Debugging memset crashes

[Dave Hylands]

Hi Praveen,

On Wed, Feb 23, 2011 at 11:58 PM, Praveen kumar <chatpravi at gmail.com> wrote:
> Hi,
> I have an embedded system on which i get crash (Caused by memset )  at long
> run( Reproducible 2/10 times ),
> I wanted to know efficient ways (materials) to handle memset crashes .
>
> Program terminated with signal 11, Segmentation fault.
>
> #0 memset (dstpp=0xd2, c=<value optimized out>, len=3374)
>     at ../../../src/Common/OS/linux/memset.c:20
> 20 ../../../src/Common/OS/linux/memset.c: No such file or directory.
> in ../../../src/Common/OS/linux/memset.c

The problem is probably that memset has been passed either a bad
pointer, or a bad size.

>From your backtrace, the dstpp of 0xd2 is almost certainly the
problem. The first 4K (and sometimes 64K) starting at location zero is
invalid. Passing a pointer in that range will cause an page fault (or
segmentation fault).

memset doesn't do any parameter checking, so if you want parameter
checking you'll need to do that yourself. You could add a function
like my_memest and use a macro to map memset to my_memset. Or you
could play with LD_PRELOAD (if your runtime library is in a shared
library) and intercept the calls that way.

http://stackoverflow.com/questions/69859/how-could-i-intercept-linux-sys-calls
http://developers.sun.com/solaris/articles/lib_interposers_code.html
http://www.jayconrod.com/cgi/view_post.py?23

Yet another technique is to use the linker's wrap capability
http://stackoverflow.com/questions/998464/function-interposition-in-linux-without-dlsym

Dave Hylands