https://sourceware.org/bugzilla/show_bug.cgi?id=29292
Rainer Orth <ro at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |nick.alcock at oracle dot com
--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> ---
The issue persists up to binutils 2.44.90. Investigating some more, I can
confirm that disabling mmap (with ac_cv_func_mmap_fixed_mapped=no) fixes the
failures.
What I found is that after arc_mmap_unmap unmaps the CTF header, when it is
read back in ctf_link_write the buffer is all zeros.
I suspected that the msync (header, headersz, MS_ASYNC) call in
arc_mmap_writeout
might be a problem (buffer not yet synced to disk when it's read back using
fread), but changing that to MS_SYNC didn't make a difference.
I'll try to create a minimal testcase (just tmpfile, mmap, fill buffer, msync,
munmap, fread) to see if the problem reproduces there.
However, I very much wonder what's the advantage of this complicated
mmap/msync/munmap
dance instead of just writing the header to the tmp file and be done with it.
--
You are receiving this mail because:
You are on the CC list for the bug.
