[Libre-soc-dev] daily kan-ban update 16dec2021

[lkcl]

dealing with fetch when MMU is enabled, is hairy as anything.

i encountered an alarming bug in nmigen Simulator where a signal was clearly delayed by three clock cycles: "fixed" by assigning the Signal combinatorially to another and used that: problem "solved".

currently the fake microcoded op FETCH_FAILED successfully runs through the MMU and installs the TLB entry.  however the re-run of the instruction is not yet triggering: the address needs to be put back into fetch apart from anything.

at least the path has been explicitly tested so it is known, as of last week, that "fetch fail lookup fetch" should work: it is just TestIssuer that now needs to do it.

l.