Re: [LUG] Size of /proc/kcore

[Simon Waters <simon@xxxxxxxxxxxxxx>]

Simon Waters wrote:
> This is universally the size of memory, being as it is a mapping of 
> memory except.....
> 64bit AMD machines (virtual) have 128TB for /proc/kcore
>
> This seems wrong, and the nice folk in the LKML were discussing a 
> similar issue in June 2009.
As far as I can establish, this is a result of the vmalloc behaviour in 
64bit AMD architecture.
They deliberately place certain parts of address space memory a long 
long way from others in terms of numbers.
kcore is deliberately a core image (not simply as the documentation says 
size of physical RAM plus a header), and so it has to be appear to be 
large enough to accommodate the addressing scheme in use.
Moral - don't trust the documentation.

I think it was clear the original method of using kcore size to 
determine RAM size (which a Debian packaged script uses) was naive 
originally, and is now clearly wrong. They should use first line 
/proc/meminfo if they are relying on the /proc interface.
Hmm soon I will understand the AMD memory management... I'd rather not 
have to.




--
The Mailing List for the Devon & Cornwall LUG
http://mailman.dclug.org.uk/listinfo/list
FAQ: http://www.dcglug.org.uk/listfaq