Re: sbrk question?

[Roman dot Pollak at Sun dot COM]

Jeff,
what you are saying is, when I do something like :

sbkr( size)
{
 return my_osmalloc(size);
}

this will not work, is this correct?

what is not clear to me from malloc_extend_top, what exactly happens,
when I use code as above and
somebody issue sbrk from different side.

It is gcc/newlib 1.12 chain on a 68020 (selfmade). The os is a very
striped down version of linux (just filesystem) no kernel
stuff,network,and others. However it fits in to 128K flash :)

However thank you for the answer

Roman



Jeff Johnston wrote On 03/16/06 21:08,:

>Roman.Pollak@Sun.COM wrote:
>  
>
>>Hi,
>>
>>I try to port newlib to a self made OS and run in to problems.
>>So far I didn't find the real problem, since it overwrites structures
>>and crashes when uses stdio functions.
>>Could it be that the malloc expects to have a continues sbrk return values?
>>    
>>
>
>Malloc is expecting sbrk to behave like sbrk is supposed to.  It asks 
>for an increment and expects the return address to increase.  If it does 
>not increase as expected, it assumes that something else is issuing sbrk 
>calls.  Take a look at malloc_extend_top in libc/stdlib/mallocr.c.
>
>If that is not your problem, you may want to look at where the stack and 
>heap are defined.  If they are colliding during your application and you 
>aren't catching this, random behavior will occur.  I can't comment on 
>your platform as you haven't given any details.  Make sure your crt0 is 
>properly clearing the .bss section and initializing the stack pointer 
>(usually top of data area and above _end linker symbol which is used as 
>start of heap).
>
>  
>
>>Or may I alloc memory from a different place (such as kernel and
>>newlib), without running in to trouble?
>>
>>    
>>
>
>You can create your own malloc family set of routines if you want. 
>Routines that use malloc/calloc/realloc don't have any expectation of 
>the memory address return value (other than non-null).  Remember that 
>newlib internally uses _r versions of a number of malloc-family 
>routines.  These versions take a reentrant structure pointer which is 
>needed to properly set errno.
>
>-- Jeff J.
>
>  
>