valgrind 3.8.1 warning in jemalloc 3.2

[Jason Evans]

On Dec 12, 2012, at 9:14 AM, Daniel Mezzatto Rodrigues <danielm at buscape-inc.com> wrote:
> Unfortunately I can't generate a simple test program that makes the valgring errors occur. Even inside my system the valgrind error messeges don't happen for every HTTP request.
> 
> Looks like only memory intense requests, that allocates 64 MB or more in a single chunk, triggers the errors. It also appears that it is related to reallocs. If I have a HTTP request that somehow generates more than 64 MB of data in a single chunk of memory and this chunck needs to me realloc'd, then valgring starts complaning.

Ah, this info narrows things down a bit.  Can you try configuring jemalloc with --enable-munmap and see if the problem persists?  It seems probable that the VALGRIND_MAKE_MEM_UNDEFINED() call in chunk_recycle() should be unconditional.

Thanks,
Jason