Crash with jemalloc 2.2.1

[Jason Evans]

On 04/18/2011 01:11 PM, Felipe Wilhelms Damasio - Taghos wrote:
> I'm using jemalloc to build a webcache system.
>
> I'm doing around 600-700 calloc/free per second. (a calloc-per-new-session).
>
> It's a linux machine with kernel 2.6.34 and glibc 2.11.2.
>
> jemalloc was compiled with:
>
>      ./configure  --enable-prof --enable-dss --enable-fill --enable-stats
> --enable-xmalloc --enable-swap
>
> I have no malloc options file.
>
> At some point, my program crashed. I generated a core dump and a gdb bt
> showed this:
>
> #0  arena_avail_comp (arena=0x1000a80, size=36864, large=true, zero=true) at
> src/arena.c:222
> 222        size_t b_size = b->bits&  ~PAGE_MASK;
> (gdb) bt
> #0  arena_avail_comp (arena=0x1000a80, size=36864, large=true, zero=true) at
> src/arena.c:222
> #1  arena_avail_tree_nsearch (arena=0x1000a80, size=36864, large=true,
> zero=true) at src/arena.c:249
> #2  arena_run_alloc (arena=0x1000a80, size=36864, large=true, zero=true) at
> src/arena.c:593
> #3  0x00007fe240427b20 in arena_malloc_large (arena=0x1000a80, size=36864,
> zero=80) at src/arena.c:1474
> #4  0x00007fe240422007 in icalloc (num=<value optimized out>, size=35472) at
> include/jemalloc/internal/jemalloc_internal.h:620
> #5  calloc (num=<value optimized out>, size=35472) at src/jemalloc.c:1185
> #6  0x0000000000408a76 in SES_new (fd=11191) at cache_session.c:11
> #7  0x000000000041563e in main (argc=3, argv=0x7fff357085c8) at server.c:2261
>
> Is this indication of a bug in jemalloc or in my program?
>
> What other info can I provide to help track this down?

This is likely an application bug, e.g. a double free.  I'd suggest 
testing with valgrind (without jemalloc linked in) if you can, or using 
a debug build of jemalloc if valgrind isn't an option.  There are 
assertions in debug builds of jemalloc that detect many (but not all) 
application memory bugs.

Thanks,
Jason