Re: 2.6.17-mm1

[Mel Gorman]

On Thu, 22 Jun 2006, Russell King wrote:

> On Thu, Jun 22, 2006 at 04:54:06PM +0100, Mel Gorman wrote:
> > On Thu, 22 Jun 2006, Franck Bui-Huu wrote:
> > > It's the default value (see memory_model.h). It means that pfn start
> > > for node 0 is 0, therefore your physical memory address starts at 0.
> >
> > I know, but what I'm getting at is that ARCH_PFN_OFFSET may be unnecessary
> > with flatmem-relax-requirement-for-memory-to-start-at-pfn-0.patch applied.
> > ARCH_PFN_OFFSET is used as
> >
> > #define page_to_pfn(page)       ((unsigned long)((page) - mem_map) + \
> >                                  ARCH_PFN_OFFSET)
> >
> > because it knew that the map may not start at PFN 0. With
> > flatmem-relax-requirement-for-memory-to-start-at-pfn-0.patch, the map will
> > start at PFN 0 even if physical memory does not start until later.
>
> Doesn't that result in a massive array of struct pages if your memory
> starts a 3GB physical and has 4K pages?

No, I should have been clear. The size of the map remains the same but 
mem_map does not point directly to NODE_DATA(0)->node_mem_map when the PFN 
of node 0 is not 0. Instead mem_map points to

NODE_DATA(0)->node_mem_map - NODE_DATA(0)->node_start_pfn

The relevant bit of code is

	map = alloc_remap(pgdat->node_id, size);
	if (!map)
		map = alloc_bootmem_node(pgdat, size);
	pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);

	/*
	 * With FLATMEM the global mem_map is used.  This is assumed
	 * to be based at pfn 0 such that 'pfn = page* - mem_map'
	 * is true. Adjust map relative to node_mem_map to
	 * maintain this relationship.
	 */
	map -= pgdat->node_start_pfn;

and later

	if (pgdat == NODE_DATA(0))
		mem_map = map;

So memory is wasted and ARCH_PFN_OFFSET isn't needed in the case where it 
is working around NODE_DATA(0)->node_start_pfn != 0

> If you have only 32MB in that
> scenario, and that was correct, you'd gobble 25MB of that just to
> store that array.  Ouch.

It would be a kick in the shins all right if that was the case.

--
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
-
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/