> Hello,
> 
> OK, sorry. Let me try to clarify my question:
> 
> First, I'm using QNX 8.0.
> For a certification process, I need to prove that a given program won't 
> allocate any more memory after a given point: So, no more malloc() or 
> equivalent after a given point (End of initialization phase).
> Analyzing the code by code review could be an option, but I wanted to automate
>  this par inspecting this program from an external C application.

As I wrote, I don't believe that can work.

> I wrote a C routine which parses the  "as_stats.rss" field from "/proc/%d/
> vmstat" file.
> This returns me the numbers of pages mapped by the process, and I could 
> multiply rss * _SC_PAGESIZE to obtain the size of the memory mapped by the 
> process.
> But what I would really need is the size of the memory currently used by the 
> process.

There is no distinction from the POV of the OS between "mapped" and "in use"  (with the exception of lazy mapping, which
 I think you're not talking about, and which we currently don't support in QNX 8) by a process. Also, mappings happen on
 per-page basis, you can't go smaller than that.
 
> For example, if my process allocates some memory after the end of its 
> initialization phase, the OS can potentially use the remaining unused memory 
> in mapped pages, without having to map a new memory page.

This doesn't make much sense to me, sorry. Perhaps other can follow better.

> And, in this specific case, parsing "as_stats.rss" field will not indicate 
> that memory has been allocated after the init phase end.
> 
> Hoping this is clearer now.

erm... not really (to me at least)

Michael

> 
> Thanks.
> Regards
> Jean-Michel

> For example, if my process allocates some memory after the end of its 
> initialization phase, the OS can potentially use the remaining unused memory 
> in mapped pages, without having to map a new memory page.
> And, in this specific case, parsing "as_stats.rss" field will not indicate 
> that memory has been allocated after the init phase end.

If you want to say that a malloc() or free() call will not necessarily change the number of mapped pages, you are right.


What about a preload library with stubs for malloc() and free().
You could count them and check then number by dumping them in a signal handler.
Maybe you can also reset the counters in a signal handler after the init phase and check if new calls are made later.

Not sure however, if this is an acceptable approach for your certification process.