makefiles: change RISC-V GCC -march to rv32imac_zicsr

[LP-HAW]

Contribution description

This PR changes RISC-V GCC -march to rv32imac_zicsr

See PR #20753. (I may have deleted the PR branch, so I cannot reopen my original PR...)

Testing procedure

Same as #20753

Issues/PRs references

#17951
#18098
#20753

[riot-ci]

Murdock results

✔️ PASSED

1025538 makefiles: change RISC-V GCC -march to rv32imac_zicsr

Success	Failures	Total	Runtime
10171	0	10175	17m:35s

Artifacts

• Documentation preview

[maribu]

Hmm, I wonder if dropping support for older toolchains is a bit rushed. At least the ESP RISC-V toolchains in our CI have a binutils version that does not support the new ISA spec. That won't cause issues now because the ESP RISC-V stuff is not well integrated with the other RISC-V code in RIOT (yet). But it may still serve as an indication that we are moving a bit fast here.

Also: I wonder if this is rather a GDB bug, with GDB uses the new ISA spec while the ELF file was build with the older. Maybe it even is not possible bto extract the ISA spec version from the ELF file, and GDB would need a hint from the user.

Hence: Maybe adding explicitly an option to GDB to use the older spec in our make debug command would be the better option?

[LP-HAW]

At least the ESP RISC-V toolchains in our CI have a binutils version that does not support the new ISA spec.

The ESP RISC-V binutils that I see in riotbuild is 2.39.

Also: I wonder if this is rather a GDB bug

Yes, this is a bug in GNU binutils. The ELF files should be rv32i2p0 but the .riscv.attributes sections say they are rv32i2p1 (though I do not expect this to get fixed). GDB is behaving correctly in this case.

Hence: Maybe adding explicitly an option to GDB to use the older spec in our make debug command would be the better option?

I am not aware of any GDB options that allow you to override the .riscv.attributes.

I thought that instead of adding more code to work around issues with -misa-spec=2.2, this is an opportunity to start removing it.