Fix memory protection in regression tests

[piotro888]

I found a bug in #466 - all memory segments were set to all RWX flags. After fixing that, regression tests started to fail.

First bug is simple - wrong flag on data memory (only in cocotb backend)
Second bug is problematic. In rv32uc-rvc test there is a data symbol and not .data (!) section that is embedded into instruction memory. This section is written by rvc tests, but not flagged as writeable in ELF file. I don't have currently solution to it (ignore for this specific test, or maybe data section can be added to linker script?)

[piotro888]

This data section is here. Does someone know how this can be added to linker script, so instruction LOAD segment would be also writeable?

[piotro888]

I decided to manually exclude tests from write protection. It will be also useful for tests with self-modifying code, which can't be detected for linker script.

[piotro888]

@tilk Can you change the https://github.com/orgs/kuznia-rdzeni/packages/container/package/riscv-toolchain package visibility from internal, please? If we are working on forks, it can't be pulled for manual run of CI benchmark workflow on forked branch.

[tilk]

Done.

[tilk]

Second bug is problematic. In rv32uc-rvc test there is a data symbol and not .data (!) section that is embedded into instruction memory.

If I understand correctly, this test has no .data section, and the data symbol is just for padding in the code section. I don't see why this should cause a problem. Unfortunately I was mistaken: there are store instructions to the region. This is not the only test with such behavior: there is also rv32ui-fence_i, which we don't run because we still don't support FENCEI. These two should probably be special-cased somewhere, as I believe a write-protected code section is a good sanity check for the other tests.

[Arusekk]

May be worth reporting upstream to the tests repo.