This is an unavoidably invasive commit, because it affects the unit tests
(which formerly exited using stop()), the test harness Verilog generator
(since it is no longer necessary), and the DRAM model (since it is no
longer connected). However, this should substantially reduce the effort
of building test harnesses in the future, since manual or semi-automatic
Verilog writing should no longer be necessary. Furthermore, there is now
very little duplication of effort between the Verilator and VCS test
harnesses.
This commit removes support for DRAMsim, which is a bit of an unfortunate
consequence. The main blocker is the lack of Verilog parameterization for
BlackBox. It would be straightforward to revive DRAMsim once support for
that feature is added to Chisel and FIRRTL. But that might not even be
necessary, as we move towards synthesizable DRAM models and FAME-1
transformations.
This is necessary to guarantee forward progress with RVC, since if the
BTB keeps mispredicting, the processor might never successfully
fetch both halves of a misaligned instruction.
It was possible that the result of a store-conditional could get lost if it
did not depend on the result of the corresponding load-reserved.
This was because the MSHR does not update the client state based on the
secondary requests. So the LR would acquire the line in clientExcusiveClean,
but then we would fail to update the metadata array to change the state
to clientExclusiveDirty.
The solution is to track whether a secondary acquire would cause the
line to be dirty. If so, use M_XWR instead of the primary command to
generate the update coherence state.
This addresses a bug in which changes in mstatus could
propagate to RoCCs before their time. Existing RoCCs that use
the status port will need to be modified to match this change.
This addresses the first half of #40.
Validating the target PC in the pipeline is cheaper than maintaining
the valid bits and control logic to guarantee the BTB won't ever
mispredict branch targets.
