Re-generating the SoC/Software headers was causing some un-expected behaviours for users not familiar
with the flow. For example doing a --load with a different configuration, was re-generating the Software
headers and messing up things when trying to run software on the SoC.
Moving CRG simplify reset with NaxRiscv debug module and is in fact more logical.
Also do some minor updates/cosmetic changes while touching CRG/SoCCore.
rst was not directly assigned/used on reset_less clock domains, so reset_less
property was not really useful. With the changes on stream.CDC, having a rst
(Even fixed at 0) is now mandatory on clock domains involved in the CDC, so this
also fixes targets.
CPU parameters are now selectively exposed to the user:
./digilent_arty.py --cpu-type=vexriscv_smp --help will show VexRiscv-SMP parameters.
./digilent_arty.py --cpu-type=naxriscv --help will show NaxRiscv parameters.
This adds support for v2 of the board via a --board-version argument
and a way to select the FPGA speed grade via a --speed-grade argument.
Note that the speed grade now defaults to -1. QMTech confirmed that
V1 of the board were made in two batches, one with -1 and one with -2,
while V2 of the board is all -1. So -1 is the safer default.
This also fixes the inversion of j10 and j11 and a typo in the pin
definition of jp3
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
- main_ram mem_map is now directly used by add_sdram when origin is None.
- max_sdram_size/min_l2_data_width are no longer exposed as targets arguments this can
still be used enforced directly in the few cases it is useful.
Collect --with-ethernet/--with-etherbone, --with-spi-sdcard/--with-sdcard,
etc. into ArgumentParser.add_mutually_exclusive_group()s. That way, we
get pretty --help output, and appropriate error messages if somebody
tries to ask for something that doesn't make sense.
Florent Kermarrec
Benjamin Herrenschmidt
Joey Bushagour
Romain Dolbeau
Gary Wong
Hans Baier
Shinken Sanada