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> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 14 | #include <linux/export.h> |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 15 | #include <linux/kmemleak.h> |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 16 | #include <linux/range.h> |
Paul McQuade | d85fbee | 2014-10-09 15:29:05 -0700 | [diff] [blame] | 17 | #include <linux/bug.h> |
| 18 | #include <linux/io.h> |
zijun_hu | 1d8bf92 | 2016-10-07 16:59:27 -0700 | [diff] [blame] | 19 | #include <linux/bootmem.h> |
Franck Bui-Huu | e786e86 | 2006-09-25 23:31:06 -0700 | [diff] [blame] | 20 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include "internal.h" |
| 22 | |
Yinghai Lu | e782ab4 | 2011-02-24 14:43:06 +0100 | [diff] [blame] | 23 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
| 24 | struct pglist_data __refdata contig_page_data = { |
| 25 | .bdata = &bootmem_node_data[0] |
| 26 | }; |
| 27 | EXPORT_SYMBOL(contig_page_data); |
| 28 | #endif |
| 29 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | unsigned long max_low_pfn; |
| 31 | unsigned long min_low_pfn; |
| 32 | unsigned long max_pfn; |
Igor Mammedov | 8dd3303 | 2015-12-04 14:07:05 +0100 | [diff] [blame] | 33 | unsigned long long max_possible_pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | |
Johannes Weiner | b61bfa3 | 2008-07-23 21:26:55 -0700 | [diff] [blame] | 35 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
| 36 | |
Johannes Weiner | 636cc40 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 37 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
| 38 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 39 | static int bootmem_debug; |
| 40 | |
| 41 | static int __init bootmem_debug_setup(char *buf) |
| 42 | { |
| 43 | bootmem_debug = 1; |
| 44 | return 0; |
| 45 | } |
| 46 | early_param("bootmem_debug", bootmem_debug_setup); |
| 47 | |
| 48 | #define bdebug(fmt, args...) ({ \ |
| 49 | if (unlikely(bootmem_debug)) \ |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 50 | pr_info("bootmem::%s " fmt, \ |
Harvey Harrison | 80a914d | 2008-10-15 22:01:25 -0700 | [diff] [blame] | 51 | __func__, ## args); \ |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 52 | }) |
| 53 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 54 | static unsigned long __init bootmap_bytes(unsigned long pages) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 55 | { |
Adygzhy Ondar | d3a9d7a3 | 2017-02-22 15:44:58 -0800 | [diff] [blame^] | 56 | unsigned long bytes = DIV_ROUND_UP(pages, BITS_PER_BYTE); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 57 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 58 | return ALIGN(bytes, sizeof(long)); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 59 | } |
| 60 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 61 | /** |
| 62 | * bootmem_bootmap_pages - calculate bitmap size in pages |
| 63 | * @pages: number of pages the bitmap has to represent |
| 64 | */ |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 65 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | { |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 67 | unsigned long bytes = bootmap_bytes(pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 69 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | } |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 71 | |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 72 | /* |
| 73 | * link bdata in order |
| 74 | */ |
Franck Bui-Huu | 69d49e6 | 2006-09-25 23:31:04 -0700 | [diff] [blame] | 75 | static void __init link_bootmem(bootmem_data_t *bdata) |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 76 | { |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 77 | bootmem_data_t *ent; |
Franck Bui-Huu | f71bf0c | 2006-09-25 23:31:08 -0700 | [diff] [blame] | 78 | |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 79 | list_for_each_entry(ent, &bdata_list, list) { |
| 80 | if (bdata->node_min_pfn < ent->node_min_pfn) { |
| 81 | list_add_tail(&bdata->list, &ent->list); |
| 82 | return; |
| 83 | } |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 84 | } |
Gavin Shan | 5c2b8a1 | 2012-05-29 15:06:46 -0700 | [diff] [blame] | 85 | |
| 86 | list_add_tail(&bdata->list, &bdata_list); |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 87 | } |
| 88 | |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 89 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | * Called once to set up the allocator itself. |
| 91 | */ |
Johannes Weiner | 8ae0446 | 2008-07-23 21:26:59 -0700 | [diff] [blame] | 92 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | unsigned long mapstart, unsigned long start, unsigned long end) |
| 94 | { |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 95 | unsigned long mapsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
Mel Gorman | 2dbb51c | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 97 | mminit_validate_memmodel_limits(&start, &end); |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 98 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 99 | bdata->node_min_pfn = start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | bdata->node_low_pfn = end; |
KAMEZAWA Hiroyuki | 679bc9f | 2006-03-27 01:15:58 -0800 | [diff] [blame] | 101 | link_bootmem(bdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | |
| 103 | /* |
| 104 | * Initially all pages are reserved - setup_arch() has to |
| 105 | * register free RAM areas explicitly. |
| 106 | */ |
Johannes Weiner | df049a5 | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 107 | mapsize = bootmap_bytes(end - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
| 109 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 110 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
| 111 | bdata - bootmem_node_data, start, mapstart, end, mapsize); |
| 112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | return mapsize; |
| 114 | } |
| 115 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 116 | /** |
| 117 | * init_bootmem_node - register a node as boot memory |
| 118 | * @pgdat: node to register |
| 119 | * @freepfn: pfn where the bitmap for this node is to be placed |
| 120 | * @startpfn: first pfn on the node |
| 121 | * @endpfn: first pfn after the node |
| 122 | * |
| 123 | * Returns the number of bytes needed to hold the bitmap for this node. |
| 124 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 125 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
| 126 | unsigned long startpfn, unsigned long endpfn) |
| 127 | { |
| 128 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); |
| 129 | } |
| 130 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 131 | /** |
| 132 | * init_bootmem - register boot memory |
| 133 | * @start: pfn where the bitmap is to be placed |
| 134 | * @pages: number of available physical pages |
| 135 | * |
| 136 | * Returns the number of bytes needed to hold the bitmap. |
| 137 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 138 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
| 139 | { |
| 140 | max_low_pfn = pages; |
| 141 | min_low_pfn = start; |
| 142 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); |
| 143 | } |
Yinghai Lu | 0932587 | 2011-02-24 14:43:05 +0100 | [diff] [blame] | 144 | |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 145 | /* |
| 146 | * free_bootmem_late - free bootmem pages directly to page allocator |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 147 | * @addr: starting physical address of the range |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 148 | * @size: size of the range in bytes |
| 149 | * |
| 150 | * This is only useful when the bootmem allocator has already been torn |
| 151 | * down, but we are still initializing the system. Pages are given directly |
| 152 | * to the page allocator, no bootmem metadata is updated because it is gone. |
| 153 | */ |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 154 | void __init free_bootmem_late(unsigned long physaddr, unsigned long size) |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 155 | { |
| 156 | unsigned long cursor, end; |
| 157 | |
Catalin Marinas | 9099dae | 2016-10-11 13:55:11 -0700 | [diff] [blame] | 158 | kmemleak_free_part_phys(physaddr, size); |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 159 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 160 | cursor = PFN_UP(physaddr); |
| 161 | end = PFN_DOWN(physaddr + size); |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 162 | |
| 163 | for (; cursor < end; cursor++) { |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 164 | __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); |
FUJITA Tomonori | 9f993ac | 2009-11-10 19:46:17 +0900 | [diff] [blame] | 165 | totalram_pages++; |
| 166 | } |
| 167 | } |
| 168 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 169 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
| 170 | { |
| 171 | struct page *page; |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 172 | unsigned long *map, start, end, pages, cur, count = 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 173 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 174 | if (!bdata->node_bootmem_map) |
| 175 | return 0; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 176 | |
Daeseok Youn | 4a099fb | 2013-11-12 15:08:14 -0800 | [diff] [blame] | 177 | map = bdata->node_bootmem_map; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 178 | start = bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 179 | end = bdata->node_low_pfn; |
| 180 | |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 181 | bdebug("nid=%td start=%lx end=%lx\n", |
| 182 | bdata - bootmem_node_data, start, end); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 183 | |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 184 | while (start < end) { |
Daeseok Youn | 4a099fb | 2013-11-12 15:08:14 -0800 | [diff] [blame] | 185 | unsigned long idx, vec; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 186 | unsigned shift; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 187 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 188 | idx = start - bdata->node_min_pfn; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 189 | shift = idx & (BITS_PER_LONG - 1); |
| 190 | /* |
| 191 | * vec holds at most BITS_PER_LONG map bits, |
| 192 | * bit 0 corresponds to start. |
| 193 | */ |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 194 | vec = ~map[idx / BITS_PER_LONG]; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 195 | |
| 196 | if (shift) { |
| 197 | vec >>= shift; |
| 198 | if (end - start >= BITS_PER_LONG) |
| 199 | vec |= ~map[idx / BITS_PER_LONG + 1] << |
| 200 | (BITS_PER_LONG - shift); |
| 201 | } |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 202 | /* |
| 203 | * If we have a properly aligned and fully unreserved |
| 204 | * BITS_PER_LONG block of pages in front of us, free |
| 205 | * it in one go. |
| 206 | */ |
| 207 | if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 208 | int order = ilog2(BITS_PER_LONG); |
| 209 | |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 210 | __free_pages_bootmem(pfn_to_page(start), start, order); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 211 | count += BITS_PER_LONG; |
Johannes Weiner | 799f933 | 2012-01-10 15:08:15 -0800 | [diff] [blame] | 212 | start += BITS_PER_LONG; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 213 | } else { |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 214 | cur = start; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 215 | |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 216 | start = ALIGN(start + 1, BITS_PER_LONG); |
| 217 | while (vec && cur != start) { |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 218 | if (vec & 1) { |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 219 | page = pfn_to_page(cur); |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 220 | __free_pages_bootmem(page, cur, 0); |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 221 | count++; |
| 222 | } |
| 223 | vec >>= 1; |
Max Filippov | 10d73e6 | 2013-01-11 14:31:52 -0800 | [diff] [blame] | 224 | ++cur; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 225 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 226 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 227 | } |
| 228 | |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 229 | cur = bdata->node_min_pfn; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 230 | page = virt_to_page(bdata->node_bootmem_map); |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 231 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
Johannes Weiner | 41546c1 | 2008-07-23 21:28:03 -0700 | [diff] [blame] | 232 | pages = bootmem_bootmap_pages(pages); |
| 233 | count += pages; |
Andrew Morton | 5576646 | 2012-11-16 14:15:06 -0800 | [diff] [blame] | 234 | while (pages--) |
Mel Gorman | d70ddd7 | 2015-06-30 14:56:52 -0700 | [diff] [blame] | 235 | __free_pages_bootmem(page++, cur++, 0); |
Chris Metcalf | 1b4ace4 | 2015-09-08 15:02:12 -0700 | [diff] [blame] | 236 | bdata->node_bootmem_map = NULL; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 237 | |
Johannes Weiner | 2e5237d | 2008-07-23 21:28:02 -0700 | [diff] [blame] | 238 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
| 239 | |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 240 | return count; |
| 241 | } |
| 242 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 243 | static int reset_managed_pages_done __initdata; |
| 244 | |
Tang Chen | f784a3f | 2014-11-13 15:19:39 -0800 | [diff] [blame] | 245 | void reset_node_managed_pages(pg_data_t *pgdat) |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 246 | { |
| 247 | struct zone *z; |
| 248 | |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 249 | 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] | 250 | z->managed_pages = 0; |
| 251 | } |
| 252 | |
| 253 | void __init reset_all_zones_managed_pages(void) |
| 254 | { |
| 255 | struct pglist_data *pgdat; |
| 256 | |
Tang Chen | f784a3f | 2014-11-13 15:19:39 -0800 | [diff] [blame] | 257 | if (reset_managed_pages_done) |
| 258 | return; |
| 259 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 260 | for_each_online_pgdat(pgdat) |
| 261 | reset_node_managed_pages(pgdat); |
Tang Chen | f784a3f | 2014-11-13 15:19:39 -0800 | [diff] [blame] | 262 | |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 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 | |
Chris Metcalf | 1b4ace4 | 2015-09-08 15:02:12 -0700 | [diff] [blame] | 295 | if (WARN_ON(bdata->node_bootmem_map == NULL)) |
| 296 | return; |
| 297 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 298 | if (bdata->hint_idx > sidx) |
| 299 | bdata->hint_idx = sidx; |
| 300 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 301 | for (idx = sidx; idx < eidx; idx++) |
| 302 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) |
| 303 | BUG(); |
| 304 | } |
| 305 | |
| 306 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, |
| 307 | unsigned long eidx, int flags) |
| 308 | { |
| 309 | unsigned long idx; |
| 310 | int exclusive = flags & BOOTMEM_EXCLUSIVE; |
| 311 | |
| 312 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", |
| 313 | bdata - bootmem_node_data, |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 314 | sidx + bdata->node_min_pfn, |
| 315 | eidx + bdata->node_min_pfn, |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 316 | flags); |
| 317 | |
Chris Metcalf | 1b4ace4 | 2015-09-08 15:02:12 -0700 | [diff] [blame] | 318 | if (WARN_ON(bdata->node_bootmem_map == NULL)) |
| 319 | return 0; |
| 320 | |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 321 | for (idx = sidx; idx < eidx; idx++) |
| 322 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { |
| 323 | if (exclusive) { |
| 324 | __free(bdata, sidx, idx); |
| 325 | return -EBUSY; |
| 326 | } |
| 327 | bdebug("silent double reserve of PFN %lx\n", |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 328 | idx + bdata->node_min_pfn); |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 329 | } |
| 330 | return 0; |
| 331 | } |
| 332 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 333 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
| 334 | unsigned long start, unsigned long end, |
| 335 | int reserve, int flags) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 336 | { |
| 337 | unsigned long sidx, eidx; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 338 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 339 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
| 340 | bdata - bootmem_node_data, start, end, reserve, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 341 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 342 | BUG_ON(start < bdata->node_min_pfn); |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 343 | BUG_ON(end > bdata->node_low_pfn); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 344 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 345 | sidx = start - bdata->node_min_pfn; |
| 346 | eidx = end - bdata->node_min_pfn; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 347 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 348 | if (reserve) |
| 349 | return __reserve(bdata, sidx, eidx, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 350 | else |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 351 | __free(bdata, sidx, eidx); |
| 352 | return 0; |
| 353 | } |
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 | static int __init mark_bootmem(unsigned long start, unsigned long end, |
| 356 | int reserve, int flags) |
| 357 | { |
| 358 | unsigned long pos; |
| 359 | bootmem_data_t *bdata; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 360 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 361 | pos = start; |
| 362 | list_for_each_entry(bdata, &bdata_list, list) { |
| 363 | int err; |
| 364 | unsigned long max; |
| 365 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 366 | if (pos < bdata->node_min_pfn || |
| 367 | pos >= bdata->node_low_pfn) { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 368 | BUG_ON(pos != start); |
| 369 | continue; |
| 370 | } |
| 371 | |
| 372 | max = min(bdata->node_low_pfn, end); |
| 373 | |
| 374 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
| 375 | if (reserve && err) { |
| 376 | mark_bootmem(start, pos, 0, 0); |
| 377 | return err; |
| 378 | } |
| 379 | |
| 380 | if (max == end) |
| 381 | return 0; |
| 382 | pos = bdata->node_low_pfn; |
| 383 | } |
| 384 | BUG(); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 385 | } |
| 386 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 387 | /** |
| 388 | * free_bootmem_node - mark a page range as usable |
| 389 | * @pgdat: node the range resides on |
| 390 | * @physaddr: starting address of the range |
| 391 | * @size: size of the range in bytes |
| 392 | * |
| 393 | * Partial pages will be considered reserved and left as they are. |
| 394 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 395 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 396 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 397 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 398 | unsigned long size) |
| 399 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 400 | unsigned long start, end; |
| 401 | |
Catalin Marinas | 9099dae | 2016-10-11 13:55:11 -0700 | [diff] [blame] | 402 | kmemleak_free_part_phys(physaddr, size); |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 403 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 404 | start = PFN_UP(physaddr); |
| 405 | end = PFN_DOWN(physaddr + size); |
| 406 | |
| 407 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 408 | } |
| 409 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 410 | /** |
| 411 | * free_bootmem - mark a page range as usable |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 412 | * @addr: starting physical address of the range |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 413 | * @size: size of the range in bytes |
| 414 | * |
| 415 | * Partial pages will be considered reserved and left as they are. |
| 416 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 417 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 418 | */ |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 419 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 420 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 421 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 422 | |
Catalin Marinas | 9099dae | 2016-10-11 13:55:11 -0700 | [diff] [blame] | 423 | kmemleak_free_part_phys(physaddr, size); |
Catalin Marinas | ec3a354 | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 424 | |
Joonsoo Kim | 81df9bf | 2012-12-11 16:03:10 -0800 | [diff] [blame] | 425 | start = PFN_UP(physaddr); |
| 426 | end = PFN_DOWN(physaddr + size); |
Yinghai Lu | a5645a6 | 2008-03-18 12:49:12 -0700 | [diff] [blame] | 427 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 428 | mark_bootmem(start, end, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | } |
| 430 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 431 | /** |
| 432 | * reserve_bootmem_node - mark a page range as reserved |
| 433 | * @pgdat: node the range resides on |
| 434 | * @physaddr: starting address of the range |
| 435 | * @size: size of the range in bytes |
| 436 | * @flags: reservation flags (see linux/bootmem.h) |
| 437 | * |
| 438 | * Partial pages will be reserved. |
| 439 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 440 | * The range must reside completely on the specified node. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 441 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 442 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 443 | unsigned long size, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 445 | unsigned long start, end; |
Franck Bui-Huu | bbc7b92 | 2006-09-25 23:31:07 -0700 | [diff] [blame] | 446 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 447 | start = PFN_DOWN(physaddr); |
| 448 | end = PFN_UP(physaddr + size); |
| 449 | |
| 450 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | } |
| 452 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 453 | /** |
Javi Merino | 0d4ba4d | 2012-08-24 18:09:31 +0100 | [diff] [blame] | 454 | * reserve_bootmem - mark a page range as reserved |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 455 | * @addr: starting address of the range |
| 456 | * @size: size of the range in bytes |
| 457 | * @flags: reservation flags (see linux/bootmem.h) |
| 458 | * |
| 459 | * Partial pages will be reserved. |
| 460 | * |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 461 | * The range must be contiguous but may span node boundaries. |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 462 | */ |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 463 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
| 464 | int flags) |
| 465 | { |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 466 | unsigned long start, end; |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 467 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 468 | start = PFN_DOWN(addr); |
| 469 | end = PFN_UP(addr + size); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 470 | |
Johannes Weiner | e2bf3ca | 2008-07-23 21:28:06 -0700 | [diff] [blame] | 471 | return mark_bootmem(start, end, 1, flags); |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 472 | } |
Johannes Weiner | 223e8dc | 2008-07-23 21:28:00 -0700 | [diff] [blame] | 473 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 474 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
| 475 | unsigned long idx, unsigned long step) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 476 | { |
| 477 | unsigned long base = bdata->node_min_pfn; |
| 478 | |
| 479 | /* |
| 480 | * Align the index with respect to the node start so that the |
| 481 | * combination of both satisfies the requested alignment. |
| 482 | */ |
| 483 | |
| 484 | return ALIGN(base + idx, step) - base; |
| 485 | } |
| 486 | |
Jan Beulich | 8aa043d | 2009-12-14 17:59:46 -0800 | [diff] [blame] | 487 | static unsigned long __init align_off(struct bootmem_data *bdata, |
| 488 | unsigned long off, unsigned long align) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 489 | { |
| 490 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); |
| 491 | |
| 492 | /* Same as align_idx for byte offsets */ |
| 493 | |
| 494 | return ALIGN(base + off, align) - base; |
| 495 | } |
| 496 | |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 497 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 498 | unsigned long size, unsigned long align, |
| 499 | unsigned long goal, unsigned long limit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 501 | unsigned long fallback = 0; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 502 | unsigned long min, max, start, sidx, midx, step; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | |
Johannes Weiner | 594fe1a | 2009-01-06 14:40:32 -0800 | [diff] [blame] | 504 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
| 505 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, |
| 506 | align, goal, limit); |
| 507 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 508 | BUG_ON(!size); |
| 509 | BUG_ON(align & (align - 1)); |
| 510 | BUG_ON(limit && goal + size > limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | |
Christian Krafft | 7c309a6 | 2006-12-06 20:32:41 -0800 | [diff] [blame] | 512 | if (!bdata->node_bootmem_map) |
| 513 | return NULL; |
| 514 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 515 | min = bdata->node_min_pfn; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 516 | max = bdata->node_low_pfn; |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 517 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 518 | goal >>= PAGE_SHIFT; |
| 519 | limit >>= PAGE_SHIFT; |
| 520 | |
| 521 | if (limit && max > limit) |
| 522 | max = limit; |
| 523 | if (max <= min) |
Yinghai Lu | 9a2dc04 | 2008-03-18 12:44:48 -0700 | [diff] [blame] | 524 | return NULL; |
| 525 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 526 | step = max(align >> PAGE_SHIFT, 1UL); |
Yasunori Goto | 281dd25 | 2005-10-19 15:52:18 -0700 | [diff] [blame] | 527 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 528 | if (goal && min < goal && goal < max) |
| 529 | start = ALIGN(goal, step); |
| 530 | else |
| 531 | start = ALIGN(min, step); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 533 | sidx = start - bdata->node_min_pfn; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 534 | midx = max - bdata->node_min_pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 536 | if (bdata->hint_idx > sidx) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 537 | /* |
| 538 | * Handle the valid case of sidx being zero and still |
| 539 | * catch the fallback below. |
| 540 | */ |
| 541 | fallback = sidx + 1; |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 542 | sidx = align_idx(bdata, bdata->hint_idx, step); |
Yinghai Lu | ad09315 | 2008-03-10 23:23:42 -0700 | [diff] [blame] | 543 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 545 | while (1) { |
| 546 | int merge; |
| 547 | void *region; |
| 548 | unsigned long eidx, i, start_off, end_off; |
| 549 | find_block: |
| 550 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 551 | sidx = align_idx(bdata, sidx, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 552 | eidx = sidx + PFN_UP(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 554 | if (sidx >= midx || eidx > midx) |
Haren Myneni | 66d43e9 | 2005-12-12 00:37:39 -0800 | [diff] [blame] | 555 | break; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 556 | |
| 557 | for (i = sidx; i < eidx; i++) |
| 558 | if (test_bit(i, bdata->node_bootmem_map)) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 559 | sidx = align_idx(bdata, i, step); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 560 | if (sidx == i) |
| 561 | sidx += step; |
| 562 | goto find_block; |
| 563 | } |
| 564 | |
Mikulas Patocka | 627240a | 2008-08-15 00:40:17 -0700 | [diff] [blame] | 565 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 566 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 567 | start_off = align_off(bdata, bdata->last_end_off, align); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 568 | else |
| 569 | start_off = PFN_PHYS(sidx); |
| 570 | |
| 571 | merge = PFN_DOWN(start_off) < sidx; |
| 572 | end_off = start_off + size; |
| 573 | |
| 574 | bdata->last_end_off = end_off; |
| 575 | bdata->hint_idx = PFN_UP(end_off); |
| 576 | |
| 577 | /* |
| 578 | * Reserve the area now: |
| 579 | */ |
Johannes Weiner | d747fa4 | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 580 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
| 581 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) |
| 582 | BUG(); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 583 | |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 584 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
| 585 | start_off); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 586 | memset(region, 0, size); |
Catalin Marinas | 008139d | 2009-08-27 14:29:17 +0100 | [diff] [blame] | 587 | /* |
| 588 | * The min_count is set to 0 so that bootmem allocated blocks |
| 589 | * are never reported as leaks. |
| 590 | */ |
| 591 | kmemleak_alloc(region, size, 0, 0); |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 592 | return region; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | } |
| 594 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 595 | if (fallback) { |
Johannes Weiner | 481ebd0 | 2008-08-20 14:09:15 -0700 | [diff] [blame] | 596 | sidx = align_idx(bdata, fallback - 1, step); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 597 | fallback = 0; |
| 598 | goto find_block; |
| 599 | } |
| 600 | |
| 601 | return NULL; |
| 602 | } |
| 603 | |
Johannes Weiner | c12ab50 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 604 | static void * __init alloc_bootmem_core(unsigned long size, |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 605 | unsigned long align, |
| 606 | unsigned long goal, |
| 607 | unsigned long limit) |
| 608 | { |
| 609 | bootmem_data_t *bdata; |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 610 | void *region; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 611 | |
Joonsoo Kim | 3f7dfe2 | 2012-12-12 13:50:43 -0800 | [diff] [blame] | 612 | if (WARN_ON_ONCE(slab_is_available())) |
| 613 | return kzalloc(size, GFP_NOWAIT); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 614 | |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 615 | list_for_each_entry(bdata, &bdata_list, list) { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 616 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
| 617 | continue; |
Johannes Weiner | 3560e24 | 2008-07-23 21:28:09 -0700 | [diff] [blame] | 618 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 619 | break; |
| 620 | |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 621 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 622 | if (region) |
| 623 | return region; |
| 624 | } |
| 625 | |
Johannes Weiner | c12ab50 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 626 | return NULL; |
| 627 | } |
| 628 | |
| 629 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
| 630 | unsigned long align, |
| 631 | unsigned long goal, |
| 632 | unsigned long limit) |
| 633 | { |
| 634 | void *ptr; |
| 635 | |
| 636 | restart: |
| 637 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 638 | if (ptr) |
| 639 | return ptr; |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 640 | if (goal) { |
| 641 | goal = 0; |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 642 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | } |
Johannes Weiner | 5f2809e | 2008-07-23 21:28:05 -0700 | [diff] [blame] | 644 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | } |
| 647 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 648 | /** |
| 649 | * __alloc_bootmem_nopanic - allocate boot memory without panicking |
| 650 | * @size: size of the request in bytes |
| 651 | * @align: alignment of the region |
| 652 | * @goal: preferred starting address of the region |
| 653 | * |
| 654 | * The goal is dropped if it can not be satisfied and the allocation will |
| 655 | * fall back to memory below @goal. |
| 656 | * |
| 657 | * Allocation may happen on any node in the system. |
| 658 | * |
| 659 | * Returns NULL on failure. |
| 660 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 661 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 662 | unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 664 | unsigned long limit = 0; |
| 665 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 666 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 669 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
| 670 | unsigned long goal, unsigned long limit) |
| 671 | { |
| 672 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 673 | |
| 674 | if (mem) |
| 675 | return mem; |
| 676 | /* |
| 677 | * Whoops, we cannot satisfy the allocation request. |
| 678 | */ |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 679 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 680 | panic("Out of memory"); |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 681 | return NULL; |
| 682 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 684 | /** |
| 685 | * __alloc_bootmem - allocate boot memory |
| 686 | * @size: size of the request in bytes |
| 687 | * @align: alignment of the region |
| 688 | * @goal: preferred starting address of the region |
| 689 | * |
| 690 | * The goal is dropped if it can not be satisfied and the allocation will |
| 691 | * fall back to memory below @goal. |
| 692 | * |
| 693 | * Allocation may happen on any node in the system. |
| 694 | * |
| 695 | * The function panics if the request can not be satisfied. |
| 696 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 697 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
| 698 | unsigned long goal) |
Andi Kleen | a806223 | 2006-04-07 19:49:21 +0200 | [diff] [blame] | 699 | { |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 700 | unsigned long limit = 0; |
| 701 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 702 | return ___alloc_bootmem(size, align, goal, limit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | } |
| 704 | |
Yinghai Lu | 99ab7b1 | 2012-07-11 14:02:53 -0700 | [diff] [blame] | 705 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 706 | unsigned long size, unsigned long align, |
| 707 | unsigned long goal, unsigned long limit) |
| 708 | { |
| 709 | void *ptr; |
| 710 | |
Joonsoo Kim | 3f7dfe2 | 2012-12-12 13:50:43 -0800 | [diff] [blame] | 711 | if (WARN_ON_ONCE(slab_is_available())) |
zijun_hu | 1d8bf92 | 2016-10-07 16:59:27 -0700 | [diff] [blame] | 712 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 713 | again: |
Tejun Heo | d0c4f57 | 2009-03-01 16:06:56 +0900 | [diff] [blame] | 714 | |
Yinghai Lu | c8f4a2d | 2012-07-17 15:47:51 -0700 | [diff] [blame] | 715 | /* do not panic in alloc_bootmem_bdata() */ |
| 716 | if (limit && goal + size > limit) |
| 717 | limit = 0; |
| 718 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 719 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 720 | if (ptr) |
| 721 | return ptr; |
| 722 | |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 723 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 724 | if (ptr) |
| 725 | return ptr; |
| 726 | |
| 727 | if (goal) { |
| 728 | goal = 0; |
| 729 | goto again; |
| 730 | } |
| 731 | |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 732 | return NULL; |
| 733 | } |
| 734 | |
| 735 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
| 736 | unsigned long align, unsigned long goal) |
| 737 | { |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 738 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 739 | } |
| 740 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 741 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 742 | unsigned long align, unsigned long goal, |
| 743 | unsigned long limit) |
| 744 | { |
| 745 | void *ptr; |
| 746 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 747 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
Johannes Weiner | 421456e | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 748 | if (ptr) |
| 749 | return ptr; |
| 750 | |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 751 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
Johannes Weiner | ab38184 | 2012-05-29 15:06:34 -0700 | [diff] [blame] | 752 | panic("Out of memory"); |
| 753 | return NULL; |
Johannes Weiner | 4cc278b | 2008-07-23 21:28:08 -0700 | [diff] [blame] | 754 | } |
| 755 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 756 | /** |
| 757 | * __alloc_bootmem_node - allocate boot memory from a specific node |
| 758 | * @pgdat: node to allocate from |
| 759 | * @size: size of the request in bytes |
| 760 | * @align: alignment of the region |
| 761 | * @goal: preferred starting address of the region |
| 762 | * |
| 763 | * The goal is dropped if it can not be satisfied and the allocation will |
| 764 | * fall back to memory below @goal. |
| 765 | * |
| 766 | * Allocation may fall back to any node in the system if the specified node |
| 767 | * can not hold the requested memory. |
| 768 | * |
| 769 | * The function panics if the request can not be satisfied. |
| 770 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 771 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 772 | unsigned long align, unsigned long goal) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 773 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 774 | if (WARN_ON_ONCE(slab_is_available())) |
| 775 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 776 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 777 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 778 | } |
| 779 | |
| 780 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, |
| 781 | unsigned long align, unsigned long goal) |
| 782 | { |
| 783 | #ifdef MAX_DMA32_PFN |
| 784 | unsigned long end_pfn; |
| 785 | |
| 786 | if (WARN_ON_ONCE(slab_is_available())) |
| 787 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 788 | |
| 789 | /* update goal according ...MAX_DMA32_PFN */ |
Xishi Qiu | 83285c7 | 2013-11-12 15:07:19 -0800 | [diff] [blame] | 790 | end_pfn = pgdat_end_pfn(pgdat); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 791 | |
| 792 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && |
| 793 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { |
| 794 | void *ptr; |
| 795 | unsigned long new_goal; |
| 796 | |
| 797 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; |
Johannes Weiner | c6785b6 | 2012-05-29 15:06:33 -0700 | [diff] [blame] | 798 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 799 | new_goal, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 800 | if (ptr) |
| 801 | return ptr; |
| 802 | } |
| 803 | #endif |
| 804 | |
| 805 | return __alloc_bootmem_node(pgdat, size, align, goal); |
| 806 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | } |
| 808 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 809 | /** |
| 810 | * __alloc_bootmem_low - allocate low boot memory |
| 811 | * @size: size of the request in bytes |
| 812 | * @align: alignment of the region |
| 813 | * @goal: preferred starting address of the region |
| 814 | * |
| 815 | * The goal is dropped if it can not be satisfied and the allocation will |
| 816 | * fall back to memory below @goal. |
| 817 | * |
| 818 | * Allocation may happen on any node in the system. |
| 819 | * |
| 820 | * The function panics if the request can not be satisfied. |
| 821 | */ |
Franck Bui-Huu | bb0923a | 2006-09-25 23:31:05 -0700 | [diff] [blame] | 822 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
| 823 | unsigned long goal) |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 824 | { |
Johannes Weiner | 0f3caba | 2008-07-23 21:28:07 -0700 | [diff] [blame] | 825 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 826 | } |
| 827 | |
Yinghai Lu | 38fa417 | 2013-01-24 12:20:15 -0800 | [diff] [blame] | 828 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
| 829 | unsigned long align, |
| 830 | unsigned long goal) |
| 831 | { |
| 832 | return ___alloc_bootmem_nopanic(size, align, goal, |
| 833 | ARCH_LOW_ADDRESS_LIMIT); |
| 834 | } |
| 835 | |
Johannes Weiner | a66fd7d | 2008-07-23 21:28:01 -0700 | [diff] [blame] | 836 | /** |
| 837 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node |
| 838 | * @pgdat: node to allocate from |
| 839 | * @size: size of the request in bytes |
| 840 | * @align: alignment of the region |
| 841 | * @goal: preferred starting address of the region |
| 842 | * |
| 843 | * The goal is dropped if it can not be satisfied and the allocation will |
| 844 | * fall back to memory below @goal. |
| 845 | * |
| 846 | * Allocation may fall back to any node in the system if the specified node |
| 847 | * can not hold the requested memory. |
| 848 | * |
| 849 | * The function panics if the request can not be satisfied. |
| 850 | */ |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 851 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
| 852 | unsigned long align, unsigned long goal) |
| 853 | { |
Pekka Enberg | c91c477 | 2009-06-11 08:10:28 +0300 | [diff] [blame] | 854 | if (WARN_ON_ONCE(slab_is_available())) |
| 855 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 856 | |
Johannes Weiner | e907991 | 2012-05-29 15:06:36 -0700 | [diff] [blame] | 857 | return ___alloc_bootmem_node(pgdat, size, align, |
| 858 | goal, ARCH_LOW_ADDRESS_LIMIT); |
Ravikiran G Thirumalai | 008857c | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 859 | } |