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> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 15 | #include <linux/export.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 | |
Yinghai Lu | e782ab4 | 2011-02-24 14:43:06 +0100 | [diff] [blame] | 26 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
| 27 | struct pglist_data __refdata contig_page_data = { |
| 28 | .bdata = &bootmem_node_data[0] |
| 29 | }; |
| 30 | EXPORT_SYMBOL(contig_page_data); |
| 31 | #endif |
| 32 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | unsigned long max_low_pfn; |
| 34 | unsigned long min_low_pfn; |
| 35 | unsigned long max_pfn; |
| 36 | |
Johannes Weiner | b61bfa3 | 2008-07-23 21:26:55 -0700 | [diff] [blame] | 37 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
| 38 | |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 39 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
| 40 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 41 | static int bootmem_debug; |
| 42 | |
| 43 | static int __init bootmem_debug_setup(char *buf) |
| 44 | { |
| 45 | bootmem_debug = 1; |
| 46 | return 0; |
| 47 | } |
| 48 | early_param("bootmem_debug", bootmem_debug_setup); |
| 49 | |
| 50 | #define bdebug(fmt, args...) ({ \ |
| 51 | if (unlikely(bootmem_debug)) \ |
| 52 | printk(KERN_INFO \ |
| 53 | "bootmem::%s " fmt, \ |
Harvey Harrison | 80a914d | 2008-10-15 22:01:25 -0700 | [diff] [blame] | 54 | __func__, ## args); \ |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 55 | }) |
| 56 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 57 | static unsigned long __init bootmap_bytes(unsigned long pages) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 58 | { |
Uwe Kleine-König | 9571a98 | 2012-01-10 15:08:00 -0800 | [diff] [blame] | 59 | unsigned long bytes = DIV_ROUND_UP(pages, 8); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 60 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 61 | return ALIGN(bytes, sizeof(long)); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 62 | } |
| 63 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 64 | /** |
| 65 | * bootmem_bootmap_pages - calculate bitmap size in pages |
| 66 | * @pages: number of pages the bitmap has to represent |
| 67 | */ |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 68 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | { |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 70 | unsigned long bytes = bootmap_bytes(pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 72 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | } |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 74 | |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 75 | /* |
| 76 | * link bdata in order |
| 77 | */ |
Franck Bui-Huu | 69d49e6 | 2006-09-25 23:31:04 -0700 | [diff] [blame] | 78 | static void __init link_bootmem(bootmem_data_t *bdata) |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 79 | { |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 80 | bootmem_data_t *ent; |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 81 | |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 82 | list_for_each_entry(ent, &bdata_list, list) { |
| 83 | if (bdata->node_min_pfn < ent->node_min_pfn) { |
| 84 | list_add_tail(&bdata->list, &ent->list); |
| 85 | return; |
| 86 | } |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 87 | } |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 88 | |
| 89 | list_add_tail(&bdata->list, &bdata_list); |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 90 | } |
| 91 | |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 92 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | * Called once to set up the allocator itself. |
| 94 | */ |
Johannes Weiner | 8ae0446 | 2008-07-23 21:26:59 -0700 | [diff] [blame] | 95 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | unsigned long mapstart, unsigned long start, unsigned long end) |
| 97 | { |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 98 | unsigned long mapsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | |
Mel Gorman | 2dbb51c | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 100 | mminit_validate_memmodel_limits(&start, &end); |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 101 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 102 | bdata->node_min_pfn = start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | bdata->node_low_pfn = end; |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 104 | link_bootmem(bdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | |
| 106 | /* |
| 107 | * Initially all pages are reserved - setup_arch() has to |
| 108 | * register free RAM areas explicitly. |
| 109 | */ |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 110 | mapsize = bootmap_bytes(end - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
| 112 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 113 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
| 114 | bdata - bootmem_node_data, start, mapstart, end, mapsize); |
| 115 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | return mapsize; |
| 117 | } |
| 118 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 119 | /** |
| 120 | * init_bootmem_node - register a node as boot memory |
| 121 | * @pgdat: node to register |
| 122 | * @freepfn: pfn where the bitmap for this node is to be placed |
| 123 | * @startpfn: first pfn on the node |
| 124 | * @endpfn: first pfn after the node |
| 125 | * |
| 126 | * Returns the number of bytes needed to hold the bitmap for this node. |
| 127 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 128 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
| 129 | unsigned long startpfn, unsigned long endpfn) |
| 130 | { |
| 131 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); |
| 132 | } |
| 133 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 134 | /** |
| 135 | * init_bootmem - register boot memory |
| 136 | * @start: pfn where the bitmap is to be placed |
| 137 | * @pages: number of available physical pages |
| 138 | * |
| 139 | * Returns the number of bytes needed to hold the bitmap. |
| 140 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 141 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
| 142 | { |
| 143 | max_low_pfn = pages; |
| 144 | min_low_pfn = start; |
| 145 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); |
| 146 | } |
Yinghai Lu | 0932587 | 2011-02-24 14:43:05 +0100 | [diff] [blame] | 147 | |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 148 | /* |
| 149 | * free_bootmem_late - free bootmem pages directly to page allocator |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 150 | * @addr: starting physical address of the range |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 151 | * @size: size of the range in bytes |
| 152 | * |
| 153 | * This is only useful when the bootmem allocator has already been torn |
| 154 | * down, but we are still initializing the system. Pages are given directly |
| 155 | * to the page allocator, no bootmem metadata is updated because it is gone. |
| 156 | */ |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 157 | void __init free_bootmem_late(unsigned long physaddr, unsigned long size) |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 158 | { |
| 159 | unsigned long cursor, end; |
| 160 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 161 | kmemleak_free_part(__va(physaddr), size); |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 162 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 163 | cursor = PFN_UP(physaddr); |
| 164 | end = PFN_DOWN(physaddr + size); |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 165 | |
| 166 | for (; cursor < end; cursor++) { |
| 167 | __free_pages_bootmem(pfn_to_page(cursor), 0); |
| 168 | totalram_pages++; |
| 169 | } |
| 170 | } |
| 171 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 172 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
| 173 | { |
| 174 | struct page *page; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 175 | unsigned long start, end, pages, count = 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 176 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 177 | if (!bdata->node_bootmem_map) |
| 178 | return 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 179 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 180 | start = bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 181 | end = bdata->node_low_pfn; |
| 182 | |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 183 | bdebug("nid=%td start=%lx end=%lx\n", |
| 184 | bdata - bootmem_node_data, start, end); |
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 | while (start < end) { |
| 187 | unsigned long *map, idx, vec; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 188 | unsigned shift; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 189 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 190 | map = bdata->node_bootmem_map; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 191 | idx = start - bdata->node_min_pfn; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 192 | shift = idx & (BITS_PER_LONG - 1); |
| 193 | /* |
| 194 | * vec holds at most BITS_PER_LONG map bits, |
| 195 | * bit 0 corresponds to start. |
| 196 | */ |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 197 | vec = ~map[idx / BITS_PER_LONG]; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 198 | |
| 199 | if (shift) { |
| 200 | vec >>= shift; |
| 201 | if (end - start >= BITS_PER_LONG) |
| 202 | vec |= ~map[idx / BITS_PER_LONG + 1] << |
| 203 | (BITS_PER_LONG - shift); |
| 204 | } |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 205 | /* |
| 206 | * If we have a properly aligned and fully unreserved |
| 207 | * BITS_PER_LONG block of pages in front of us, free |
| 208 | * it in one go. |
| 209 | */ |
| 210 | if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 211 | int order = ilog2(BITS_PER_LONG); |
| 212 | |
| 213 | __free_pages_bootmem(pfn_to_page(start), order); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 214 | count += BITS_PER_LONG; |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 215 | start += BITS_PER_LONG; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 216 | } else { |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 217 | unsigned long cur = start; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 218 | |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 219 | start = ALIGN(start + 1, BITS_PER_LONG); |
| 220 | while (vec && cur != start) { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 221 | if (vec & 1) { |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 222 | page = pfn_to_page(cur); |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 223 | __free_pages_bootmem(page, 0); |
| 224 | count++; |
| 225 | } |
| 226 | vec >>= 1; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 227 | ++cur; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 228 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 229 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 230 | } |
| 231 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 232 | page = virt_to_page(bdata->node_bootmem_map); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 233 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 234 | pages = bootmem_bootmap_pages(pages); |
| 235 | count += pages; |
Andrew Morton | 5576646 | 2012-11-16 14:15:06 -0800 | [diff] [blame] | 236 | while (pages--) |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 237 | __free_pages_bootmem(page++, 0); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 238 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 239 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
| 240 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 241 | return count; |
| 242 | } |
| 243 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 244 | static int reset_managed_pages_done __initdata; |
| 245 | |
| 246 | static inline void __init reset_node_managed_pages(pg_data_t *pgdat) |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 247 | { |
| 248 | struct zone *z; |
| 249 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 250 | if (reset_managed_pages_done) |
| 251 | return; |
| 252 | |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 253 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 254 | z->managed_pages = 0; |
| 255 | } |
| 256 | |
| 257 | void __init reset_all_zones_managed_pages(void) |
| 258 | { |
| 259 | struct pglist_data *pgdat; |
| 260 | |
| 261 | for_each_online_pgdat(pgdat) |
| 262 | reset_node_managed_pages(pgdat); |
| 263 | reset_managed_pages_done = 1; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 264 | } |
| 265 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 266 | /** |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 267 | * free_all_bootmem - release free pages to the buddy allocator |
| 268 | * |
| 269 | * Returns the number of pages actually released. |
| 270 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 271 | unsigned long __init free_all_bootmem(void) |
| 272 | { |
Yinghai Lu | aa235fc | 2010-03-31 20:45:27 -0700 | [diff] [blame] | 273 | unsigned long total_pages = 0; |
| 274 | bootmem_data_t *bdata; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 275 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 276 | reset_all_zones_managed_pages(); |
Yinghai Lu | aa235fc | 2010-03-31 20:45:27 -0700 | [diff] [blame] | 277 | |
| 278 | list_for_each_entry(bdata, &bdata_list, list) |
| 279 | total_pages += free_all_bootmem_core(bdata); |
| 280 | |
Jiang Liu | 0c98853 | 2013-07-03 15:03:24 -0700 | [diff] [blame] | 281 | totalram_pages += total_pages; |
| 282 | |
Yinghai Lu | aa235fc | 2010-03-31 20:45:27 -0700 | [diff] [blame] | 283 | return total_pages; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 284 | } |
| 285 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 286 | static void __init __free(bootmem_data_t *bdata, |
| 287 | unsigned long sidx, unsigned long eidx) |
| 288 | { |
| 289 | unsigned long idx; |
| 290 | |
| 291 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 292 | sidx + bdata->node_min_pfn, |
| 293 | eidx + bdata->node_min_pfn); |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 294 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 295 | if (bdata->hint_idx > sidx) |
| 296 | bdata->hint_idx = sidx; |
| 297 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 298 | for (idx = sidx; idx < eidx; idx++) |
| 299 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) |
| 300 | BUG(); |
| 301 | } |
| 302 | |
| 303 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, |
| 304 | unsigned long eidx, int flags) |
| 305 | { |
| 306 | unsigned long idx; |
| 307 | int exclusive = flags & BOOTMEM_EXCLUSIVE; |
| 308 | |
| 309 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", |
| 310 | bdata - bootmem_node_data, |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 311 | sidx + bdata->node_min_pfn, |
| 312 | eidx + bdata->node_min_pfn, |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 313 | flags); |
| 314 | |
| 315 | for (idx = sidx; idx < eidx; idx++) |
| 316 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { |
| 317 | if (exclusive) { |
| 318 | __free(bdata, sidx, idx); |
| 319 | return -EBUSY; |
| 320 | } |
| 321 | bdebug("silent double reserve of PFN %lx\n", |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 322 | idx + bdata->node_min_pfn); |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 323 | } |
| 324 | return 0; |
| 325 | } |
| 326 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 327 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
| 328 | unsigned long start, unsigned long end, |
| 329 | int reserve, int flags) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 330 | { |
| 331 | unsigned long sidx, eidx; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 332 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 333 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
| 334 | bdata - bootmem_node_data, start, end, reserve, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 335 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 336 | BUG_ON(start < bdata->node_min_pfn); |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 337 | BUG_ON(end > bdata->node_low_pfn); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 338 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 339 | sidx = start - bdata->node_min_pfn; |
| 340 | eidx = end - bdata->node_min_pfn; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 341 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 342 | if (reserve) |
| 343 | return __reserve(bdata, sidx, eidx, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 344 | else |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 345 | __free(bdata, sidx, eidx); |
| 346 | return 0; |
| 347 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 348 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 349 | static int __init mark_bootmem(unsigned long start, unsigned long end, |
| 350 | int reserve, int flags) |
| 351 | { |
| 352 | unsigned long pos; |
| 353 | bootmem_data_t *bdata; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 354 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 355 | pos = start; |
| 356 | list_for_each_entry(bdata, &bdata_list, list) { |
| 357 | int err; |
| 358 | unsigned long max; |
| 359 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 360 | if (pos < bdata->node_min_pfn || |
| 361 | pos >= bdata->node_low_pfn) { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 362 | BUG_ON(pos != start); |
| 363 | continue; |
| 364 | } |
| 365 | |
| 366 | max = min(bdata->node_low_pfn, end); |
| 367 | |
| 368 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
| 369 | if (reserve && err) { |
| 370 | mark_bootmem(start, pos, 0, 0); |
| 371 | return err; |
| 372 | } |
| 373 | |
| 374 | if (max == end) |
| 375 | return 0; |
| 376 | pos = bdata->node_low_pfn; |
| 377 | } |
| 378 | BUG(); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 379 | } |
| 380 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 381 | /** |
| 382 | * free_bootmem_node - mark a page range as usable |
| 383 | * @pgdat: node the range resides on |
| 384 | * @physaddr: starting address of the range |
| 385 | * @size: size of the range in bytes |
| 386 | * |
| 387 | * Partial pages will be considered reserved and left as they are. |
| 388 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 389 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 390 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 391 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 392 | unsigned long size) |
| 393 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 394 | unsigned long start, end; |
| 395 | |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 396 | kmemleak_free_part(__va(physaddr), size); |
| 397 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 398 | start = PFN_UP(physaddr); |
| 399 | end = PFN_DOWN(physaddr + size); |
| 400 | |
| 401 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 402 | } |
| 403 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 404 | /** |
| 405 | * free_bootmem - mark a page range as usable |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 406 | * @addr: starting physical address of the range |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 407 | * @size: size of the range in bytes |
| 408 | * |
| 409 | * Partial pages will be considered reserved and left as they are. |
| 410 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 411 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 412 | */ |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 413 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 414 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 415 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 416 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 417 | kmemleak_free_part(__va(physaddr), size); |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 418 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 419 | start = PFN_UP(physaddr); |
| 420 | end = PFN_DOWN(physaddr + size); |
Yinghai Lu | a5645a6 | 2008-03-18 12:49:12 -0700 | [diff] [blame] | 421 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 422 | mark_bootmem(start, end, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | } |
| 424 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 425 | /** |
| 426 | * reserve_bootmem_node - mark a page range as reserved |
| 427 | * @pgdat: node the range resides on |
| 428 | * @physaddr: starting address of the range |
| 429 | * @size: size of the range in bytes |
| 430 | * @flags: reservation flags (see linux/bootmem.h) |
| 431 | * |
| 432 | * Partial pages will be reserved. |
| 433 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 434 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 435 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 436 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 437 | unsigned long size, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 439 | unsigned long start, end; |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 440 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 441 | start = PFN_DOWN(physaddr); |
| 442 | end = PFN_UP(physaddr + size); |
| 443 | |
| 444 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | } |
| 446 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 447 | /** |
Javi Merino | 0d4ba4d | 2012-08-24 18:09:31 +0100 | [diff] [blame] | 448 | * reserve_bootmem - mark a page range as reserved |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 449 | * @addr: starting address of the range |
| 450 | * @size: size of the range in bytes |
| 451 | * @flags: reservation flags (see linux/bootmem.h) |
| 452 | * |
| 453 | * Partial pages will be reserved. |
| 454 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 455 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 456 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 457 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
| 458 | int flags) |
| 459 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 460 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 461 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 462 | start = PFN_DOWN(addr); |
| 463 | end = PFN_UP(addr + size); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 464 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 465 | return mark_bootmem(start, end, 1, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 466 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 467 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 468 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
| 469 | unsigned long idx, unsigned long step) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 470 | { |
| 471 | unsigned long base = bdata->node_min_pfn; |
| 472 | |
| 473 | /* |
| 474 | * Align the index with respect to the node start so that the |
| 475 | * combination of both satisfies the requested alignment. |
| 476 | */ |
| 477 | |
| 478 | return ALIGN(base + idx, step) - base; |
| 479 | } |
| 480 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 481 | static unsigned long __init align_off(struct bootmem_data *bdata, |
| 482 | unsigned long off, unsigned long align) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 483 | { |
| 484 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); |
| 485 | |
| 486 | /* Same as align_idx for byte offsets */ |
| 487 | |
| 488 | return ALIGN(base + off, align) - base; |
| 489 | } |
| 490 | |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 491 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 492 | unsigned long size, unsigned long align, |
| 493 | unsigned long goal, unsigned long limit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 495 | unsigned long fallback = 0; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 496 | unsigned long min, max, start, sidx, midx, step; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | |
Johannes Weiner | 594fe1a | 2009-01-06 14:40:32 -0800 | [diff] [blame] | 498 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
| 499 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, |
| 500 | align, goal, limit); |
| 501 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 502 | BUG_ON(!size); |
| 503 | BUG_ON(align & (align - 1)); |
| 504 | BUG_ON(limit && goal + size > limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | |
Christian Krafft | 7c309a6 | 2006-12-06 20:32:41 -0800 | [diff] [blame] | 506 | if (!bdata->node_bootmem_map) |
| 507 | return NULL; |
| 508 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 509 | min = bdata->node_min_pfn; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 510 | max = bdata->node_low_pfn; |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 511 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 512 | goal >>= PAGE_SHIFT; |
| 513 | limit >>= PAGE_SHIFT; |
| 514 | |
| 515 | if (limit && max > limit) |
| 516 | max = limit; |
| 517 | if (max <= min) |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 518 | return NULL; |
| 519 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 520 | step = max(align >> PAGE_SHIFT, 1UL); |
Yasunori Goto | 281dd25 | 2005-10-19 15:52:18 -0700 | [diff] [blame] | 521 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 522 | if (goal && min < goal && goal < max) |
| 523 | start = ALIGN(goal, step); |
| 524 | else |
| 525 | start = ALIGN(min, step); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 527 | sidx = start - bdata->node_min_pfn; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 528 | midx = max - bdata->node_min_pfn; |
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 (bdata->hint_idx > sidx) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 531 | /* |
| 532 | * Handle the valid case of sidx being zero and still |
| 533 | * catch the fallback below. |
| 534 | */ |
| 535 | fallback = sidx + 1; |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 536 | sidx = align_idx(bdata, bdata->hint_idx, step); |
Yinghai Lu | ad09315 | 2008-03-10 23:23:42 -0700 | [diff] [blame] | 537 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 539 | while (1) { |
| 540 | int merge; |
| 541 | void *region; |
| 542 | unsigned long eidx, i, start_off, end_off; |
| 543 | find_block: |
| 544 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 545 | sidx = align_idx(bdata, sidx, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 546 | eidx = sidx + PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 548 | if (sidx >= midx || eidx > midx) |
Haren Myneni | 66d43e9 | 2005-12-12 00:37:39 -0800 | [diff] [blame] | 549 | break; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 550 | |
| 551 | for (i = sidx; i < eidx; i++) |
| 552 | if (test_bit(i, bdata->node_bootmem_map)) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 553 | sidx = align_idx(bdata, i, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 554 | if (sidx == i) |
| 555 | sidx += step; |
| 556 | goto find_block; |
| 557 | } |
| 558 | |
Mikulas Patocka | 627240a | 2008-08-15 00:40:17 -0700 | [diff] [blame] | 559 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 560 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 561 | start_off = align_off(bdata, bdata->last_end_off, align); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 562 | else |
| 563 | start_off = PFN_PHYS(sidx); |
| 564 | |
| 565 | merge = PFN_DOWN(start_off) < sidx; |
| 566 | end_off = start_off + size; |
| 567 | |
| 568 | bdata->last_end_off = end_off; |
| 569 | bdata->hint_idx = PFN_UP(end_off); |
| 570 | |
| 571 | /* |
| 572 | * Reserve the area now: |
| 573 | */ |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 574 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
| 575 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) |
| 576 | BUG(); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 577 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 578 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
| 579 | start_off); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 580 | memset(region, 0, size); |
Catalin Marinas | 008139d | 2009-08-27 14:29:17 +0100 | [diff] [blame] | 581 | /* |
| 582 | * The min_count is set to 0 so that bootmem allocated blocks |
| 583 | * are never reported as leaks. |
| 584 | */ |
| 585 | kmemleak_alloc(region, size, 0, 0); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 586 | return region; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | } |
| 588 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 589 | if (fallback) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 590 | sidx = align_idx(bdata, fallback - 1, step); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 591 | fallback = 0; |
| 592 | goto find_block; |
| 593 | } |
| 594 | |
| 595 | return NULL; |
| 596 | } |
| 597 | |
Johannes Weiner | c12ab50 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 598 | static void * __init alloc_bootmem_core(unsigned long size, |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 599 | unsigned long align, |
| 600 | unsigned long goal, |
| 601 | unsigned long limit) |
| 602 | { |
| 603 | bootmem_data_t *bdata; |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 604 | void *region; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 605 | |
Joonsoo Kim | 3f7dfe2 | 2012-12-12 13:50:43 -0800 | [diff] [blame] | 606 | if (WARN_ON_ONCE(slab_is_available())) |
| 607 | return kzalloc(size, GFP_NOWAIT); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 608 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 609 | list_for_each_entry(bdata, &bdata_list, list) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 610 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
| 611 | continue; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 612 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 613 | break; |
| 614 | |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 615 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 616 | if (region) |
| 617 | return region; |
| 618 | } |
| 619 | |
Johannes Weiner | c12ab50 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 620 | return NULL; |
| 621 | } |
| 622 | |
| 623 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
| 624 | unsigned long align, |
| 625 | unsigned long goal, |
| 626 | unsigned long limit) |
| 627 | { |
| 628 | void *ptr; |
| 629 | |
| 630 | restart: |
| 631 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 632 | if (ptr) |
| 633 | return ptr; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 634 | if (goal) { |
| 635 | goal = 0; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 636 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | } |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 638 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | } |
| 641 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 642 | /** |
| 643 | * __alloc_bootmem_nopanic - allocate boot memory without panicking |
| 644 | * @size: size of the request in bytes |
| 645 | * @align: alignment of the region |
| 646 | * @goal: preferred starting address of the region |
| 647 | * |
| 648 | * The goal is dropped if it can not be satisfied and the allocation will |
| 649 | * fall back to memory below @goal. |
| 650 | * |
| 651 | * Allocation may happen on any node in the system. |
| 652 | * |
| 653 | * Returns NULL on failure. |
| 654 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 655 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 656 | unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 658 | unsigned long limit = 0; |
| 659 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 660 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 661 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 663 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
| 664 | unsigned long goal, unsigned long limit) |
| 665 | { |
| 666 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 667 | |
| 668 | if (mem) |
| 669 | return mem; |
| 670 | /* |
| 671 | * Whoops, we cannot satisfy the allocation request. |
| 672 | */ |
| 673 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
| 674 | panic("Out of memory"); |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 675 | return NULL; |
| 676 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 678 | /** |
| 679 | * __alloc_bootmem - allocate boot memory |
| 680 | * @size: size of the request in bytes |
| 681 | * @align: alignment of the region |
| 682 | * @goal: preferred starting address of the region |
| 683 | * |
| 684 | * The goal is dropped if it can not be satisfied and the allocation will |
| 685 | * fall back to memory below @goal. |
| 686 | * |
| 687 | * Allocation may happen on any node in the system. |
| 688 | * |
| 689 | * The function panics if the request can not be satisfied. |
| 690 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 691 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
| 692 | unsigned long goal) |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 693 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 694 | unsigned long limit = 0; |
| 695 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 696 | return ___alloc_bootmem(size, align, goal, limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | } |
| 698 | |
Yinghai Lu | 99ab7b1 | 2012-07-11 14:02:53 -0700 | [diff] [blame] | 699 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 700 | unsigned long size, unsigned long align, |
| 701 | unsigned long goal, unsigned long limit) |
| 702 | { |
| 703 | void *ptr; |
| 704 | |
Joonsoo Kim | 3f7dfe2 | 2012-12-12 13:50:43 -0800 | [diff] [blame] | 705 | if (WARN_ON_ONCE(slab_is_available())) |
| 706 | return kzalloc(size, GFP_NOWAIT); |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 707 | again: |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 708 | |
Yinghai Lu | c8f4a2d | 2012-07-17 15:47:51 -0700 | [diff] [blame] | 709 | /* do not panic in alloc_bootmem_bdata() */ |
| 710 | if (limit && goal + size > limit) |
| 711 | limit = 0; |
| 712 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 713 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 714 | if (ptr) |
| 715 | return ptr; |
| 716 | |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 717 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 718 | if (ptr) |
| 719 | return ptr; |
| 720 | |
| 721 | if (goal) { |
| 722 | goal = 0; |
| 723 | goto again; |
| 724 | } |
| 725 | |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 726 | return NULL; |
| 727 | } |
| 728 | |
| 729 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
| 730 | unsigned long align, unsigned long goal) |
| 731 | { |
| 732 | if (WARN_ON_ONCE(slab_is_available())) |
| 733 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 734 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 735 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 736 | } |
| 737 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 738 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 739 | unsigned long align, unsigned long goal, |
| 740 | unsigned long limit) |
| 741 | { |
| 742 | void *ptr; |
| 743 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 744 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 745 | if (ptr) |
| 746 | return ptr; |
| 747 | |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 748 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
| 749 | panic("Out of memory"); |
| 750 | return NULL; |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 751 | } |
| 752 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 753 | /** |
| 754 | * __alloc_bootmem_node - allocate boot memory from a specific node |
| 755 | * @pgdat: node to allocate from |
| 756 | * @size: size of the request in bytes |
| 757 | * @align: alignment of the region |
| 758 | * @goal: preferred starting address of the region |
| 759 | * |
| 760 | * The goal is dropped if it can not be satisfied and the allocation will |
| 761 | * fall back to memory below @goal. |
| 762 | * |
| 763 | * Allocation may fall back to any node in the system if the specified node |
| 764 | * can not hold the requested memory. |
| 765 | * |
| 766 | * The function panics if the request can not be satisfied. |
| 767 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 768 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 769 | unsigned long align, unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 771 | if (WARN_ON_ONCE(slab_is_available())) |
| 772 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 773 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 774 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 775 | } |
| 776 | |
| 777 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, |
| 778 | unsigned long align, unsigned long goal) |
| 779 | { |
| 780 | #ifdef MAX_DMA32_PFN |
| 781 | unsigned long end_pfn; |
| 782 | |
| 783 | if (WARN_ON_ONCE(slab_is_available())) |
| 784 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 785 | |
| 786 | /* update goal according ...MAX_DMA32_PFN */ |
| 787 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 788 | |
| 789 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && |
| 790 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { |
| 791 | void *ptr; |
| 792 | unsigned long new_goal; |
| 793 | |
| 794 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 795 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 796 | new_goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 797 | if (ptr) |
| 798 | return ptr; |
| 799 | } |
| 800 | #endif |
| 801 | |
| 802 | return __alloc_bootmem_node(pgdat, size, align, goal); |
| 803 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | } |
| 805 | |
Heiko Carstens | dfd54cb | 2006-09-25 23:31:33 -0700 | [diff] [blame] | 806 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
| 807 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL |
| 808 | #endif |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 809 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 810 | /** |
| 811 | * __alloc_bootmem_low - allocate low boot memory |
| 812 | * @size: size of the request in bytes |
| 813 | * @align: alignment of the region |
| 814 | * @goal: preferred starting address of the region |
| 815 | * |
| 816 | * The goal is dropped if it can not be satisfied and the allocation will |
| 817 | * fall back to memory below @goal. |
| 818 | * |
| 819 | * Allocation may happen on any node in the system. |
| 820 | * |
| 821 | * The function panics if the request can not be satisfied. |
| 822 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 823 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
| 824 | unsigned long goal) |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 825 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 826 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 827 | } |
| 828 | |
Yinghai Lu | 38fa417 | 2013-01-24 12:20:15 -0800 | [diff] [blame] | 829 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
| 830 | unsigned long align, |
| 831 | unsigned long goal) |
| 832 | { |
| 833 | return ___alloc_bootmem_nopanic(size, align, goal, |
| 834 | ARCH_LOW_ADDRESS_LIMIT); |
| 835 | } |
| 836 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 837 | /** |
| 838 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node |
| 839 | * @pgdat: node to allocate from |
| 840 | * @size: size of the request in bytes |
| 841 | * @align: alignment of the region |
| 842 | * @goal: preferred starting address of the region |
| 843 | * |
| 844 | * The goal is dropped if it can not be satisfied and the allocation will |
| 845 | * fall back to memory below @goal. |
| 846 | * |
| 847 | * Allocation may fall back to any node in the system if the specified node |
| 848 | * can not hold the requested memory. |
| 849 | * |
| 850 | * The function panics if the request can not be satisfied. |
| 851 | */ |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 852 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
| 853 | unsigned long align, unsigned long goal) |
| 854 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 855 | if (WARN_ON_ONCE(slab_is_available())) |
| 856 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 857 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 858 | return ___alloc_bootmem_node(pgdat, size, align, |
| 859 | goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 860 | } |