Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | #ifndef __LINUX_GFP_H |
| 3 | #define __LINUX_GFP_H |
| 4 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 5 | #include <linux/mmdebug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6 | #include <linux/mmzone.h> |
| 7 | #include <linux/stddef.h> |
| 8 | #include <linux/linkage.h> |
Rusty Russell | 082edb7 | 2009-03-13 23:43:37 +1030 | [diff] [blame] | 9 | #include <linux/topology.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | |
| 11 | struct vm_area_struct; |
| 12 | |
Vlastimil Babka | 1f7866b | 2016-03-15 14:55:45 -0700 | [diff] [blame] | 13 | /* |
| 14 | * In case of changes, please don't forget to update |
Vlastimil Babka | 420adbe9 | 2016-03-15 14:55:52 -0700 | [diff] [blame] | 15 | * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c |
Vlastimil Babka | 1f7866b | 2016-03-15 14:55:45 -0700 | [diff] [blame] | 16 | */ |
| 17 | |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 18 | /* Plain integer GFP bitmasks. Do not use this directly. */ |
| 19 | #define ___GFP_DMA 0x01u |
| 20 | #define ___GFP_HIGHMEM 0x02u |
| 21 | #define ___GFP_DMA32 0x04u |
| 22 | #define ___GFP_MOVABLE 0x08u |
Mel Gorman | 016c13d | 2015-11-06 16:28:18 -0800 | [diff] [blame] | 23 | #define ___GFP_RECLAIMABLE 0x10u |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 24 | #define ___GFP_HIGH 0x20u |
| 25 | #define ___GFP_IO 0x40u |
| 26 | #define ___GFP_FS 0x80u |
Shakeel Butt | e67d4ca | 2018-06-07 17:08:11 -0700 | [diff] [blame] | 27 | #define ___GFP_WRITE 0x100u |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 28 | #define ___GFP_NOWARN 0x200u |
Michal Hocko | dcda9b0 | 2017-07-12 14:36:45 -0700 | [diff] [blame] | 29 | #define ___GFP_RETRY_MAYFAIL 0x400u |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 30 | #define ___GFP_NOFAIL 0x800u |
| 31 | #define ___GFP_NORETRY 0x1000u |
Mel Gorman | b37f1dd | 2012-07-31 16:44:03 -0700 | [diff] [blame] | 32 | #define ___GFP_MEMALLOC 0x2000u |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 33 | #define ___GFP_COMP 0x4000u |
| 34 | #define ___GFP_ZERO 0x8000u |
| 35 | #define ___GFP_NOMEMALLOC 0x10000u |
| 36 | #define ___GFP_HARDWALL 0x20000u |
| 37 | #define ___GFP_THISNODE 0x40000u |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 38 | #define ___GFP_ATOMIC 0x80000u |
Vladimir Davydov | a9bb7e6 | 2016-01-14 15:18:12 -0800 | [diff] [blame] | 39 | #define ___GFP_ACCOUNT 0x100000u |
Shakeel Butt | e67d4ca | 2018-06-07 17:08:11 -0700 | [diff] [blame] | 40 | #define ___GFP_DIRECT_RECLAIM 0x200000u |
| 41 | #define ___GFP_KSWAPD_RECLAIM 0x400000u |
Heesub Shin | 46f8fca | 2013-01-07 11:10:13 +0900 | [diff] [blame] | 42 | #define ___GFP_CMA 0x800000u |
Michal Hocko | 7e78442 | 2017-05-03 14:53:09 -0700 | [diff] [blame] | 43 | #ifdef CONFIG_LOCKDEP |
Heesub Shin | 46f8fca | 2013-01-07 11:10:13 +0900 | [diff] [blame] | 44 | #define ___GFP_NOLOCKDEP 0x1000000u |
Michal Hocko | 7e78442 | 2017-05-03 14:53:09 -0700 | [diff] [blame] | 45 | #else |
| 46 | #define ___GFP_NOLOCKDEP 0 |
| 47 | #endif |
Andrew Morton | 05b0afd | 2012-12-12 13:51:56 -0800 | [diff] [blame] | 48 | /* If the above are modified, __GFP_BITS_SHIFT may need updating */ |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 49 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | /* |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 51 | * Physical address zone modifiers (see linux/mmzone.h - low four bits) |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 52 | * |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 53 | * Do not put any conditional on these. If necessary modify the definitions |
matt mooney | 263ff5d | 2010-05-24 14:32:44 -0700 | [diff] [blame] | 54 | * without the underscores and use them consistently. The definitions here may |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 55 | * be used in bit comparisons. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | */ |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 57 | #define __GFP_DMA ((__force gfp_t)___GFP_DMA) |
| 58 | #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) |
| 59 | #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 60 | #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ |
Heesub Shin | 46f8fca | 2013-01-07 11:10:13 +0900 | [diff] [blame] | 61 | #define __GFP_CMA ((__force gfp_t)___GFP_CMA) |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 62 | #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 63 | |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 64 | /** |
| 65 | * DOC: Page mobility and placement hints |
| 66 | * |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 67 | * Page mobility and placement hints |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 68 | * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 69 | * |
| 70 | * These flags provide hints about how mobile the page is. Pages with similar |
| 71 | * mobility are placed within the same pageblocks to minimise problems due |
| 72 | * to external fragmentation. |
| 73 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 74 | * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be |
| 75 | * moved by page migration during memory compaction or can be reclaimed. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 76 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 77 | * %__GFP_RECLAIMABLE is used for slab allocations that specify |
| 78 | * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 79 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 80 | * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, |
| 81 | * these pages will be spread between local zones to avoid all the dirty |
| 82 | * pages being in one zone (fair zone allocation policy). |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 83 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 84 | * %__GFP_HARDWALL enforces the cpuset memory allocation policy. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 85 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 86 | * %__GFP_THISNODE forces the allocation to be satisified from the requested |
| 87 | * node with no fallbacks or placement policy enforcements. |
Vladimir Davydov | a9bb7e6 | 2016-01-14 15:18:12 -0800 | [diff] [blame] | 88 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 89 | * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 90 | */ |
| 91 | #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) |
| 92 | #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) |
| 93 | #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) |
| 94 | #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) |
Vladimir Davydov | a9bb7e6 | 2016-01-14 15:18:12 -0800 | [diff] [blame] | 95 | #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 96 | |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 97 | /** |
| 98 | * DOC: Watermark modifiers |
| 99 | * |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 100 | * Watermark modifiers -- controls access to emergency reserves |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 101 | * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 102 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 103 | * %__GFP_HIGH indicates that the caller is high-priority and that granting |
| 104 | * the request is necessary before the system can make forward progress. |
| 105 | * For example, creating an IO context to clean pages. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 106 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 107 | * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is |
| 108 | * high priority. Users are typically interrupt handlers. This may be |
| 109 | * used in conjunction with %__GFP_HIGH |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 110 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 111 | * %__GFP_MEMALLOC allows access to all memory. This should only be used when |
| 112 | * the caller guarantees the allocation will allow more memory to be freed |
| 113 | * very shortly e.g. process exiting or swapping. Users either should |
| 114 | * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 115 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 116 | * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. |
| 117 | * This takes precedence over the %__GFP_MEMALLOC flag if both are set. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 118 | */ |
| 119 | #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) |
| 120 | #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) |
| 121 | #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) |
| 122 | #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 123 | |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 124 | /** |
| 125 | * DOC: Reclaim modifiers |
| 126 | * |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 127 | * Reclaim modifiers |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 128 | * ~~~~~~~~~~~~~~~~~ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 129 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 130 | * %__GFP_IO can start physical IO. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 131 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 132 | * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the |
| 133 | * allocator recursing into the filesystem which might already be holding |
| 134 | * locks. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 135 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 136 | * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. |
| 137 | * This flag can be cleared to avoid unnecessary delays when a fallback |
| 138 | * option is available. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 139 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 140 | * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when |
| 141 | * the low watermark is reached and have it reclaim pages until the high |
| 142 | * watermark is reached. A caller may wish to clear this flag when fallback |
| 143 | * options are available and the reclaim is likely to disrupt the system. The |
| 144 | * canonical example is THP allocation where a fallback is cheap but |
| 145 | * reclaim/compaction may cause indirect stalls. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 146 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 147 | * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | * |
Michal Hocko | dcda9b0 | 2017-07-12 14:36:45 -0700 | [diff] [blame] | 149 | * The default allocator behavior depends on the request size. We have a concept |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 150 | * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). |
Michal Hocko | dcda9b0 | 2017-07-12 14:36:45 -0700 | [diff] [blame] | 151 | * !costly allocations are too essential to fail so they are implicitly |
| 152 | * non-failing by default (with some exceptions like OOM victims might fail so |
| 153 | * the caller still has to check for failures) while costly requests try to be |
| 154 | * not disruptive and back off even without invoking the OOM killer. |
| 155 | * The following three modifiers might be used to override some of these |
| 156 | * implicit rules |
| 157 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 158 | * %__GFP_NORETRY: The VM implementation will try only very lightweight |
| 159 | * memory direct reclaim to get some memory under memory pressure (thus |
| 160 | * it can sleep). It will avoid disruptive actions like OOM killer. The |
| 161 | * caller must handle the failure which is quite likely to happen under |
| 162 | * heavy memory pressure. The flag is suitable when failure can easily be |
| 163 | * handled at small cost, such as reduced throughput |
Michal Hocko | dcda9b0 | 2017-07-12 14:36:45 -0700 | [diff] [blame] | 164 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 165 | * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim |
| 166 | * procedures that have previously failed if there is some indication |
| 167 | * that progress has been made else where. It can wait for other |
| 168 | * tasks to attempt high level approaches to freeing memory such as |
| 169 | * compaction (which removes fragmentation) and page-out. |
| 170 | * There is still a definite limit to the number of retries, but it is |
| 171 | * a larger limit than with %__GFP_NORETRY. |
| 172 | * Allocations with this flag may fail, but only when there is |
| 173 | * genuinely little unused memory. While these allocations do not |
| 174 | * directly trigger the OOM killer, their failure indicates that |
| 175 | * the system is likely to need to use the OOM killer soon. The |
| 176 | * caller must handle failure, but can reasonably do so by failing |
| 177 | * a higher-level request, or completing it only in a much less |
| 178 | * efficient manner. |
| 179 | * If the allocation does fail, and the caller is in a position to |
| 180 | * free some non-essential memory, doing so could benefit the system |
| 181 | * as a whole. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 183 | * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller |
| 184 | * cannot handle allocation failures. The allocation could block |
| 185 | * indefinitely but will never return with failure. Testing for |
| 186 | * failure is pointless. |
| 187 | * New users should be evaluated carefully (and the flag should be |
| 188 | * used only when there is no reasonable failure policy) but it is |
| 189 | * definitely preferable to use the flag rather than opencode endless |
| 190 | * loop around allocator. |
| 191 | * Using this flag for costly allocations is _highly_ discouraged. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | */ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 193 | #define __GFP_IO ((__force gfp_t)___GFP_IO) |
| 194 | #define __GFP_FS ((__force gfp_t)___GFP_FS) |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 195 | #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ |
| 196 | #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 197 | #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) |
Michal Hocko | dcda9b0 | 2017-07-12 14:36:45 -0700 | [diff] [blame] | 198 | #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 199 | #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) |
| 200 | #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 201 | |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 202 | /** |
| 203 | * DOC: Action modifiers |
| 204 | * |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 205 | * Action modifiers |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 206 | * ~~~~~~~~~~~~~~~~ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 207 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 208 | * %__GFP_NOWARN suppresses allocation failure reports. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 209 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 210 | * %__GFP_COMP address compound page metadata. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 211 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 212 | * %__GFP_ZERO returns a zeroed page on success. |
Vegard Nossum | 2dff440 | 2008-05-31 15:56:17 +0200 | [diff] [blame] | 213 | */ |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 214 | #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) |
| 215 | #define __GFP_COMP ((__force gfp_t)___GFP_COMP) |
| 216 | #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) |
Vegard Nossum | 2dff440 | 2008-05-31 15:56:17 +0200 | [diff] [blame] | 217 | |
Michal Hocko | 7e78442 | 2017-05-03 14:53:09 -0700 | [diff] [blame] | 218 | /* Disable lockdep for GFP context tracking */ |
| 219 | #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) |
| 220 | |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 221 | /* Room for N __GFP_FOO bits */ |
Heesub Shin | 46f8fca | 2013-01-07 11:10:13 +0900 | [diff] [blame] | 222 | #define __GFP_BITS_SHIFT (24 + IS_ENABLED(CONFIG_LOCKDEP)) |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 223 | #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 225 | /** |
| 226 | * DOC: Useful GFP flag combinations |
| 227 | * |
| 228 | * Useful GFP flag combinations |
| 229 | * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 230 | * |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 231 | * Useful GFP flag combinations that are commonly used. It is recommended |
| 232 | * that subsystems start with one of these combinations and then set/clear |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 233 | * %__GFP_FOO flags as necessary. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 234 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 235 | * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower |
| 236 | * watermark is applied to allow access to "atomic reserves" |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 237 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 238 | * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires |
| 239 | * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 240 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 241 | * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is |
| 242 | * accounted to kmemcg. |
Vladimir Davydov | a9bb7e6 | 2016-01-14 15:18:12 -0800 | [diff] [blame] | 243 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 244 | * %GFP_NOWAIT is for kernel allocations that should not stall for direct |
| 245 | * reclaim, start physical IO or use any filesystem callback. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 246 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 247 | * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages |
| 248 | * that do not require the starting of any physical IO. |
| 249 | * Please try to avoid using this flag directly and instead use |
| 250 | * memalloc_noio_{save,restore} to mark the whole scope which cannot |
| 251 | * perform any IO with a short explanation why. All allocation requests |
| 252 | * will inherit GFP_NOIO implicitly. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 253 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 254 | * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. |
| 255 | * Please try to avoid using this flag directly and instead use |
| 256 | * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't |
| 257 | * recurse into the FS layer with a short explanation why. All allocation |
| 258 | * requests will inherit GFP_NOFS implicitly. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 259 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 260 | * %GFP_USER is for userspace allocations that also need to be directly |
| 261 | * accessibly by the kernel or hardware. It is typically used by hardware |
| 262 | * for buffers that are mapped to userspace (e.g. graphics) that hardware |
| 263 | * still must DMA to. cpuset limits are enforced for these allocations. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 264 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 265 | * %GFP_DMA exists for historical reasons and should be avoided where possible. |
| 266 | * The flags indicates that the caller requires that the lowest zone be |
| 267 | * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but |
| 268 | * it would require careful auditing as some users really require it and |
| 269 | * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the |
| 270 | * lowest zone as a type of emergency reserve. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 271 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 272 | * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit |
| 273 | * address. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 274 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 275 | * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, |
| 276 | * do not need to be directly accessible by the kernel but that cannot |
| 277 | * move once in use. An example may be a hardware allocation that maps |
| 278 | * data directly into userspace but has no addressing limitations. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 279 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 280 | * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not |
| 281 | * need direct access to but can use kmap() when access is required. They |
| 282 | * are expected to be movable via page reclaim or page migration. Typically, |
| 283 | * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 284 | * |
Mike Rapoport | 263fade | 2018-08-23 17:01:15 -0700 | [diff] [blame] | 285 | * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They |
| 286 | * are compound allocations that will generally fail quickly if memory is not |
| 287 | * available and will not wake kswapd/kcompactd on failure. The _LIGHT |
| 288 | * version does not attempt reclaim/compaction at all and is by default used |
| 289 | * in page fault path, while the non-light is used by khugepaged. |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 290 | */ |
| 291 | #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 292 | #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) |
Vladimir Davydov | a9bb7e6 | 2016-01-14 15:18:12 -0800 | [diff] [blame] | 293 | #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 294 | #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 295 | #define GFP_NOIO (__GFP_RECLAIM) |
| 296 | #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 297 | #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 298 | #define GFP_DMA __GFP_DMA |
| 299 | #define GFP_DMA32 __GFP_DMA32 |
Jianyu Zhan | 2d48366 | 2014-12-12 16:55:43 -0800 | [diff] [blame] | 300 | #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) |
| 301 | #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE) |
Vlastimil Babka | 2516035 | 2016-07-28 15:49:25 -0700 | [diff] [blame] | 302 | #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ |
| 303 | __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) |
| 304 | #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 306 | /* Convert GFP flags to their corresponding migrate type */ |
Mel Gorman | e12ba74 | 2007-10-16 01:25:52 -0700 | [diff] [blame] | 307 | #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) |
Mel Gorman | 016c13d | 2015-11-06 16:28:18 -0800 | [diff] [blame] | 308 | #define GFP_MOVABLE_SHIFT 3 |
Christoph Lameter | 6cb0622 | 2007-10-16 01:25:41 -0700 | [diff] [blame] | 309 | |
David Rientjes | 43e7a34 | 2014-10-09 15:27:25 -0700 | [diff] [blame] | 310 | static inline int gfpflags_to_migratetype(const gfp_t gfp_flags) |
Mel Gorman | 467c996 | 2007-10-16 01:26:02 -0700 | [diff] [blame] | 311 | { |
Mel Gorman | 016c13d | 2015-11-06 16:28:18 -0800 | [diff] [blame] | 312 | VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); |
| 313 | BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); |
| 314 | BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); |
Mel Gorman | 467c996 | 2007-10-16 01:26:02 -0700 | [diff] [blame] | 315 | |
| 316 | if (unlikely(page_group_by_mobility_disabled)) |
| 317 | return MIGRATE_UNMOVABLE; |
| 318 | |
| 319 | /* Group based on mobility */ |
Mel Gorman | 016c13d | 2015-11-06 16:28:18 -0800 | [diff] [blame] | 320 | return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; |
Mel Gorman | 467c996 | 2007-10-16 01:26:02 -0700 | [diff] [blame] | 321 | } |
Mel Gorman | dd56b04 | 2015-11-06 16:28:43 -0800 | [diff] [blame] | 322 | #undef GFP_MOVABLE_MASK |
| 323 | #undef GFP_MOVABLE_SHIFT |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 324 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 325 | static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) |
| 326 | { |
Joshua Clayton | 543dfb2 | 2016-01-14 15:22:10 -0800 | [diff] [blame] | 327 | return !!(gfp_flags & __GFP_DIRECT_RECLAIM); |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 328 | } |
| 329 | |
Tejun Heo | 1d5cb12 | 2019-10-24 13:50:27 -0700 | [diff] [blame] | 330 | /** |
| 331 | * gfpflags_normal_context - is gfp_flags a normal sleepable context? |
| 332 | * @gfp_flags: gfp_flags to test |
| 333 | * |
| 334 | * Test whether @gfp_flags indicates that the allocation is from the |
| 335 | * %current context and allowed to sleep. |
| 336 | * |
| 337 | * An allocation being allowed to block doesn't mean it owns the %current |
| 338 | * context. When direct reclaim path tries to allocate memory, the |
| 339 | * allocation context is nested inside whatever %current was doing at the |
| 340 | * time of the original allocation. The nested allocation may be allowed |
| 341 | * to block but modifying anything %current owns can corrupt the outer |
| 342 | * context's expectations. |
| 343 | * |
| 344 | * %true result from this function indicates that the allocation context |
| 345 | * can sleep and use anything that's associated with %current. |
| 346 | */ |
| 347 | static inline bool gfpflags_normal_context(const gfp_t gfp_flags) |
| 348 | { |
| 349 | return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) == |
| 350 | __GFP_DIRECT_RECLAIM; |
| 351 | } |
| 352 | |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 353 | #ifdef CONFIG_HIGHMEM |
| 354 | #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM |
| 355 | #else |
| 356 | #define OPT_ZONE_HIGHMEM ZONE_NORMAL |
| 357 | #endif |
| 358 | |
| 359 | #ifdef CONFIG_ZONE_DMA |
| 360 | #define OPT_ZONE_DMA ZONE_DMA |
| 361 | #else |
| 362 | #define OPT_ZONE_DMA ZONE_NORMAL |
| 363 | #endif |
| 364 | |
| 365 | #ifdef CONFIG_ZONE_DMA32 |
| 366 | #define OPT_ZONE_DMA32 ZONE_DMA32 |
| 367 | #else |
| 368 | #define OPT_ZONE_DMA32 ZONE_NORMAL |
| 369 | #endif |
| 370 | |
| 371 | /* |
| 372 | * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the |
Hao Lee | ac2e8e4 | 2017-05-03 14:54:51 -0700 | [diff] [blame] | 373 | * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT |
| 374 | * bits long and there are 16 of them to cover all possible combinations of |
matt mooney | 263ff5d | 2010-05-24 14:32:44 -0700 | [diff] [blame] | 375 | * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 376 | * |
| 377 | * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. |
| 378 | * But GFP_MOVABLE is not only a zone specifier but also an allocation |
| 379 | * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. |
matt mooney | 263ff5d | 2010-05-24 14:32:44 -0700 | [diff] [blame] | 380 | * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 381 | * |
| 382 | * bit result |
| 383 | * ================= |
| 384 | * 0x0 => NORMAL |
| 385 | * 0x1 => DMA or NORMAL |
| 386 | * 0x2 => HIGHMEM or NORMAL |
| 387 | * 0x3 => BAD (DMA+HIGHMEM) |
Huaisheng Ye | 4b33b69 | 2018-06-07 17:09:36 -0700 | [diff] [blame] | 388 | * 0x4 => DMA32 or NORMAL |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 389 | * 0x5 => BAD (DMA+DMA32) |
| 390 | * 0x6 => BAD (HIGHMEM+DMA32) |
| 391 | * 0x7 => BAD (HIGHMEM+DMA32+DMA) |
| 392 | * 0x8 => NORMAL (MOVABLE+0) |
| 393 | * 0x9 => DMA or NORMAL (MOVABLE+DMA) |
| 394 | * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) |
| 395 | * 0xb => BAD (MOVABLE+HIGHMEM+DMA) |
Huaisheng Ye | 4b33b69 | 2018-06-07 17:09:36 -0700 | [diff] [blame] | 396 | * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 397 | * 0xd => BAD (MOVABLE+DMA32+DMA) |
| 398 | * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) |
| 399 | * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) |
| 400 | * |
Dan Williams | b11a7b9 | 2016-03-17 14:19:41 -0700 | [diff] [blame] | 401 | * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 402 | */ |
| 403 | |
Dan Williams | b11a7b9 | 2016-03-17 14:19:41 -0700 | [diff] [blame] | 404 | #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 |
| 405 | /* ZONE_DEVICE is not a valid GFP zone specifier */ |
| 406 | #define GFP_ZONES_SHIFT 2 |
| 407 | #else |
| 408 | #define GFP_ZONES_SHIFT ZONES_SHIFT |
| 409 | #endif |
| 410 | |
| 411 | #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG |
| 412 | #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 413 | #endif |
| 414 | |
| 415 | #define GFP_ZONE_TABLE ( \ |
Dan Williams | b11a7b9 | 2016-03-17 14:19:41 -0700 | [diff] [blame] | 416 | (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ |
| 417 | | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ |
| 418 | | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ |
| 419 | | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ |
| 420 | | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ |
| 421 | | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ |
| 422 | | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ |
| 423 | | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 424 | ) |
| 425 | |
| 426 | /* |
matt mooney | 263ff5d | 2010-05-24 14:32:44 -0700 | [diff] [blame] | 427 | * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 428 | * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per |
| 429 | * entry starting with bit 0. Bit is set if the combination is not |
| 430 | * allowed. |
| 431 | */ |
| 432 | #define GFP_ZONE_BAD ( \ |
Namhyung Kim | 16b56cf | 2010-10-26 14:22:04 -0700 | [diff] [blame] | 433 | 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ |
| 434 | | 1 << (___GFP_DMA | ___GFP_DMA32) \ |
| 435 | | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ |
| 436 | | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ |
| 437 | | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ |
| 438 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ |
| 439 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ |
| 440 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 441 | ) |
| 442 | |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 443 | static inline enum zone_type gfp_zone(gfp_t flags) |
Christoph Lameter | 4e4785b | 2006-09-25 23:31:17 -0700 | [diff] [blame] | 444 | { |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 445 | enum zone_type z; |
Liam Mark | 0157fbd | 2018-07-25 10:24:55 -0700 | [diff] [blame] | 446 | int bit; |
| 447 | |
| 448 | if (!IS_ENABLED(CONFIG_HIGHMEM)) { |
| 449 | if ((flags & __GFP_MOVABLE) && !(flags & __GFP_CMA)) |
| 450 | flags &= ~__GFP_HIGHMEM; |
| 451 | } |
| 452 | |
| 453 | bit = (__force int) (flags & GFP_ZONEMASK); |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 454 | |
Dan Williams | b11a7b9 | 2016-03-17 14:19:41 -0700 | [diff] [blame] | 455 | z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & |
| 456 | ((1 << GFP_ZONES_SHIFT) - 1); |
Dave Hansen | 82d4b57 | 2011-05-24 17:11:42 -0700 | [diff] [blame] | 457 | VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); |
Christoph Lameter | b70d94e | 2009-06-16 15:32:46 -0700 | [diff] [blame] | 458 | return z; |
Christoph Lameter | 4e4785b | 2006-09-25 23:31:17 -0700 | [diff] [blame] | 459 | } |
| 460 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | /* |
| 462 | * There is only one page-allocator function, and two main namespaces to |
| 463 | * it. The alloc_page*() variants return 'struct page *' and as such |
| 464 | * can allocate highmem pages, the *get*page*() variants return |
| 465 | * virtual kernel addresses to the allocated page(s). |
| 466 | */ |
| 467 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 468 | static inline int gfp_zonelist(gfp_t flags) |
| 469 | { |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 470 | #ifdef CONFIG_NUMA |
| 471 | if (unlikely(flags & __GFP_THISNODE)) |
| 472 | return ZONELIST_NOFALLBACK; |
| 473 | #endif |
| 474 | return ZONELIST_FALLBACK; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 475 | } |
| 476 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | /* |
| 478 | * We get the zone list from the current node and the gfp_mask. |
| 479 | * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones. |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 480 | * There are two zonelists per node, one for all zones with memory and |
| 481 | * one containing just zones from the node the zonelist belongs to. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | * |
| 483 | * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets |
| 484 | * optimized to &contig_page_data at compile-time. |
| 485 | */ |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 486 | static inline struct zonelist *node_zonelist(int nid, gfp_t flags) |
| 487 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 488 | return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 489 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | |
| 491 | #ifndef HAVE_ARCH_FREE_PAGE |
| 492 | static inline void arch_free_page(struct page *page, int order) { } |
| 493 | #endif |
Nick Piggin | cc102509 | 2006-12-06 20:32:00 -0800 | [diff] [blame] | 494 | #ifndef HAVE_ARCH_ALLOC_PAGE |
| 495 | static inline void arch_alloc_page(struct page *page, int order) { } |
| 496 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | |
KOSAKI Motohiro | e4048e5 | 2008-07-23 21:27:01 -0700 | [diff] [blame] | 498 | struct page * |
Vlastimil Babka | 04ec626 | 2017-07-06 15:40:03 -0700 | [diff] [blame] | 499 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
| 500 | nodemask_t *nodemask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | |
KOSAKI Motohiro | e4048e5 | 2008-07-23 21:27:01 -0700 | [diff] [blame] | 502 | static inline struct page * |
Vlastimil Babka | 04ec626 | 2017-07-06 15:40:03 -0700 | [diff] [blame] | 503 | __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid) |
KOSAKI Motohiro | e4048e5 | 2008-07-23 21:27:01 -0700 | [diff] [blame] | 504 | { |
Vlastimil Babka | 04ec626 | 2017-07-06 15:40:03 -0700 | [diff] [blame] | 505 | return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL); |
KOSAKI Motohiro | e4048e5 | 2008-07-23 21:27:01 -0700 | [diff] [blame] | 506 | } |
| 507 | |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 508 | /* |
| 509 | * Allocate pages, preferring the node given as nid. The node must be valid and |
| 510 | * online. For more general interface, see alloc_pages_node(). |
| 511 | */ |
| 512 | static inline struct page * |
| 513 | __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) |
| 514 | { |
Vlastimil Babka | 0bc35a9 | 2015-09-08 15:03:53 -0700 | [diff] [blame] | 515 | VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); |
Michal Hocko | 8addc2d | 2018-05-25 14:47:46 -0700 | [diff] [blame] | 516 | VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 517 | |
Vlastimil Babka | 04ec626 | 2017-07-06 15:40:03 -0700 | [diff] [blame] | 518 | return __alloc_pages(gfp_mask, order, nid); |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 519 | } |
| 520 | |
| 521 | /* |
| 522 | * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, |
Vlastimil Babka | 82c1fc7 | 2015-09-08 15:03:56 -0700 | [diff] [blame] | 523 | * prefer the current CPU's closest node. Otherwise node must be valid and |
| 524 | * online. |
Vlastimil Babka | 96db800 | 2015-09-08 15:03:50 -0700 | [diff] [blame] | 525 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 526 | static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | unsigned int order) |
| 528 | { |
Vlastimil Babka | 0bc35a9 | 2015-09-08 15:03:53 -0700 | [diff] [blame] | 529 | if (nid == NUMA_NO_NODE) |
Vlastimil Babka | 82c1fc7 | 2015-09-08 15:03:56 -0700 | [diff] [blame] | 530 | nid = numa_mem_id(); |
Andi Kleen | 819a692 | 2006-01-11 22:43:45 +0100 | [diff] [blame] | 531 | |
Vlastimil Babka | 0bc35a9 | 2015-09-08 15:03:53 -0700 | [diff] [blame] | 532 | return __alloc_pages_node(nid, gfp_mask, order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | } |
| 534 | |
| 535 | #ifdef CONFIG_NUMA |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 536 | extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 537 | |
| 538 | static inline struct page * |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 539 | alloc_pages(gfp_t gfp_mask, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | return alloc_pages_current(gfp_mask, order); |
| 542 | } |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 543 | extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, |
Andi Kleen | 2f5f948 | 2011-03-04 17:36:29 -0800 | [diff] [blame] | 544 | struct vm_area_struct *vma, unsigned long addr, |
Vlastimil Babka | be97a41 | 2015-02-11 15:27:15 -0800 | [diff] [blame] | 545 | int node, bool hugepage); |
| 546 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ |
| 547 | alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | #else |
| 549 | #define alloc_pages(gfp_mask, order) \ |
| 550 | alloc_pages_node(numa_node_id(), gfp_mask, order) |
Vlastimil Babka | be97a41 | 2015-02-11 15:27:15 -0800 | [diff] [blame] | 551 | #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\ |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 552 | alloc_pages(gfp_mask, order) |
Aneesh Kumar K.V | 077fcf1 | 2015-02-11 15:27:12 -0800 | [diff] [blame] | 553 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ |
| 554 | alloc_pages(gfp_mask, order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | #endif |
| 556 | #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) |
Andi Kleen | 2f5f948 | 2011-03-04 17:36:29 -0800 | [diff] [blame] | 557 | #define alloc_page_vma(gfp_mask, vma, addr) \ |
Vlastimil Babka | be97a41 | 2015-02-11 15:27:15 -0800 | [diff] [blame] | 558 | alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false) |
Andi Kleen | 236344d | 2011-03-04 17:36:30 -0800 | [diff] [blame] | 559 | #define alloc_page_vma_node(gfp_mask, vma, addr, node) \ |
Vlastimil Babka | be97a41 | 2015-02-11 15:27:15 -0800 | [diff] [blame] | 560 | alloc_pages_vma(gfp_mask, 0, vma, addr, node, false) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | |
Harvey Harrison | b3c9752 | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 562 | extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); |
| 563 | extern unsigned long get_zeroed_page(gfp_t gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 565 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask); |
| 566 | void free_pages_exact(void *virt, size_t size); |
Fabian Frederick | e193181 | 2014-08-06 16:04:59 -0700 | [diff] [blame] | 567 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask); |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 568 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | #define __get_free_page(gfp_mask) \ |
matt mooney | fd23855 | 2010-05-24 14:32:45 -0700 | [diff] [blame] | 570 | __get_free_pages((gfp_mask), 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | |
| 572 | #define __get_dma_pages(gfp_mask, order) \ |
matt mooney | fd23855 | 2010-05-24 14:32:45 -0700 | [diff] [blame] | 573 | __get_free_pages((gfp_mask) | GFP_DMA, (order)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
Harvey Harrison | b3c9752 | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 575 | extern void __free_pages(struct page *page, unsigned int order); |
| 576 | extern void free_pages(unsigned long addr, unsigned int order); |
Mel Gorman | 2d4894b | 2017-11-15 17:37:59 -0800 | [diff] [blame] | 577 | extern void free_unref_page(struct page *page); |
| 578 | extern void free_unref_page_list(struct list_head *list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 580 | struct page_frag_cache; |
Alexander Duyck | 2976db8 | 2017-01-10 16:58:09 -0800 | [diff] [blame] | 581 | extern void __page_frag_cache_drain(struct page *page, unsigned int count); |
Alexander Duyck | 8c2dd3e | 2017-01-10 16:58:06 -0800 | [diff] [blame] | 582 | extern void *page_frag_alloc(struct page_frag_cache *nc, |
| 583 | unsigned int fragsz, gfp_t gfp_mask); |
| 584 | extern void page_frag_free(void *addr); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 585 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | #define __free_page(page) __free_pages((page), 0) |
matt mooney | fd23855 | 2010-05-24 14:32:45 -0700 | [diff] [blame] | 587 | #define free_page(addr) free_pages((addr), 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | |
| 589 | void page_alloc_init(void); |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 590 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 591 | void drain_all_pages(struct zone *zone); |
| 592 | void drain_local_pages(struct zone *zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 594 | void page_alloc_init_late(void); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 595 | |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 596 | /* |
| 597 | * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what |
| 598 | * GFP flags are used before interrupts are enabled. Once interrupts are |
| 599 | * enabled, it is set to __GFP_BITS_MASK while the system is running. During |
| 600 | * hibernation, it is used by PM to avoid I/O during memory allocation while |
| 601 | * devices are suspended. |
| 602 | */ |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 603 | extern gfp_t gfp_allowed_mask; |
| 604 | |
Mel Gorman | c93bdd0 | 2012-07-31 16:44:19 -0700 | [diff] [blame] | 605 | /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ |
| 606 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); |
| 607 | |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 608 | extern void pm_restrict_gfp_mask(void); |
| 609 | extern void pm_restore_gfp_mask(void); |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 610 | |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 611 | #ifdef CONFIG_PM_SLEEP |
| 612 | extern bool pm_suspended_storage(void); |
| 613 | #else |
| 614 | static inline bool pm_suspended_storage(void) |
| 615 | { |
| 616 | return false; |
| 617 | } |
| 618 | #endif /* CONFIG_PM_SLEEP */ |
| 619 | |
Vlastimil Babka | 080fe20 | 2016-02-05 15:36:41 -0800 | [diff] [blame] | 620 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 621 | /* The below functions must be run on a range from a single zone. */ |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 622 | extern int alloc_contig_range(unsigned long start, unsigned long end, |
Lucas Stach | ca96b62 | 2017-02-24 14:58:37 -0800 | [diff] [blame] | 623 | unsigned migratetype, gfp_t gfp_mask); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 624 | extern void free_contig_range(unsigned long pfn, unsigned nr_pages); |
Vlastimil Babka | 080fe20 | 2016-02-05 15:36:41 -0800 | [diff] [blame] | 625 | #endif |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 626 | |
Vlastimil Babka | 080fe20 | 2016-02-05 15:36:41 -0800 | [diff] [blame] | 627 | #ifdef CONFIG_CMA |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 628 | /* CMA stuff */ |
| 629 | extern void init_cma_reserved_pageblock(struct page *page); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 630 | #endif |
| 631 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | #endif /* __LINUX_GFP_H */ |