Debugging requires many system components to all work together. The tests here perform an end-to-end test, communicating with gdb and OpenOCD. If a simulator or hardware passes all these tests, then you can be pretty confident that the actual debug interface is functioning correctly.
The following should be in the user's path:
rvv-0.9.x branch for riscv-gnu-toolchain should work if master does not have vector support yet)
--gdb when running gdbserver.py manually), which should be the latest from git://sourceware.org/git/binutils-gdb.git.
--sim_cmd when running gdbserver.py manually), which should be the latest from https://github.com/riscv/riscv-isa-sim.git.
--server_cmd when running gdbserver.py manually), which should be the latest from https://github.com/riscv/riscv-openocd.git.
To run a quick smoke test against spike, run
make. For a more comprehensive test against a variety of spike configurations, run
To run tests against hardware, or a specific spike configuration, manually invoke gdbserver.py:
You can run just a single test by specifying any part of its name on the command line, eg:
./gdbserver.py targets/RISC-V/spike64.py S0 runs SimpleS0Test. Once that test has failed, you can look at the log file to get an idea of what might have gone wrong.
For custom targets, you can create a .py file anywhere and pass its path on the command line. The Targets class in
targets.py contains documentation on what every variable means.
All output from tests ends up in the
logs/ subdirectory, with one log file per test. If a test fails, this is where to look.
You can see what spike is doing by adding
-l to the spike command, eg.:
./gdbserver.py --sim_cmd "$RISCV/bin/spike -l" targets/RISC-V/spike32.py Breakpoint
You can see what OpenOCD is doing by adding
-d to the OpenOCD command, eg.:
./gdbserver.py --server_cmd "openocd -d" targets/RISC-V/spike32.py Breakpoint
You can run gdb under valgrind by passing --gdb, eg.:
./gdbserver.py --gdb "valgrind riscv64-unknown-elf-gdb" targets/RISC-V/spike64.py DownloadTest