On Tue, 2004-12-21 at 19:16, Hiroyuki KAMEZAWA wrote:
> I'm sorry for not understanding what is done in alloc_bootmem_node_remap().
> This one wiil meet current -mhp tree.
> this one uses HAVE_ARCH_ALLOC_REMAP and never modifies mm/page_alloc.c
This does look better, especially since it stays in arch/ia64. A few
tiny comments, though.
> diff -puN arch/ia64/mm/contig.c~contig3 arch/ia64/mm/contig.c
> --- test-kernel/arch/ia64/mm/contig.c~contig3 2004-12-22 12:04:45.639948552 +0900
> +++ test-kernel-kamezawa/arch/ia64/mm/contig.c 2004-12-22 12:04:45.648947184 +0900
> @@ -231,7 +231,6 @@ paging_init (void)
> unsigned long zones_size[MAX_NR_ZONES];
> #ifdef CONFIG_VIRTUAL_MEM_MAP
> unsigned long zholes_size[MAX_NR_ZONES];
> - unsigned long max_gap;
> #endif
>
> /* initialize mem_map[] */
> @@ -262,29 +261,9 @@ paging_init (void)
> (num_physpages - num_dma_physpages));
> }
> }
> + vmem_map = 0;
NULL, please.
> + free_area_init_node(0, &contig_page_data, zones_size, 0, zholes_size);
Please use NODE_DATA(0) instead of contig_page_data. We've been trying
to slowly replace all of those.
> +struct page *__init alloc_vmem_map(unsigned long size)
> +{
> +#ifndef CONFIG_DISCONTIGMEM
> + unsigned long max_gap;
> + efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
> + if (max_gap < LARGE_GAP) {
> + vmem_map = (struct page *)0;
Using NULL here will avoid the cast.
> + return alloc_bootmem(size);
> + }
> +#endif
It seems a little bit odd to me that the efi walk is done under
CONFIG_DISCONTIGMEM. I just question whether this is completely
necessary. But, I don't really care all that much about ia64. :)
> + size = PAGE_ALIGN(max_low_pfn * sizeof(struct page));
> + vmalloc_end -= size;
> + vmem_map = (struct page *)vmalloc_end;
> + efi_memmap_walk(create_mem_map_page_table, NULL);
> +
> + printk("Virtual mem_map starts at 0x%p\n", vmem_map);
> + return vmem_map;
> +}
> +
> +void * alloc_remap(int nid, unsigned long size)
> +{
> + struct page *lmem_map;
> + if (nid == 0)
> + return alloc_vmem_map(size);
> + return vmem_map + NODE_DATA(nid)->node_start_pfn;
> +}
Are my eyes just missing where lmem_map is used, or is it really unused
here?
It does seem a wee bit hackish to special-case node 0 to be the
effective initalization call, but I think I can live with it if that
other stuff is addressed.
-- Dave
