Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Johannes Weiner | 57cfc29 | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 2 | * bootmem - A boot-time physical memory allocator and configurator |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * Copyright (C) 1999 Ingo Molnar |
Johannes Weiner | 57cfc29 | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 5 | * 1999 Kanoj Sarcar, SGI |
| 6 | * 2008 Johannes Weiner |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * |
Johannes Weiner | 57cfc29 | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 8 | * Access to this subsystem has to be serialized externally (which is true |
| 9 | * for the boot process anyway). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | #include <linux/init.h> |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 12 | #include <linux/pfn.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 13 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/bootmem.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | #include <linux/module.h> |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 16 | #include <linux/kmemleak.h> |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 17 | #include <linux/range.h> |
Yinghai Lu | 72d7c3b | 2010-08-25 13:39:17 -0700 | [diff] [blame] | 18 | #include <linux/memblock.h> |
Franck Bui-Huu | e786e86 | 2006-09-25 23:31:06 -0700 | [diff] [blame] | 19 | |
| 20 | #include <asm/bug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <asm/io.h> |
Heiko Carstens | dfd54cb | 2006-09-25 23:31:33 -0700 | [diff] [blame] | 22 | #include <asm/processor.h> |
Franck Bui-Huu | e786e86 | 2006-09-25 23:31:06 -0700 | [diff] [blame] | 23 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "internal.h" |
| 25 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | unsigned long max_low_pfn; |
| 27 | unsigned long min_low_pfn; |
| 28 | unsigned long max_pfn; |
| 29 | |
Vivek Goyal | 92aa63a | 2005-06-25 14:58:18 -0700 | [diff] [blame] | 30 | #ifdef CONFIG_CRASH_DUMP |
| 31 | /* |
| 32 | * If we have booted due to a crash, max_pfn will be a very low value. We need |
| 33 | * to know the amount of memory that the previous kernel used. |
| 34 | */ |
| 35 | unsigned long saved_max_pfn; |
| 36 | #endif |
| 37 | |
Johannes Weiner | b61bfa3 | 2008-07-23 21:26:55 -0700 | [diff] [blame] | 38 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
| 39 | |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 40 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
| 41 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 42 | static int bootmem_debug; |
| 43 | |
| 44 | static int __init bootmem_debug_setup(char *buf) |
| 45 | { |
| 46 | bootmem_debug = 1; |
| 47 | return 0; |
| 48 | } |
| 49 | early_param("bootmem_debug", bootmem_debug_setup); |
| 50 | |
| 51 | #define bdebug(fmt, args...) ({ \ |
| 52 | if (unlikely(bootmem_debug)) \ |
| 53 | printk(KERN_INFO \ |
| 54 | "bootmem::%s " fmt, \ |
Harvey Harrison | 80a914d | 2008-10-15 22:01:25 -0700 | [diff] [blame] | 55 | __func__, ## args); \ |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 56 | }) |
| 57 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 58 | static unsigned long __init bootmap_bytes(unsigned long pages) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 59 | { |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 60 | unsigned long bytes = (pages + 7) / 8; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 61 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 62 | return ALIGN(bytes, sizeof(long)); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 63 | } |
| 64 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 65 | /** |
| 66 | * bootmem_bootmap_pages - calculate bitmap size in pages |
| 67 | * @pages: number of pages the bitmap has to represent |
| 68 | */ |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 69 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | { |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 71 | unsigned long bytes = bootmap_bytes(pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 73 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | } |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 75 | |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 76 | /* |
| 77 | * link bdata in order |
| 78 | */ |
Franck Bui-Huu | 69d49e6 | 2006-09-25 23:31:04 -0700 | [diff] [blame] | 79 | static void __init link_bootmem(bootmem_data_t *bdata) |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 80 | { |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 81 | struct list_head *iter; |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 82 | |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 83 | list_for_each(iter, &bdata_list) { |
| 84 | bootmem_data_t *ent; |
| 85 | |
| 86 | ent = list_entry(iter, bootmem_data_t, list); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 87 | if (bdata->node_min_pfn < ent->node_min_pfn) |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 88 | break; |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 89 | } |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 90 | list_add_tail(&bdata->list, iter); |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 91 | } |
| 92 | |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 93 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | * Called once to set up the allocator itself. |
| 95 | */ |
Johannes Weiner | 8ae0446 | 2008-07-23 21:26:59 -0700 | [diff] [blame] | 96 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | unsigned long mapstart, unsigned long start, unsigned long end) |
| 98 | { |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 99 | unsigned long mapsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | |
Mel Gorman | 2dbb51c | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 101 | mminit_validate_memmodel_limits(&start, &end); |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 102 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 103 | bdata->node_min_pfn = start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | bdata->node_low_pfn = end; |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 105 | link_bootmem(bdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | |
| 107 | /* |
| 108 | * Initially all pages are reserved - setup_arch() has to |
| 109 | * register free RAM areas explicitly. |
| 110 | */ |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 111 | mapsize = bootmap_bytes(end - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
| 113 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 114 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
| 115 | bdata - bootmem_node_data, start, mapstart, end, mapsize); |
| 116 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | return mapsize; |
| 118 | } |
| 119 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 120 | /** |
| 121 | * init_bootmem_node - register a node as boot memory |
| 122 | * @pgdat: node to register |
| 123 | * @freepfn: pfn where the bitmap for this node is to be placed |
| 124 | * @startpfn: first pfn on the node |
| 125 | * @endpfn: first pfn after the node |
| 126 | * |
| 127 | * Returns the number of bytes needed to hold the bitmap for this node. |
| 128 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 129 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
| 130 | unsigned long startpfn, unsigned long endpfn) |
| 131 | { |
| 132 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); |
| 133 | } |
| 134 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 135 | /** |
| 136 | * init_bootmem - register boot memory |
| 137 | * @start: pfn where the bitmap is to be placed |
| 138 | * @pages: number of available physical pages |
| 139 | * |
| 140 | * Returns the number of bytes needed to hold the bitmap. |
| 141 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 142 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
| 143 | { |
| 144 | max_low_pfn = pages; |
| 145 | min_low_pfn = start; |
| 146 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); |
| 147 | } |
Yinghai Lu | 0932587 | 2011-02-24 14:43:05 +0100 | [diff] [blame^] | 148 | |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 149 | /* |
| 150 | * free_bootmem_late - free bootmem pages directly to page allocator |
| 151 | * @addr: starting address of the range |
| 152 | * @size: size of the range in bytes |
| 153 | * |
| 154 | * This is only useful when the bootmem allocator has already been torn |
| 155 | * down, but we are still initializing the system. Pages are given directly |
| 156 | * to the page allocator, no bootmem metadata is updated because it is gone. |
| 157 | */ |
| 158 | void __init free_bootmem_late(unsigned long addr, unsigned long size) |
| 159 | { |
| 160 | unsigned long cursor, end; |
| 161 | |
| 162 | kmemleak_free_part(__va(addr), size); |
| 163 | |
| 164 | cursor = PFN_UP(addr); |
| 165 | end = PFN_DOWN(addr + size); |
| 166 | |
| 167 | for (; cursor < end; cursor++) { |
| 168 | __free_pages_bootmem(pfn_to_page(cursor), 0); |
| 169 | totalram_pages++; |
| 170 | } |
| 171 | } |
| 172 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 173 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
| 174 | { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 175 | int aligned; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 176 | struct page *page; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 177 | unsigned long start, end, pages, count = 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 178 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 179 | if (!bdata->node_bootmem_map) |
| 180 | return 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 181 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 182 | start = bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 183 | end = bdata->node_low_pfn; |
| 184 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 185 | /* |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 186 | * If the start is aligned to the machines wordsize, we might |
| 187 | * be able to free pages in bulks of that order. |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 188 | */ |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 189 | aligned = !(start & (BITS_PER_LONG - 1)); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 190 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 191 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
| 192 | bdata - bootmem_node_data, start, end, aligned); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 193 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 194 | while (start < end) { |
| 195 | unsigned long *map, idx, vec; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 196 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 197 | map = bdata->node_bootmem_map; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 198 | idx = start - bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 199 | vec = ~map[idx / BITS_PER_LONG]; |
| 200 | |
| 201 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { |
| 202 | int order = ilog2(BITS_PER_LONG); |
| 203 | |
| 204 | __free_pages_bootmem(pfn_to_page(start), order); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 205 | count += BITS_PER_LONG; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 206 | } else { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 207 | unsigned long off = 0; |
| 208 | |
| 209 | while (vec && off < BITS_PER_LONG) { |
| 210 | if (vec & 1) { |
| 211 | page = pfn_to_page(start + off); |
| 212 | __free_pages_bootmem(page, 0); |
| 213 | count++; |
| 214 | } |
| 215 | vec >>= 1; |
| 216 | off++; |
| 217 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 218 | } |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 219 | start += BITS_PER_LONG; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 220 | } |
| 221 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 222 | page = virt_to_page(bdata->node_bootmem_map); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 223 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 224 | pages = bootmem_bootmap_pages(pages); |
| 225 | count += pages; |
| 226 | while (pages--) |
| 227 | __free_pages_bootmem(page++, 0); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 228 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 229 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
| 230 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 231 | return count; |
| 232 | } |
| 233 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 234 | /** |
| 235 | * free_all_bootmem_node - release a node's free pages to the buddy allocator |
| 236 | * @pgdat: node to be released |
| 237 | * |
| 238 | * Returns the number of pages actually released. |
| 239 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 240 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
| 241 | { |
| 242 | register_page_bootmem_info_node(pgdat); |
| 243 | return free_all_bootmem_core(pgdat->bdata); |
| 244 | } |
| 245 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 246 | /** |
| 247 | * free_all_bootmem - release free pages to the buddy allocator |
| 248 | * |
| 249 | * Returns the number of pages actually released. |
| 250 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 251 | unsigned long __init free_all_bootmem(void) |
| 252 | { |
Yinghai Lu | aa235fc | 2010-03-31 20:45:27 -0700 | [diff] [blame] | 253 | unsigned long total_pages = 0; |
| 254 | bootmem_data_t *bdata; |
| 255 | |
| 256 | list_for_each_entry(bdata, &bdata_list, list) |
| 257 | total_pages += free_all_bootmem_core(bdata); |
| 258 | |
| 259 | return total_pages; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 260 | } |
| 261 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 262 | static void __init __free(bootmem_data_t *bdata, |
| 263 | unsigned long sidx, unsigned long eidx) |
| 264 | { |
| 265 | unsigned long idx; |
| 266 | |
| 267 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 268 | sidx + bdata->node_min_pfn, |
| 269 | eidx + bdata->node_min_pfn); |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 270 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 271 | if (bdata->hint_idx > sidx) |
| 272 | bdata->hint_idx = sidx; |
| 273 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 274 | for (idx = sidx; idx < eidx; idx++) |
| 275 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) |
| 276 | BUG(); |
| 277 | } |
| 278 | |
| 279 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, |
| 280 | unsigned long eidx, int flags) |
| 281 | { |
| 282 | unsigned long idx; |
| 283 | int exclusive = flags & BOOTMEM_EXCLUSIVE; |
| 284 | |
| 285 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", |
| 286 | bdata - bootmem_node_data, |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 287 | sidx + bdata->node_min_pfn, |
| 288 | eidx + bdata->node_min_pfn, |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 289 | flags); |
| 290 | |
| 291 | for (idx = sidx; idx < eidx; idx++) |
| 292 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { |
| 293 | if (exclusive) { |
| 294 | __free(bdata, sidx, idx); |
| 295 | return -EBUSY; |
| 296 | } |
| 297 | bdebug("silent double reserve of PFN %lx\n", |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 298 | idx + bdata->node_min_pfn); |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 299 | } |
| 300 | return 0; |
| 301 | } |
| 302 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 303 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
| 304 | unsigned long start, unsigned long end, |
| 305 | int reserve, int flags) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 306 | { |
| 307 | unsigned long sidx, eidx; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 308 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 309 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
| 310 | bdata - bootmem_node_data, start, end, reserve, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 311 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 312 | BUG_ON(start < bdata->node_min_pfn); |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 313 | BUG_ON(end > bdata->node_low_pfn); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 314 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 315 | sidx = start - bdata->node_min_pfn; |
| 316 | eidx = end - bdata->node_min_pfn; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 317 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 318 | if (reserve) |
| 319 | return __reserve(bdata, sidx, eidx, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 320 | else |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 321 | __free(bdata, sidx, eidx); |
| 322 | return 0; |
| 323 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 324 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 325 | static int __init mark_bootmem(unsigned long start, unsigned long end, |
| 326 | int reserve, int flags) |
| 327 | { |
| 328 | unsigned long pos; |
| 329 | bootmem_data_t *bdata; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 330 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 331 | pos = start; |
| 332 | list_for_each_entry(bdata, &bdata_list, list) { |
| 333 | int err; |
| 334 | unsigned long max; |
| 335 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 336 | if (pos < bdata->node_min_pfn || |
| 337 | pos >= bdata->node_low_pfn) { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 338 | BUG_ON(pos != start); |
| 339 | continue; |
| 340 | } |
| 341 | |
| 342 | max = min(bdata->node_low_pfn, end); |
| 343 | |
| 344 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
| 345 | if (reserve && err) { |
| 346 | mark_bootmem(start, pos, 0, 0); |
| 347 | return err; |
| 348 | } |
| 349 | |
| 350 | if (max == end) |
| 351 | return 0; |
| 352 | pos = bdata->node_low_pfn; |
| 353 | } |
| 354 | BUG(); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 355 | } |
| 356 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 357 | /** |
| 358 | * free_bootmem_node - mark a page range as usable |
| 359 | * @pgdat: node the range resides on |
| 360 | * @physaddr: starting address of the range |
| 361 | * @size: size of the range in bytes |
| 362 | * |
| 363 | * Partial pages will be considered reserved and left as they are. |
| 364 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 365 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 366 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 367 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 368 | unsigned long size) |
| 369 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 370 | unsigned long start, end; |
| 371 | |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 372 | kmemleak_free_part(__va(physaddr), size); |
| 373 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 374 | start = PFN_UP(physaddr); |
| 375 | end = PFN_DOWN(physaddr + size); |
| 376 | |
| 377 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 378 | } |
| 379 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 380 | /** |
| 381 | * free_bootmem - mark a page range as usable |
| 382 | * @addr: starting address of the range |
| 383 | * @size: size of the range in bytes |
| 384 | * |
| 385 | * Partial pages will be considered reserved and left as they are. |
| 386 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 387 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 388 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 389 | void __init free_bootmem(unsigned long addr, unsigned long size) |
| 390 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 391 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 392 | |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 393 | kmemleak_free_part(__va(addr), size); |
| 394 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 395 | start = PFN_UP(addr); |
| 396 | end = PFN_DOWN(addr + size); |
Yinghai Lu | a5645a6 | 2008-03-18 12:49:12 -0700 | [diff] [blame] | 397 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 398 | mark_bootmem(start, end, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | } |
| 400 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 401 | /** |
| 402 | * reserve_bootmem_node - mark a page range as reserved |
| 403 | * @pgdat: node the range resides on |
| 404 | * @physaddr: starting address of the range |
| 405 | * @size: size of the range in bytes |
| 406 | * @flags: reservation flags (see linux/bootmem.h) |
| 407 | * |
| 408 | * Partial pages will be reserved. |
| 409 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 410 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 411 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 412 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 413 | unsigned long size, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 415 | unsigned long start, end; |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 416 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 417 | start = PFN_DOWN(physaddr); |
| 418 | end = PFN_UP(physaddr + size); |
| 419 | |
| 420 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | } |
| 422 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 423 | /** |
| 424 | * reserve_bootmem - mark a page range as usable |
| 425 | * @addr: starting address of the range |
| 426 | * @size: size of the range in bytes |
| 427 | * @flags: reservation flags (see linux/bootmem.h) |
| 428 | * |
| 429 | * Partial pages will be reserved. |
| 430 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 431 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 432 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 433 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
| 434 | int flags) |
| 435 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 436 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 437 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 438 | start = PFN_DOWN(addr); |
| 439 | end = PFN_UP(addr + size); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 440 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 441 | return mark_bootmem(start, end, 1, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 442 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 443 | |
Yinghai Lu | f88eff7 | 2010-08-25 13:39:15 -0700 | [diff] [blame] | 444 | int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
| 445 | int flags) |
| 446 | { |
| 447 | return reserve_bootmem(phys, len, flags); |
| 448 | } |
| 449 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 450 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
| 451 | unsigned long idx, unsigned long step) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 452 | { |
| 453 | unsigned long base = bdata->node_min_pfn; |
| 454 | |
| 455 | /* |
| 456 | * Align the index with respect to the node start so that the |
| 457 | * combination of both satisfies the requested alignment. |
| 458 | */ |
| 459 | |
| 460 | return ALIGN(base + idx, step) - base; |
| 461 | } |
| 462 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 463 | static unsigned long __init align_off(struct bootmem_data *bdata, |
| 464 | unsigned long off, unsigned long align) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 465 | { |
| 466 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); |
| 467 | |
| 468 | /* Same as align_idx for byte offsets */ |
| 469 | |
| 470 | return ALIGN(base + off, align) - base; |
| 471 | } |
| 472 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 473 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
| 474 | unsigned long size, unsigned long align, |
| 475 | unsigned long goal, unsigned long limit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 477 | unsigned long fallback = 0; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 478 | unsigned long min, max, start, sidx, midx, step; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | |
Johannes Weiner | 594fe1a | 2009-01-06 14:40:32 -0800 | [diff] [blame] | 480 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
| 481 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, |
| 482 | align, goal, limit); |
| 483 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 484 | BUG_ON(!size); |
| 485 | BUG_ON(align & (align - 1)); |
| 486 | BUG_ON(limit && goal + size > limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | |
Christian Krafft | 7c309a6 | 2006-12-06 20:32:41 -0800 | [diff] [blame] | 488 | if (!bdata->node_bootmem_map) |
| 489 | return NULL; |
| 490 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 491 | min = bdata->node_min_pfn; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 492 | max = bdata->node_low_pfn; |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 493 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 494 | goal >>= PAGE_SHIFT; |
| 495 | limit >>= PAGE_SHIFT; |
| 496 | |
| 497 | if (limit && max > limit) |
| 498 | max = limit; |
| 499 | if (max <= min) |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 500 | return NULL; |
| 501 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 502 | step = max(align >> PAGE_SHIFT, 1UL); |
Yasunori Goto | 281dd25 | 2005-10-19 15:52:18 -0700 | [diff] [blame] | 503 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 504 | if (goal && min < goal && goal < max) |
| 505 | start = ALIGN(goal, step); |
| 506 | else |
| 507 | start = ALIGN(min, step); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 509 | sidx = start - bdata->node_min_pfn; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 510 | midx = max - bdata->node_min_pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 512 | if (bdata->hint_idx > sidx) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 513 | /* |
| 514 | * Handle the valid case of sidx being zero and still |
| 515 | * catch the fallback below. |
| 516 | */ |
| 517 | fallback = sidx + 1; |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 518 | sidx = align_idx(bdata, bdata->hint_idx, step); |
Yinghai Lu | ad09315 | 2008-03-10 23:23:42 -0700 | [diff] [blame] | 519 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 521 | while (1) { |
| 522 | int merge; |
| 523 | void *region; |
| 524 | unsigned long eidx, i, start_off, end_off; |
| 525 | find_block: |
| 526 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 527 | sidx = align_idx(bdata, sidx, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 528 | eidx = sidx + PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 530 | if (sidx >= midx || eidx > midx) |
Haren Myneni | 66d43e9 | 2005-12-12 00:37:39 -0800 | [diff] [blame] | 531 | break; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 532 | |
| 533 | for (i = sidx; i < eidx; i++) |
| 534 | if (test_bit(i, bdata->node_bootmem_map)) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 535 | sidx = align_idx(bdata, i, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 536 | if (sidx == i) |
| 537 | sidx += step; |
| 538 | goto find_block; |
| 539 | } |
| 540 | |
Mikulas Patocka | 627240a | 2008-08-15 00:40:17 -0700 | [diff] [blame] | 541 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 542 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 543 | start_off = align_off(bdata, bdata->last_end_off, align); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 544 | else |
| 545 | start_off = PFN_PHYS(sidx); |
| 546 | |
| 547 | merge = PFN_DOWN(start_off) < sidx; |
| 548 | end_off = start_off + size; |
| 549 | |
| 550 | bdata->last_end_off = end_off; |
| 551 | bdata->hint_idx = PFN_UP(end_off); |
| 552 | |
| 553 | /* |
| 554 | * Reserve the area now: |
| 555 | */ |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 556 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
| 557 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) |
| 558 | BUG(); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 559 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 560 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
| 561 | start_off); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 562 | memset(region, 0, size); |
Catalin Marinas | 008139d | 2009-08-27 14:29:17 +0100 | [diff] [blame] | 563 | /* |
| 564 | * The min_count is set to 0 so that bootmem allocated blocks |
| 565 | * are never reported as leaks. |
| 566 | */ |
| 567 | kmemleak_alloc(region, size, 0, 0); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 568 | return region; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | } |
| 570 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 571 | if (fallback) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 572 | sidx = align_idx(bdata, fallback - 1, step); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 573 | fallback = 0; |
| 574 | goto find_block; |
| 575 | } |
| 576 | |
| 577 | return NULL; |
| 578 | } |
| 579 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 580 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
| 581 | unsigned long size, unsigned long align, |
| 582 | unsigned long goal, unsigned long limit) |
| 583 | { |
Pekka Enberg | 441c7e0 | 2009-06-10 20:05:53 +0300 | [diff] [blame] | 584 | if (WARN_ON_ONCE(slab_is_available())) |
| 585 | return kzalloc(size, GFP_NOWAIT); |
| 586 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 587 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
Joe Perches | 433f13a | 2009-06-18 16:48:58 -0700 | [diff] [blame] | 588 | { |
| 589 | bootmem_data_t *p_bdata; |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 590 | |
Joe Perches | 433f13a | 2009-06-18 16:48:58 -0700 | [diff] [blame] | 591 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, |
| 592 | goal, limit); |
| 593 | if (p_bdata) |
| 594 | return alloc_bootmem_core(p_bdata, size, align, |
| 595 | goal, limit); |
| 596 | } |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 597 | #endif |
| 598 | return NULL; |
| 599 | } |
| 600 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 601 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
| 602 | unsigned long align, |
| 603 | unsigned long goal, |
| 604 | unsigned long limit) |
| 605 | { |
| 606 | bootmem_data_t *bdata; |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 607 | void *region; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 608 | |
| 609 | restart: |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 610 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
| 611 | if (region) |
| 612 | return region; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 613 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 614 | list_for_each_entry(bdata, &bdata_list, list) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 615 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
| 616 | continue; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 617 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 618 | break; |
| 619 | |
| 620 | region = alloc_bootmem_core(bdata, size, align, goal, limit); |
| 621 | if (region) |
| 622 | return region; |
| 623 | } |
| 624 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 625 | if (goal) { |
| 626 | goal = 0; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 627 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | } |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 629 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | } |
| 632 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 633 | /** |
| 634 | * __alloc_bootmem_nopanic - allocate boot memory without panicking |
| 635 | * @size: size of the request in bytes |
| 636 | * @align: alignment of the region |
| 637 | * @goal: preferred starting address of the region |
| 638 | * |
| 639 | * The goal is dropped if it can not be satisfied and the allocation will |
| 640 | * fall back to memory below @goal. |
| 641 | * |
| 642 | * Allocation may happen on any node in the system. |
| 643 | * |
| 644 | * Returns NULL on failure. |
| 645 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 646 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 647 | unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 649 | unsigned long limit = 0; |
| 650 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 651 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 652 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 653 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 654 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
| 655 | unsigned long goal, unsigned long limit) |
| 656 | { |
| 657 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 658 | |
| 659 | if (mem) |
| 660 | return mem; |
| 661 | /* |
| 662 | * Whoops, we cannot satisfy the allocation request. |
| 663 | */ |
| 664 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
| 665 | panic("Out of memory"); |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 666 | return NULL; |
| 667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 669 | /** |
| 670 | * __alloc_bootmem - allocate boot memory |
| 671 | * @size: size of the request in bytes |
| 672 | * @align: alignment of the region |
| 673 | * @goal: preferred starting address of the region |
| 674 | * |
| 675 | * The goal is dropped if it can not be satisfied and the allocation will |
| 676 | * fall back to memory below @goal. |
| 677 | * |
| 678 | * Allocation may happen on any node in the system. |
| 679 | * |
| 680 | * The function panics if the request can not be satisfied. |
| 681 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 682 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
| 683 | unsigned long goal) |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 684 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 685 | unsigned long limit = 0; |
| 686 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 687 | return ___alloc_bootmem(size, align, goal, limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | } |
| 689 | |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 690 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
| 691 | unsigned long size, unsigned long align, |
| 692 | unsigned long goal, unsigned long limit) |
| 693 | { |
| 694 | void *ptr; |
| 695 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 696 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
| 697 | if (ptr) |
| 698 | return ptr; |
| 699 | |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 700 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
| 701 | if (ptr) |
| 702 | return ptr; |
| 703 | |
| 704 | return ___alloc_bootmem(size, align, goal, limit); |
| 705 | } |
| 706 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 707 | /** |
| 708 | * __alloc_bootmem_node - allocate boot memory from a specific node |
| 709 | * @pgdat: node to allocate from |
| 710 | * @size: size of the request in bytes |
| 711 | * @align: alignment of the region |
| 712 | * @goal: preferred starting address of the region |
| 713 | * |
| 714 | * The goal is dropped if it can not be satisfied and the allocation will |
| 715 | * fall back to memory below @goal. |
| 716 | * |
| 717 | * Allocation may fall back to any node in the system if the specified node |
| 718 | * can not hold the requested memory. |
| 719 | * |
| 720 | * The function panics if the request can not be satisfied. |
| 721 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 722 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 723 | unsigned long align, unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 725 | if (WARN_ON_ONCE(slab_is_available())) |
| 726 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 727 | |
Yinghai Lu | 0932587 | 2011-02-24 14:43:05 +0100 | [diff] [blame^] | 728 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 729 | } |
| 730 | |
| 731 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, |
| 732 | unsigned long align, unsigned long goal) |
| 733 | { |
| 734 | #ifdef MAX_DMA32_PFN |
| 735 | unsigned long end_pfn; |
| 736 | |
| 737 | if (WARN_ON_ONCE(slab_is_available())) |
| 738 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 739 | |
| 740 | /* update goal according ...MAX_DMA32_PFN */ |
| 741 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 742 | |
| 743 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && |
| 744 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { |
| 745 | void *ptr; |
| 746 | unsigned long new_goal; |
| 747 | |
| 748 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 749 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, |
| 750 | new_goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 751 | if (ptr) |
| 752 | return ptr; |
| 753 | } |
| 754 | #endif |
| 755 | |
| 756 | return __alloc_bootmem_node(pgdat, size, align, goal); |
| 757 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | } |
| 759 | |
Yasunori Goto | e70260a | 2008-04-28 02:13:32 -0700 | [diff] [blame] | 760 | #ifdef CONFIG_SPARSEMEM |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 761 | /** |
| 762 | * alloc_bootmem_section - allocate boot memory from a specific section |
| 763 | * @size: size of the request in bytes |
| 764 | * @section_nr: sparse map section to allocate from |
| 765 | * |
| 766 | * Return NULL on failure. |
| 767 | */ |
Yasunori Goto | e70260a | 2008-04-28 02:13:32 -0700 | [diff] [blame] | 768 | void * __init alloc_bootmem_section(unsigned long size, |
| 769 | unsigned long section_nr) |
| 770 | { |
Johannes Weiner | 75a56cf | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 771 | bootmem_data_t *bdata; |
| 772 | unsigned long pfn, goal, limit; |
Yasunori Goto | e70260a | 2008-04-28 02:13:32 -0700 | [diff] [blame] | 773 | |
| 774 | pfn = section_nr_to_pfn(section_nr); |
Johannes Weiner | 75a56cf | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 775 | goal = pfn << PAGE_SHIFT; |
| 776 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; |
| 777 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; |
Yasunori Goto | e70260a | 2008-04-28 02:13:32 -0700 | [diff] [blame] | 778 | |
Johannes Weiner | 75a56cf | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 779 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
Yasunori Goto | e70260a | 2008-04-28 02:13:32 -0700 | [diff] [blame] | 780 | } |
| 781 | #endif |
| 782 | |
Andi Kleen | b54bbf7 | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 783 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
| 784 | unsigned long align, unsigned long goal) |
| 785 | { |
| 786 | void *ptr; |
| 787 | |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 788 | if (WARN_ON_ONCE(slab_is_available())) |
| 789 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 790 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 791 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
| 792 | if (ptr) |
| 793 | return ptr; |
| 794 | |
Andi Kleen | b54bbf7 | 2008-07-23 21:27:45 -0700 | [diff] [blame] | 795 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
| 796 | if (ptr) |
| 797 | return ptr; |
| 798 | |
| 799 | return __alloc_bootmem_nopanic(size, align, goal); |
| 800 | } |
| 801 | |
Heiko Carstens | dfd54cb | 2006-09-25 23:31:33 -0700 | [diff] [blame] | 802 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
| 803 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL |
| 804 | #endif |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 805 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 806 | /** |
| 807 | * __alloc_bootmem_low - allocate low boot memory |
| 808 | * @size: size of the request in bytes |
| 809 | * @align: alignment of the region |
| 810 | * @goal: preferred starting address of the region |
| 811 | * |
| 812 | * The goal is dropped if it can not be satisfied and the allocation will |
| 813 | * fall back to memory below @goal. |
| 814 | * |
| 815 | * Allocation may happen on any node in the system. |
| 816 | * |
| 817 | * The function panics if the request can not be satisfied. |
| 818 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 819 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
| 820 | unsigned long goal) |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 821 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 822 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 823 | } |
| 824 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 825 | /** |
| 826 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node |
| 827 | * @pgdat: node to allocate from |
| 828 | * @size: size of the request in bytes |
| 829 | * @align: alignment of the region |
| 830 | * @goal: preferred starting address of the region |
| 831 | * |
| 832 | * The goal is dropped if it can not be satisfied and the allocation will |
| 833 | * fall back to memory below @goal. |
| 834 | * |
| 835 | * Allocation may fall back to any node in the system if the specified node |
| 836 | * can not hold the requested memory. |
| 837 | * |
| 838 | * The function panics if the request can not be satisfied. |
| 839 | */ |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 840 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
| 841 | unsigned long align, unsigned long goal) |
| 842 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 843 | if (WARN_ON_ONCE(slab_is_available())) |
| 844 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 845 | |
Yinghai Lu | 0932587 | 2011-02-24 14:43:05 +0100 | [diff] [blame^] | 846 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
Yinghai Lu | b8ab9f8 | 2010-07-20 13:24:31 -0700 | [diff] [blame] | 847 | goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 848 | } |