follow correctly.
Added LDO settings to allow setting the scaling to support high frequencies.
In this commit, support for the different base addresses for different
F7 parts is added, but the mechanism is now in place for L1 and others.
Reviewed-by: Karl Palsson <karlp@tweak.net.au>
(whitespace fixed, commit msg reworded)
We don't need groupings around each enum, they format nicely into a
section already. Likewise, the doxygen _is_ documentation, so we don't
need extra versions of it in places. Also fix a few warnings generated.
According to L4/L0/G0 RM, in case of clock error, interrupt flag must
be cleared, and CECS flag should be cleared as soon as clock meets
requirement.
Reviewed-on: https://github.com/libopencm3/libopencm3/pull/1062
If noise source error occurs, flag must be cleared and data register
must be discarded (at least 12 reads to flush pipeline on G0). Other
device mention start/restart of chip, so, do both (better safe than
sorry).
Reviewed-on: https://github.com/libopencm3/libopencm3/pull/1062
- match the same logic as the f0 api
- use ADC_CFGR1_AWD1CH_VAL macro to mask the channel bits
- don't check if channel is < 18q
- enable the awd in addition to setting the selection to single channel monitoring (in following with the signature and @brief 'enable'
Unlike with the doxygen source generation, we can't autoguess which of
the class files are eligible automatically. Instead, make a stub hid
file, (which we can now start adding to, if desired) and include it in
all builds that include other class stubs.
Devan Lai