[Maverick] [ti-omap4] SRU: A workaround for highmem issue on OMAP4 platform

[Nicolas Pitre]

On Fri, 1 Oct 2010, Dechesne, Nicolas wrote:

> Nicolas
> 
> > Beware.  While the 2G:2G is a good workaround for the time being, it may
> > cause problems to user space applications expecting a larger vm space to
> > their disposal.  Good testing with large programs is recommended.  Or at
> > least add this caveat to the release notes.
> > 
> 
> my understanding is that with 2G/2G we don't loose anything compared to
> highmem until we have at least 2G of RAM. is that correct?

No, that's not correct.  User applications may use their address space 
to mmap() files or hardware peripheral memory in addition to standard 
RAM.  You don't need to have 2G of physical memory to exhaust the 
address space.  In fact, you may use up to 2G of virtual address space 
even if you have far less than 1G of RAM. Furthermore, the address space 
allocations are not all contiguous and you may get extra shortage from 
unused holes.

Affected applications are typically heavy GUI based ones, or the native 
compilation of large programs.

> For now the Ubuntu 10.10 on OMAP4 is targeting boards with 1G of RAM. 
> is there anything I am missing?

See above.  And note that everything might be just fine too.  But this 
deserves some testing to be sure.


Nicolas