Avoid to wrongly allocate memory in map_file()

[RinHizakura]

In the case that HAVE_MMAP is 0, the emulator crashes.

$ make ENABLE_SYSTEM=1 system
rv32emu: src/emulate.c:626: block_translate: Assertion `insn' failed.
Aborted

The map_file() logic when HAVE_MMAP == 0 is weird as it allocates extra resources on the emulated memory. After this change, the aborting issue is gone, so I believe the original is not the correct implementation.

[RinHizakura]

BTW, I guess that we don't have coverage for HAVE_MMAP = 0. Do we need to have it?

[jserv]

Benchmarks

Benchmark suite	Current: f554a00	Previous: cb5eb10	Ratio
Dhrystone	1287 Average DMIPS over 10 runs	1328 Average DMIPS over 10 runs	1.03
Coremark	962.761 Average iterations/sec over 10 runs	964.938 Average iterations/sec over 10 runs	1.00

This comment was automatically generated by workflow using github-action-benchmark.

[vacantron]

Please refine the commit message that might contain "avoid reallocating memory" (or something like that) to indicate your change.

[jserv]

BTW, I guess that we don't have coverage for HAVE_MMAP = 0. Do we need to have it?

The condition HAVE_MMAP=0 is useful when profiling, as it is not straightforward to measure the system memory consumption when mmap is heavily used during emulation.

[ChinYikMing]

@RinHizakura Thanks for the fix. It indeed an implementation bug.

[RinHizakura]

The condition HAVE_MMAP=0 is useful when profiling, as it is not straightforward to measure the system memory consumption when mmap is heavily used during emulation.

Sorry for the unclear question, I was asking about the CI/CD coverage in fact.

[jserv]

I was asking about the CI/CD coverage in fact.

You can simply skip the condition HAVE_MMAP=0 in CI/CD pipeline.

[jserv]

Breakage reported by CI:

src/jit.c:1760:17: warning: ‘do_sret’ defined but not used [-Wunused-function]
 1760 |     static void do_##inst(struct jit_state *state UNUSED, riscv_t *rv UNUSED, \
      |                 ^~~
src/rv32_jit.c:355:1: note: in expansion of macro ‘GEN’
  355 | GEN(sret, {assert(NULL);
      | ^~~

[RinHizakura]

Breakage reported by CI:

This is not related to this change. The same warning can be found in other CI/CD jobs https://github.com/sysprog21/rv32emu/actions/runs/12540405349/job/34967656388 even if it passes. Not sure if any issues on the CI/CD justify this as a fail.

[jserv]

Thank @RinHizakura for contributing!