Re: MPS codegen

["Gerd Möllmann" <gerd.moellmann@gmail.com>]

Gerd Möllmann <gerd.moellmann@gmail.com> writes:

> Helmut Eller <eller.helmut@gmail.com> writes:
>
>> On Sat, Jun 15 2024, Gerd Möllmann wrote:
>>
>>> I can't munmap because that
>>> would write back the changes done by the reloc phase back to the .pdmp
>>> file.
>>
>> I think that's a misunderstanding.  dump_map_file_posix sets MAP_PRIVATE
>> for DUMP_MEMORY_ACCESS_READWRITE.  So whatever changes the reloc phase
>> makes, they are written to copy-on-write pages and never to disk.
>
> Hm, I can try that a bit later. So, the theory would be that mprotect
> PROT_NONE does not work somehow on madvice'd not needed memory?  

Indeed, after this change

modified   src/pdumper.c
@@ -4931,9 +4931,10 @@ dump_discard_mem (void *mem, size_t size)
       (void) posix_madvise (mem, size, POSIX_MADV_DONTNEED);
 # elif defined HAVE_MADVISE
       (void) madvise (mem, size, MADV_DONTNEED);
-#endif
+#  endif
       /* Release the commit charge for the mapping.  */
-      (void) mprotect (mem, size, PROT_NONE);
+      int rc = mprotect (mem, size, PROT_NONE);
+      eassert (rc == 0);
 #endif
 }

I get

  pdumper.c:4937: Emacs fatal error: assertion failed: rc == 0
  gmake[3]: *** [Makefile:335: emacs-lisp/byte-opt.elc] Abort trap: 6
  gmake[3]: *** Waiting for unfinished jobs....

which is soothing.

I'm reluctant to invest more time though because this all si obsolete
with your change.