From: Yinghai Lu vmemmap allocation current got [ffffe20000000000-ffffe200001fffff] PMD ->ffff810001400000 on node 0 [ffffe20000200000-ffffe200003fffff] PMD ->ffff810001800000 on node 0 [ffffe20000400000-ffffe200005fffff] PMD ->ffff810001c00000 on node 0 [ffffe20000600000-ffffe200007fffff] PMD ->ffff810002000000 on node 0 [ffffe20000800000-ffffe200009fffff] PMD ->ffff810002400000 on node 0 ... there is 2M hole between them. the rootcause is that usemap (24 bytes) will be allocated after every 2M mem_map. and it will push next vmemmap (2M) to next align (2M). solution: try to allocate mem_map continously. after patch, will get [ffffe20000000000-ffffe200001fffff] PMD ->ffff810001400000 on node 0 [ffffe20000200000-ffffe200003fffff] PMD ->ffff810001600000 on node 0 [ffffe20000400000-ffffe200005fffff] PMD ->ffff810001800000 on node 0 [ffffe20000600000-ffffe200007fffff] PMD ->ffff810001a00000 on node 0 [ffffe20000800000-ffffe200009fffff] PMD ->ffff810001c00000 on node 0 ... and usemap will share in page because of they are allocated continuously too. sparse_early_usemap_alloc: usemap = ffff810024e00000 size = 24 sparse_early_usemap_alloc: usemap = ffff810024e00080 size = 24 sparse_early_usemap_alloc: usemap = ffff810024e00100 size = 24 sparse_early_usemap_alloc: usemap = ffff810024e00180 size = 24 ... so we make the bootmem allocation more compact and use less memory for usemap. for power pc Badari Pulavarty wrote: > You have to call sparse_init_one_section() on each pmap and usemap > as we allocate - since valid_section() depends on it (which is needed > by vmemmap_populate() to check if the section is populated or not). > On ppc, we need to call htab_bolted_mapping() on each section and > we need to skip existing sections. so try to allocate usemap at first altogether. Signed-off-by: Yinghai Lu Cc: Andy Whitcroft Cc: Yasunori Goto Cc: Dave Hansen Cc: Bob Picco Cc: Christoph Lameter Cc: Ingo Molnar Cc: Badari Pulavarty Signed-off-by: Andrew Morton --- mm/sparse.c | 32 +++++++++++++++++++++++++++++--- 1 file changed, 29 insertions(+), 3 deletions(-) diff -puN mm/sparse.c~mm-make-mem_map-allocation-continuous-v2 mm/sparse.c --- a/mm/sparse.c~mm-make-mem_map-allocation-continuous-v2 +++ a/mm/sparse.c @@ -295,22 +295,48 @@ void __init sparse_init(void) unsigned long pnum; struct page *map; unsigned long *usemap; + unsigned long **usemap_map; + int size; + + /* + * map is using big page (aka 2M in x86 64 bit) + * usemap is less one page (aka 24 bytes) + * so alloc 2M (with 2M align) and 24 bytes in turn will + * make next 2M slip to one more 2M later. + * then in big system, the memory will have a lot of holes... + * here try to allocate 2M pages continously. + * + * powerpc need to call sparse_init_one_section right after each + * sparse_early_mem_map_alloc, so allocate usemap_map at first. + */ + size = sizeof(unsigned long *) * NR_MEM_SECTIONS; + usemap_map = alloc_bootmem(size); + if (!usemap_map) + panic("can not allocate usemap_map\n"); for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { if (!present_section_nr(pnum)) continue; + usemap_map[pnum] = sparse_early_usemap_alloc(pnum); + } - map = sparse_early_mem_map_alloc(pnum); - if (!map) + for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { + if (!present_section_nr(pnum)) continue; - usemap = sparse_early_usemap_alloc(pnum); + usemap = usemap_map[pnum]; if (!usemap) continue; + map = sparse_early_mem_map_alloc(pnum); + if (!map) + continue; + sparse_init_one_section(__nr_to_section(pnum), pnum, map, usemap); } + + free_bootmem(__pa(usemap_map), size); } #ifdef CONFIG_MEMORY_HOTPLUG _