problem with memory allocation

[joel at OARcorp.com (joel at OARcorp.com)]

On Tue, 3 Aug 1999, Emmanuel Raguet wrote:

> Hi Joel, hi everybody
> 
> During the port of RDBG on powerPc target, we have seen that a
> problem occured with the socket (unable to bind).
> 
> I have investigated and found it seems to be a problem with malloc.
> The returned value is in a memory space which is already allocated.
> 
> This happens during the MALLOC located in the function socreate
> (file uipc_socket.c). A pointer in a network stucture is corrupted 
> by the bzero because the returned address belongs to the structure.
> 
> I have several questions to understand better some mechanism to
> try to solve our problem :
> 
> 1 - Where does the memory allocated by a malloc come from :
>     the workspace or the heap ?
> 
>     My heap is from 0x60000 to 0x3f7ffff.
>     My workspace is from 3f80000 to the memory end.
>   
>     When rtems_bsdnet_malloc is called, the returned value is in
>     the heap.
>     When malloc is called directly (in my dec21140 driver, due to
>     undef malloc and free), the returned value is in the workspace.

The heap and workspace are separate entities.  rtems_bsdnet_malloc should
be based on malloc and both should return memory from the heap.


> 2 - How does the malloc mechanism work ?
> 
>     I have seen that the heap is seen as a region. To obtain memory,
>     a call to Region_Allocate_Segment is performed, which called
>     _Heap_Allocate. 
>     I have not well understood the Heap management (block, circular
>     buffers,... ?)    

It's OK. :)

malloc is built on top of Region which is built on top of a Heap.

Workspace is internal to RTEMS and is a heap directly.

> 3 - When is the workspace used ?

During system initialization, task/thread creation (stack and FP
contexts), and message buffer allocation.

> 4 - When is the heap used ?

The workspace is a heap -- so I assume that you mean the "C Program Heap"
aka malloc.  

The answer is any time the application allocates memory.

> FYI, all this code works well on pc386, our other target.

Which gives me a nice place to tell you waht is more than likely wrong. :)

The RTEMS Workspace and malloc heap are overlappign for the mcp750 BSP.

Check the calls to RTEMS_Malloc_Initialize( start, length, ...) and the
settings of Configuration.work_space_start and
Configuration.work_space_size.

Also I notice that the bspstart.c code indicates that INIT_STACK_SIZE and
INTR_STACK_SIZE somehow play into the memory layout on this BSP.  

My hunch is that these memory areas overlap.  When they do, all bets are
off.

--joel