Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #ifndef _LINUX_COMPACTION_H |
| 3 | #define _LINUX_COMPACTION_H |
| 4 | |
| 5 | /* |
| 6 | * Determines how hard direct compaction should try to succeed. |
| 7 | * Lower value means higher priority, analogically to reclaim priority. |
| 8 | */ |
| 9 | enum compact_priority { |
| 10 | COMPACT_PRIO_SYNC_FULL, |
| 11 | MIN_COMPACT_PRIORITY = COMPACT_PRIO_SYNC_FULL, |
| 12 | COMPACT_PRIO_SYNC_LIGHT, |
| 13 | MIN_COMPACT_COSTLY_PRIORITY = COMPACT_PRIO_SYNC_LIGHT, |
| 14 | DEF_COMPACT_PRIORITY = COMPACT_PRIO_SYNC_LIGHT, |
| 15 | COMPACT_PRIO_ASYNC, |
| 16 | INIT_COMPACT_PRIORITY = COMPACT_PRIO_ASYNC |
| 17 | }; |
| 18 | |
| 19 | /* Return values for compact_zone() and try_to_compact_pages() */ |
| 20 | /* When adding new states, please adjust include/trace/events/compaction.h */ |
| 21 | enum compact_result { |
| 22 | /* For more detailed tracepoint output - internal to compaction */ |
| 23 | COMPACT_NOT_SUITABLE_ZONE, |
| 24 | /* |
| 25 | * compaction didn't start as it was not possible or direct reclaim |
| 26 | * was more suitable |
| 27 | */ |
| 28 | COMPACT_SKIPPED, |
| 29 | /* compaction didn't start as it was deferred due to past failures */ |
| 30 | COMPACT_DEFERRED, |
| 31 | |
| 32 | /* compaction not active last round */ |
| 33 | COMPACT_INACTIVE = COMPACT_DEFERRED, |
| 34 | |
| 35 | /* For more detailed tracepoint output - internal to compaction */ |
| 36 | COMPACT_NO_SUITABLE_PAGE, |
| 37 | /* compaction should continue to another pageblock */ |
| 38 | COMPACT_CONTINUE, |
| 39 | |
| 40 | /* |
| 41 | * The full zone was compacted scanned but wasn't successfull to compact |
| 42 | * suitable pages. |
| 43 | */ |
| 44 | COMPACT_COMPLETE, |
| 45 | /* |
| 46 | * direct compaction has scanned part of the zone but wasn't successfull |
| 47 | * to compact suitable pages. |
| 48 | */ |
| 49 | COMPACT_PARTIAL_SKIPPED, |
| 50 | |
| 51 | /* compaction terminated prematurely due to lock contentions */ |
| 52 | COMPACT_CONTENDED, |
| 53 | |
| 54 | /* |
| 55 | * direct compaction terminated after concluding that the allocation |
| 56 | * should now succeed |
| 57 | */ |
| 58 | COMPACT_SUCCESS, |
| 59 | }; |
| 60 | |
| 61 | struct alloc_context; /* in mm/internal.h */ |
| 62 | |
| 63 | /* |
| 64 | * Number of free order-0 pages that should be available above given watermark |
| 65 | * to make sure compaction has reasonable chance of not running out of free |
| 66 | * pages that it needs to isolate as migration target during its work. |
| 67 | */ |
| 68 | static inline unsigned long compact_gap(unsigned int order) |
| 69 | { |
| 70 | /* |
| 71 | * Although all the isolations for migration are temporary, compaction |
| 72 | * free scanner may have up to 1 << order pages on its list and then |
| 73 | * try to split an (order - 1) free page. At that point, a gap of |
| 74 | * 1 << order might not be enough, so it's safer to require twice that |
| 75 | * amount. Note that the number of pages on the list is also |
| 76 | * effectively limited by COMPACT_CLUSTER_MAX, as that's the maximum |
| 77 | * that the migrate scanner can have isolated on migrate list, and free |
| 78 | * scanner is only invoked when the number of isolated free pages is |
| 79 | * lower than that. But it's not worth to complicate the formula here |
| 80 | * as a bigger gap for higher orders than strictly necessary can also |
| 81 | * improve chances of compaction success. |
| 82 | */ |
| 83 | return 2UL << order; |
| 84 | } |
| 85 | |
| 86 | #ifdef CONFIG_COMPACTION |
| 87 | extern int sysctl_compact_memory; |
| 88 | extern int sysctl_compaction_handler(struct ctl_table *table, int write, |
| 89 | void __user *buffer, size_t *length, loff_t *ppos); |
| 90 | extern int sysctl_extfrag_threshold; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 91 | extern int sysctl_compact_unevictable_allowed; |
| 92 | |
| 93 | extern int fragmentation_index(struct zone *zone, unsigned int order); |
| 94 | extern enum compact_result try_to_compact_pages(gfp_t gfp_mask, |
| 95 | unsigned int order, unsigned int alloc_flags, |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 96 | const struct alloc_context *ac, enum compact_priority prio, |
| 97 | struct page **page); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 98 | extern void reset_isolation_suitable(pg_data_t *pgdat); |
| 99 | extern enum compact_result compaction_suitable(struct zone *zone, int order, |
| 100 | unsigned int alloc_flags, int classzone_idx); |
| 101 | |
| 102 | extern void defer_compaction(struct zone *zone, int order); |
| 103 | extern bool compaction_deferred(struct zone *zone, int order); |
| 104 | extern void compaction_defer_reset(struct zone *zone, int order, |
| 105 | bool alloc_success); |
| 106 | extern bool compaction_restarting(struct zone *zone, int order); |
| 107 | |
| 108 | /* Compaction has made some progress and retrying makes sense */ |
| 109 | static inline bool compaction_made_progress(enum compact_result result) |
| 110 | { |
| 111 | /* |
| 112 | * Even though this might sound confusing this in fact tells us |
| 113 | * that the compaction successfully isolated and migrated some |
| 114 | * pageblocks. |
| 115 | */ |
| 116 | if (result == COMPACT_SUCCESS) |
| 117 | return true; |
| 118 | |
| 119 | return false; |
| 120 | } |
| 121 | |
| 122 | /* Compaction has failed and it doesn't make much sense to keep retrying. */ |
| 123 | static inline bool compaction_failed(enum compact_result result) |
| 124 | { |
| 125 | /* All zones were scanned completely and still not result. */ |
| 126 | if (result == COMPACT_COMPLETE) |
| 127 | return true; |
| 128 | |
| 129 | return false; |
| 130 | } |
| 131 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 132 | /* Compaction needs reclaim to be performed first, so it can continue. */ |
| 133 | static inline bool compaction_needs_reclaim(enum compact_result result) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 134 | { |
| 135 | /* |
| 136 | * Compaction backed off due to watermark checks for order-0 |
| 137 | * so the regular reclaim has to try harder and reclaim something. |
| 138 | */ |
| 139 | if (result == COMPACT_SKIPPED) |
| 140 | return true; |
| 141 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 142 | return false; |
| 143 | } |
| 144 | |
| 145 | /* |
| 146 | * Compaction has backed off for some reason after doing some work or none |
| 147 | * at all. It might be throttling or lock contention. Retrying might be still |
| 148 | * worthwhile, but with a higher priority if allowed. |
| 149 | */ |
| 150 | static inline bool compaction_withdrawn(enum compact_result result) |
| 151 | { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 152 | /* |
| 153 | * If compaction is deferred for high-order allocations, it is |
| 154 | * because sync compaction recently failed. If this is the case |
| 155 | * and the caller requested a THP allocation, we do not want |
| 156 | * to heavily disrupt the system, so we fail the allocation |
| 157 | * instead of entering direct reclaim. |
| 158 | */ |
| 159 | if (result == COMPACT_DEFERRED) |
| 160 | return true; |
| 161 | |
| 162 | /* |
| 163 | * If compaction in async mode encounters contention or blocks higher |
| 164 | * priority task we back off early rather than cause stalls. |
| 165 | */ |
| 166 | if (result == COMPACT_CONTENDED) |
| 167 | return true; |
| 168 | |
| 169 | /* |
| 170 | * Page scanners have met but we haven't scanned full zones so this |
| 171 | * is a back off in fact. |
| 172 | */ |
| 173 | if (result == COMPACT_PARTIAL_SKIPPED) |
| 174 | return true; |
| 175 | |
| 176 | return false; |
| 177 | } |
| 178 | |
| 179 | |
| 180 | bool compaction_zonelist_suitable(struct alloc_context *ac, int order, |
| 181 | int alloc_flags); |
| 182 | |
| 183 | extern int kcompactd_run(int nid); |
| 184 | extern void kcompactd_stop(int nid); |
| 185 | extern void wakeup_kcompactd(pg_data_t *pgdat, int order, int classzone_idx); |
| 186 | |
| 187 | #else |
| 188 | static inline void reset_isolation_suitable(pg_data_t *pgdat) |
| 189 | { |
| 190 | } |
| 191 | |
| 192 | static inline enum compact_result compaction_suitable(struct zone *zone, int order, |
| 193 | int alloc_flags, int classzone_idx) |
| 194 | { |
| 195 | return COMPACT_SKIPPED; |
| 196 | } |
| 197 | |
| 198 | static inline void defer_compaction(struct zone *zone, int order) |
| 199 | { |
| 200 | } |
| 201 | |
| 202 | static inline bool compaction_deferred(struct zone *zone, int order) |
| 203 | { |
| 204 | return true; |
| 205 | } |
| 206 | |
| 207 | static inline bool compaction_made_progress(enum compact_result result) |
| 208 | { |
| 209 | return false; |
| 210 | } |
| 211 | |
| 212 | static inline bool compaction_failed(enum compact_result result) |
| 213 | { |
| 214 | return false; |
| 215 | } |
| 216 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 217 | static inline bool compaction_needs_reclaim(enum compact_result result) |
| 218 | { |
| 219 | return false; |
| 220 | } |
| 221 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 222 | static inline bool compaction_withdrawn(enum compact_result result) |
| 223 | { |
| 224 | return true; |
| 225 | } |
| 226 | |
| 227 | static inline int kcompactd_run(int nid) |
| 228 | { |
| 229 | return 0; |
| 230 | } |
| 231 | static inline void kcompactd_stop(int nid) |
| 232 | { |
| 233 | } |
| 234 | |
| 235 | static inline void wakeup_kcompactd(pg_data_t *pgdat, int order, int classzone_idx) |
| 236 | { |
| 237 | } |
| 238 | |
| 239 | #endif /* CONFIG_COMPACTION */ |
| 240 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 241 | struct node; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 242 | #if defined(CONFIG_COMPACTION) && defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 243 | extern int compaction_register_node(struct node *node); |
| 244 | extern void compaction_unregister_node(struct node *node); |
| 245 | |
| 246 | #else |
| 247 | |
| 248 | static inline int compaction_register_node(struct node *node) |
| 249 | { |
| 250 | return 0; |
| 251 | } |
| 252 | |
| 253 | static inline void compaction_unregister_node(struct node *node) |
| 254 | { |
| 255 | } |
| 256 | #endif /* CONFIG_COMPACTION && CONFIG_SYSFS && CONFIG_NUMA */ |
| 257 | |
| 258 | #endif /* _LINUX_COMPACTION_H */ |