Re: [Lhms-devel] Re: Merging Nonlinear and Numa style memoryhotplug

[Dave Hansen]

Whoops.  Hit CTRL-Enter during my ASCII art :)

On Fri, 2004-06-25 at 13:45, Yasunori Goto wrote:
> > > Are you sure that all architectures need phys_section?
> > 
> > You don't *need* it, but the alternative is a scan of the mem_section[]
> > array, which would be much, much slower.
> > 
> > Do you have an idea for an alternate implementation?
> 
> I didn't find that scan of the mem_section[] is necessary.
> I thought just that mem_section index = phys_section index.
> May I ask why scan of mem_section is necessary?
> I might still have misunderstood something.

For now, the indexes happen to be the same.  However, for discontiguous
memory systems, this will not be the case

Consider a system with 3 GB of RAM, 2GB@0x00000000 and 1 GB@0xC0000000
with 1 GB sections.  The arrays would look like this:
 
mem | phys
----+-----
 0  |  0
 1  |  1
 2  |  3

See the B-lpfn patch that I posted today for why this is important.  It
basically allows us to represent sparse physical addresses in a much
more linear fashion.

-- Dave

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/