litex seems to be a common standard way to set up socs. both microwatt and minerva use it. the combination of both should get us started.
not quite the same thing, may prove useful.
(In reply to Luke Kenneth Casson Leighton from comment #1)
> not quite the same thing, may prove useful.
contains a UART example which should do well after adaptation, for the helloworld test.
hi florent, thank you for offering to help with this.
* we have a POWER9 compliant core written in nmigen
* therefore we need a hybrid combination of how both microwatt
and Minerva are set up
* the do_finalize from "litex/soc/cores/cpu/minerva/core.py"
* the reset_address gcc_flags and setup from
* the wishbone buses are *64* bit data
* we use the *Minerva* Load/Store and Fetch code therefore the
ibus and dbus can be set up according to how *Minerva* does it,
*not* the "stall" method from microwatt.
* we actually support big *and* little-endian however please assume
little for now.
* we do not unfortunately have a Debug bus (examining microwatt core.py)
* nor interrupts (yet)
now, i did notice that there is a general practice of creating a
repository that takes a *copy* of the core, in a repository named
the libre-soc codebase is so large (45,000 for the IEEE754 FP unit,
20,000 for the main core, 5,000 for the support library, nmutil)
that it would not only be impractical, it would be detrimental to
development to follow this pattern.
if it is absolutely necessary to have this subdirectory (named
pythondata-cpu-libre-soc) then what i would advocate is that we
add such a subdirectory *DIRECTLY* to the libre-soc main git
repository, populate it with the necessary files, and for development
purposes create a symbolic link in the litex directory to the
libre-soc directory, pythondata-cpu-libre-soc.
i am happy to give you commit access to the relevant libre-soc git
repositories, if you send me an id_rsa.pub.
copy of test_issuer.v, generated with
$ python3 issuer.py generate -t v > test_issuer.v
(followed by hand-editing to get rid of some debug print statements)
> To be able to do the LiteX integration and use the different features i need from you:
> - the nmigen command to run to generate the top level of the CPU (is it what test_issuer is doing?)
yes, although it spews additional debug output at the moment. let us instead go with "soc/simple/issuer_verilog.py filename.v"
> - a description of the different signals of the CPU and the ones mandatory/ the ones optional.
these should be exactly as in Minerva: i really did copy the code as-is (except to make it 64 bit)
hmm however it seems i have some additional signals that get it "going". a sort-of debug interface, these are, mandatory:
these are optional:
> - the example you already simulated with the RAM and code from Microwatt.
line 40 is where the unit test is added to the list of binaries to run.
line 89 shows how to kick it out of "halted"
lines at 119 show how to set the pc externally. it defaults to zero including after reset so this is not strictly necessary.
lines at 124 show how to kick it into running instructions. go_insn_i setting to 1 is mandatory
when we have a "proper" debug interface (JTAG, other) this will be a little different
if you can help get us started i can send you patches.
florent, i think it is possible to literally copy minerva cpu_params, the full set of ibus and dbus names exactly the same, but removing interrupts (for now).
and i believe this should do the trick of allowing the core to start running immediately from the default pc=0
self.cpu_params = dict(
i_go_insn_i = 1
Created attachment 74 [details]
first version core.py for litex libre-soc
this is a first cut at combining minerva and microwatt to create
a libre-soc litex core.py. taking minerva i/d cache cpu signal
definitions, widening them to 64 bit, taking microwatt's ppc64le
flags and so on.
It's possible to plug your CPU to LiteX without modify LiteX itself, to so do, you could have a look at https://github.com/enjoy-digital/litex_vexriscv_smp and see how the CPU wrapper/software is created in the vexriscv_smp repository.
You can also look at https://github.com/enjoy-digital/litex_vexriscv_smp/blob/master/sim.py for a Verilator simulation. (capable of loading binaries and booting linux in the case of VexRiscv SMP).
That will probably be easier for you to develop like this, this will avoid forking LiteX. Once you have something that compiles correctly with Verilator (just execute sim.py on the case of VexRiscv SMP), I could help you getting this running if this is not already the case. (You can add --trace to the simulation to generate the waveform).
added to soc for now, following Florian's advice
commit 7ef2953a3787258d218941d1ab3b91d705670248 (HEAD -> master)
Author: Luke Kenneth Casson Leighton <firstname.lastname@example.org>
Date: Wed Jul 22 16:30:35 2020 +0100
first version of litex core (to be submitted upstream once tested)
encountering this error (which was first encountered when using
platform "None", and is coming up again when trying to use platform
Traceback (most recent call last):
File "litex/sim.py", line 155, in <module>
File "litex/sim.py", line 147, in main
trace_fst = 1)
File "/home/lkcl/src/libresoc/litex/litex/soc/integration/builder.py", line 219, in build
vns = self.soc.build(build_dir=self.gateware_dir, **kwargs)
File "/home/lkcl/src/libresoc/litex/litex/soc/integration/soc.py", line 922, in build
return self.platform.build(self, *args, **kwargs)
File "/home/lkcl/src/libresoc/litex/litex/build/sim/platform.py", line 44, in build
return self.toolchain.build(self, *args, **kwargs)
File "/home/lkcl/src/libresoc/litex/litex/build/sim/verilator.py", line 187, in build
File "/home/lkcl/src/libresoc/litex/litex/build/generic_platform.py", line 314, in finalize
"No default clock and no clock domain defined")
NotImplementedError: No default clock and no clock domain defined
trying this - taken from a random/arbitrary file:
+from migen import ClockDomain
from litex.build.generic_platform import Pins, Subsignal
from litex.build.sim import SimPlatform
from litex.build.sim.config import SimConfig
@@ -73,6 +76,12 @@ class SoCSMP(SoCCore):
self.platform.name = "sim"
+ self.clock_domains.cd_sys = ClockDomain()
+ self.comb += [
It's possible to avoid the interrupts in a first time, if your CPU wrapper does not have an interrupt Signal, UART POLLING mode will be used: https://github.com/enjoy-digital/litex/blob/master/litex/soc/integration/soc.py#L1063
I'm not sure which platform/target you are using, but you can find a minimal platform here:
Or can adapt one of the existing targets.
On Wed, Jul 29, 2020 at 12:04 PM Florent Kermarrec
> would you mind sending me the last libre_soc.v file you used for the simulation. I'm having a look at it but just want to use the verilog in a first time and have compilation issues with the first test_issuer.v you provided.
attached. yes, the first version i found doesn't "start" (remains in
"halted" state). this one is "up-to-date" with core.py and sim.py.
one thing: do note that we do full 8-bit (byte level) granularity on
the wishbone buses. i do not know if microwatt does that.
the current bug that i encountered - apart from MEMTEST_BUS_xxx not
being correctly set up for 64-bit - is that if the litex bus is set to
32 bit this _does_ actually work except that the PC is read - even for
an even word - into the *UPPER* 32 bits of the 64-bit Wishbone dat_r,
*not* the lower 32 bits.
getting unit tests against microwatt (1.bin, 2.bin) was sufficient to get up and running, here. running on an ECP5 FPGA has been confirmed https://www.youtube.com/watch?v=72QmWro9BSE