This is the mail archive of the newlib@sourceware.org mailing list for the newlib project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: malloc crashes for values >3500


On 01/02/2013 05:29 PM, Martin Osterloh wrote:
Hi all,

I have a working newlib port for my x86_64 OS. I can do malloc() without
any problems if the size of the memory I request is less than ~3500. I
have enough memory available - far more than 3500.

I am just wondering if anyone of you guys had a hint? My sbrk
implementation is the one that gets shipped with newib:

char * sbrk(int nbytes);
{
   char        *base;

   if (!heap_ptr)
     heap_ptr = (char *)&_end;
   base = heap_ptr;
   heap_ptr += nbytes;

   return base;
}

My page size is 4096. One thought that comes to my mind is that I reach
the end of a page and I fail to allocate a new page?

I am keen to hear your opinions.

Thanks,
--Martin

The memory/sbrk model you are using assumes that the heap / stack is allotted a single piece of contiguous space at the start of the program with the heap growing up and the stack growing down within this space. If the heap + stack size go above the allotted space, you will have a collision (e.g. your malloc'd space is overwriting the stack or vice versa). If you need more storage, you are out of luck so you must make a good guess at the start.


So, you need to verify where the heap starts (&_end symbol), where the stack starts, (i.e. how much contiguous space has been allocated at the start of the program), and where the stack and heap pointers are when you have a crash. You can prevent a collision during a malloc by adding a check inside the sbrk routine, but if stack growth ultimately causes the collision (e.g. recursive routine), there isn't a way to prevent this since the stack change will just be a change to the stack pointer register and no checks are made in the generated code. You might be able to catch when it occurs using a debugger or something like SystemTap, but this will be very intrusive to your program's performance.

One possibility would be to try upping the amount of storage you start with. Another thing you could do is separate the stack and heap and replace sbrk with a more dynamic version that gets chunks of storage from your OS (if it is more sophisticated) when needed thereby allowing you to increase the heap as much as possible and avoid a heap/stack collision which you might not see the effects of right away.

-- Jeff J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]