primarily modify the loading method to mark pages with A/D flags
as needed to work around lack of A/D handling in the trap handler
of Xous.
Also inside main.rs is a todo list for polishing off the vex-ii
port and turning into something actually sustainable for Denarius
once we have more of a full SoC model available.
the switch based on target is now absorbed into the new() function.
This paves the way for us to add more CSR ranges into the global
configuration object, i.e. IRQ csrs for I2C.
cramium-soc i think should be deprecated everywhere because
it is redundant with the board targets: anywhere you see a board
target that includes the crossbar soc it should also transitively
pull in the cramium-soc dependency equivalents.
we'll chase this out more later, but for now we're just trying
to get the PDDB to compile
add a note about the dirty flag - it's not "dead" per se, I
think it is helpful to leave a bread crumb for future hw revs
that might have hardware-enforced dirty bit setting on the PTE.
KB_C2 pin on this test board just got shorted to ground internally
in the chip. I think it may have been due to power glitches
when the ammeter was switching modes feeding through and causing
trouble for the system.
Anyawys, disable that row and remap some keys so we can continue
with testing. MPW boards will hopefully be retired soon enough...
This loop was useful in bringing up the GC2145 camera, so
we should keep this code around. had to be refactored/shuffled
around a bit to compensate for e.g. bitrot.
for some reason the loader has a lot of code devoted to formatting
debug prints of floating point numbers. In the version of the loader
that's being built for production I think we shouldn't need any
of this?
I think we're OK to get away with this, but, it's a pretty big
set of assumptions we're making so let's write it up in a location
that we're likely to look at if/when things break.
these currently only make sense in a simulation environment,
but they allow us to work the significant configuration test
cases for practical execution to make sure we haven't missed
something in the coreuser spec.
- disable IRQs on departure
- put EV enable outside of the HAL function
this is necessary because we need all HAL calls to be finished
before an interrupt happens to avoid locking problems
- updates enables or disables updates in the loader
- verilator-only will turn off routines that access hardware not
present in verilator (e.g. leads to a bus lockup)
- simulation-only turns of features that take a long amount of time
and aren't necessary in simulation, such as bulk clearing of
all RAM - in simulation, we can configure the simulator to do this
for us prior to boot; but in HW, of course, RAM is just random,
or worse, has the previous session's secrets in it.
cramium-soc is moving closer to the platonic SoC specifier,
and the board-* features are for the specific boards that use
cramium-soc
This division of feature flags feels a lot nicer and cleaner -
if it holds up to further development, I think this style might
become the new standard for Xous (as opposed to a single level
of feature flag where both SoC and board target are conflated in
terms of meaning - which kind of made sense when the CPU was
an FPGA and could actually have different features for different
boards, but for ASICs this wouldn't happen).
I felt uncomfortable leaving the oscillator setting entirely
outside the PLL routine, so as a compromise I defer the oscillator
switch-over until after the initial PLL settings are loaded.
This makes both the simulation model and the actual hardware
happy. I think probably what the simulation model *should* have
done is consider the "powerdown" pin when checking the validity
of the signals going into the PLL model - but this is impossible
to modify/check because the PLL is provided as an encrypted blob :(
this transition was in the reference C code, but in verilog simulation
it turns out it activates the PLL in an invalid mode which can lead
to glitching. Removing this makes both the simulation happy and
presumably the PLL happier as well.
this doesn't protect against malicious tampering (because
the public key is the developer key) but it will at least
catch correctness errors in USB upload and filesystem
behavior.
we can now recover from an I2C timeout and get the PMIC data
out, but mysteriously, when we leave the routine, the CPU hangs.
I think fundamentally the I2C instability is due to something
very wonky inside the chip itself, so the recovery only papers
over an issue that is only fixed with a reboot...
still have to resolve the static mut issue of getting the USB
controller state into the interrupt controller.
I'm not sure there *is* a good way to do it - this might actually
be "the" use case for static mut's - but the point of this
exercise was to just diff out the driver in machine mode versus
OS mode to try and figure out if there's a hardware problem
causing instability or if it's a driver issue
the conclusion so far is that the USB stack requires extremely
tight interrupt handling timing, and so, for better stability
we may have to mirgate away from the heavily user-space dependent
implementation currently in use.
disk enumerates now, and initiates r/w transactions but
the reference driver does not match the implementation in that
the USB stack can't access all of RAM, so we have to add
an intermediate stage that splits the command requests into
smaller chunks
this is in preparation for loader-based updates
may have to increase space even more when we pull in e.g.
firmware signing and verification primitives.