1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * linux/mm/page_isolation.c 3 * linux/mm/page_isolation.c
4 */ 4 */
5 5
6 #include <linux/mm.h> 6 #include <linux/mm.h>
7 #include <linux/page-isolation.h> 7 #include <linux/page-isolation.h>
8 #include <linux/pageblock-flags.h> 8 #include <linux/pageblock-flags.h>
9 #include <linux/memory.h> 9 #include <linux/memory.h>
10 #include <linux/hugetlb.h> 10 #include <linux/hugetlb.h>
11 #include <linux/page_owner.h> 11 #include <linux/page_owner.h>
12 #include <linux/migrate.h> 12 #include <linux/migrate.h>
13 #include "internal.h" 13 #include "internal.h"
14 14
15 #define CREATE_TRACE_POINTS 15 #define CREATE_TRACE_POINTS
16 #include <trace/events/page_isolation.h> 16 #include <trace/events/page_isolation.h>
17 17
18 /* 18 /*
19 * This function checks whether the range [sta 19 * This function checks whether the range [start_pfn, end_pfn) includes
20 * unmovable pages or not. The range must fall 20 * unmovable pages or not. The range must fall into a single pageblock and
21 * consequently belong to a single zone. 21 * consequently belong to a single zone.
22 * 22 *
23 * PageLRU check without isolation or lru_lock 23 * PageLRU check without isolation or lru_lock could race so that
24 * MIGRATE_MOVABLE block might include unmovab 24 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
25 * check without lock_page also may miss some 25 * check without lock_page also may miss some movable non-lru pages at
26 * race condition. So you can't expect this fu 26 * race condition. So you can't expect this function should be exact.
27 * 27 *
28 * Returns a page without holding a reference. 28 * Returns a page without holding a reference. If the caller wants to
29 * dereference that page (e.g., dumping), it h 29 * dereference that page (e.g., dumping), it has to make sure that it
30 * cannot get removed (e.g., via memory unplug 30 * cannot get removed (e.g., via memory unplug) concurrently.
31 * 31 *
32 */ 32 */
33 static struct page *has_unmovable_pages(unsign 33 static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long end_pfn,
34 int migratetyp 34 int migratetype, int flags)
35 { 35 {
36 struct page *page = pfn_to_page(start_ 36 struct page *page = pfn_to_page(start_pfn);
37 struct zone *zone = page_zone(page); 37 struct zone *zone = page_zone(page);
38 unsigned long pfn; 38 unsigned long pfn;
39 39
40 VM_BUG_ON(pageblock_start_pfn(start_pf !! 40 VM_BUG_ON(ALIGN_DOWN(start_pfn, pageblock_nr_pages) !=
41 pageblock_start_pfn(end_pfn !! 41 ALIGN_DOWN(end_pfn - 1, pageblock_nr_pages));
42 42
43 if (is_migrate_cma_page(page)) { 43 if (is_migrate_cma_page(page)) {
44 /* 44 /*
45 * CMA allocations (alloc_cont 45 * CMA allocations (alloc_contig_range) really need to mark
46 * isolate CMA pageblocks even 46 * isolate CMA pageblocks even when they are not movable in fact
47 * so consider them movable he 47 * so consider them movable here.
48 */ 48 */
49 if (is_migrate_cma(migratetype 49 if (is_migrate_cma(migratetype))
50 return NULL; 50 return NULL;
51 51
52 return page; 52 return page;
53 } 53 }
54 54
55 for (pfn = start_pfn; pfn < end_pfn; p 55 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
56 page = pfn_to_page(pfn); 56 page = pfn_to_page(pfn);
57 57
58 /* 58 /*
59 * Both, bootmem allocations a 59 * Both, bootmem allocations and memory holes are marked
60 * PG_reserved and are unmovab 60 * PG_reserved and are unmovable. We can even have unmovable
61 * allocations inside ZONE_MOV 61 * allocations inside ZONE_MOVABLE, for example when
62 * specifying "movablecore". 62 * specifying "movablecore".
63 */ 63 */
64 if (PageReserved(page)) 64 if (PageReserved(page))
65 return page; 65 return page;
66 66
67 /* 67 /*
68 * If the zone is movable and 68 * If the zone is movable and we have ruled out all reserved
69 * pages then it should be rea 69 * pages then it should be reasonably safe to assume the rest
70 * is movable. 70 * is movable.
71 */ 71 */
72 if (zone_idx(zone) == ZONE_MOV 72 if (zone_idx(zone) == ZONE_MOVABLE)
73 continue; 73 continue;
74 74
75 /* 75 /*
76 * Hugepages are not in LRU li 76 * Hugepages are not in LRU lists, but they're movable.
77 * THPs are on the LRU, but ne 77 * THPs are on the LRU, but need to be counted as #small pages.
78 * We need not scan over tail 78 * We need not scan over tail pages because we don't
79 * handle each tail page indiv 79 * handle each tail page individually in migration.
80 */ 80 */
81 if (PageHuge(page) || PageTran 81 if (PageHuge(page) || PageTransCompound(page)) {
82 struct folio *folio = !! 82 struct page *head = compound_head(page);
83 unsigned int skip_page 83 unsigned int skip_pages;
84 84
85 if (PageHuge(page)) { 85 if (PageHuge(page)) {
86 if (!hugepage_ !! 86 if (!hugepage_migration_supported(page_hstate(head)))
87 return 87 return page;
88 } else if (!folio_test !! 88 } else if (!PageLRU(head) && !__PageMovable(head)) {
89 return page; 89 return page;
90 } 90 }
91 91
92 skip_pages = folio_nr_ !! 92 skip_pages = compound_nr(head) - (page - head);
93 pfn += skip_pages - 1; 93 pfn += skip_pages - 1;
94 continue; 94 continue;
95 } 95 }
96 96
97 /* 97 /*
98 * We can't use page_count wit 98 * We can't use page_count without pin a page
99 * because another CPU can fre 99 * because another CPU can free compound page.
100 * This check already skips co 100 * This check already skips compound tails of THP
101 * because their page->_refcou 101 * because their page->_refcount is zero at all time.
102 */ 102 */
103 if (!page_ref_count(page)) { 103 if (!page_ref_count(page)) {
104 if (PageBuddy(page)) 104 if (PageBuddy(page))
105 pfn += (1 << b 105 pfn += (1 << buddy_order(page)) - 1;
106 continue; 106 continue;
107 } 107 }
108 108
109 /* 109 /*
110 * The HWPoisoned page may be 110 * The HWPoisoned page may be not in buddy system, and
111 * page_count() is not 0. 111 * page_count() is not 0.
112 */ 112 */
113 if ((flags & MEMORY_OFFLINE) & 113 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
114 continue; 114 continue;
115 115
116 /* 116 /*
117 * We treat all PageOffline() 117 * We treat all PageOffline() pages as movable when offlining
118 * to give drivers a chance to 118 * to give drivers a chance to decrement their reference count
119 * in MEM_GOING_OFFLINE in ord 119 * in MEM_GOING_OFFLINE in order to indicate that these pages
120 * can be offlined as there ar 120 * can be offlined as there are no direct references anymore.
121 * For actually unmovable Page 121 * For actually unmovable PageOffline() where the driver does
122 * not support this, we will f 122 * not support this, we will fail later when trying to actually
123 * move these pages that still 123 * move these pages that still have a reference count > 0.
124 * (false negatives in this fu 124 * (false negatives in this function only)
125 */ 125 */
126 if ((flags & MEMORY_OFFLINE) & 126 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
127 continue; 127 continue;
128 128
129 if (__PageMovable(page) || Pag 129 if (__PageMovable(page) || PageLRU(page))
130 continue; 130 continue;
131 131
132 /* 132 /*
133 * If there are RECLAIMABLE pa 133 * If there are RECLAIMABLE pages, we need to check
134 * it. But now, memory offlin 134 * it. But now, memory offline itself doesn't call
135 * shrink_node_slabs() and it 135 * shrink_node_slabs() and it still to be fixed.
136 */ 136 */
137 return page; 137 return page;
138 } 138 }
139 return NULL; 139 return NULL;
140 } 140 }
141 141
142 /* 142 /*
143 * This function set pageblock migratetype to 143 * This function set pageblock migratetype to isolate if no unmovable page is
144 * present in [start_pfn, end_pfn). The pagebl 144 * present in [start_pfn, end_pfn). The pageblock must intersect with
145 * [start_pfn, end_pfn). 145 * [start_pfn, end_pfn).
146 */ 146 */
147 static int set_migratetype_isolate(struct page 147 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags,
148 unsigned long start_pf 148 unsigned long start_pfn, unsigned long end_pfn)
149 { 149 {
150 struct zone *zone = page_zone(page); 150 struct zone *zone = page_zone(page);
151 struct page *unmovable; 151 struct page *unmovable;
152 unsigned long flags; 152 unsigned long flags;
153 unsigned long check_unmovable_start, c 153 unsigned long check_unmovable_start, check_unmovable_end;
154 154
155 if (PageUnaccepted(page)) <<
156 accept_page(page); <<
157 <<
158 spin_lock_irqsave(&zone->lock, flags); 155 spin_lock_irqsave(&zone->lock, flags);
159 156
160 /* 157 /*
161 * We assume the caller intended to SE 158 * We assume the caller intended to SET migrate type to isolate.
162 * If it is already set, then someone 159 * If it is already set, then someone else must have raced and
163 * set it before us. 160 * set it before us.
164 */ 161 */
165 if (is_migrate_isolate_page(page)) { 162 if (is_migrate_isolate_page(page)) {
166 spin_unlock_irqrestore(&zone-> 163 spin_unlock_irqrestore(&zone->lock, flags);
167 return -EBUSY; 164 return -EBUSY;
168 } 165 }
169 166
170 /* 167 /*
171 * FIXME: Now, memory hotplug doesn't 168 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
172 * We just check MOVABLE pages. 169 * We just check MOVABLE pages.
173 * 170 *
174 * Pass the intersection of [start_pfn 171 * Pass the intersection of [start_pfn, end_pfn) and the page's pageblock
175 * to avoid redundant checks. 172 * to avoid redundant checks.
176 */ 173 */
177 check_unmovable_start = max(page_to_pf 174 check_unmovable_start = max(page_to_pfn(page), start_pfn);
178 check_unmovable_end = min(pageblock_en !! 175 check_unmovable_end = min(ALIGN(page_to_pfn(page) + 1, pageblock_nr_pages),
179 end_pfn); 176 end_pfn);
180 177
181 unmovable = has_unmovable_pages(check_ 178 unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end,
182 migratetype, isol_flag 179 migratetype, isol_flags);
183 if (!unmovable) { 180 if (!unmovable) {
184 if (!move_freepages_block_isol !! 181 unsigned long nr_pages;
185 spin_unlock_irqrestore !! 182 int mt = get_pageblock_migratetype(page);
186 return -EBUSY; !! 183
187 } !! 184 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
188 zone->nr_isolate_pageblock++; 185 zone->nr_isolate_pageblock++;
>> 186 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
>> 187 NULL);
>> 188
>> 189 __mod_zone_freepage_state(zone, -nr_pages, mt);
189 spin_unlock_irqrestore(&zone-> 190 spin_unlock_irqrestore(&zone->lock, flags);
190 return 0; 191 return 0;
191 } 192 }
192 193
193 spin_unlock_irqrestore(&zone->lock, fl 194 spin_unlock_irqrestore(&zone->lock, flags);
194 if (isol_flags & REPORT_FAILURE) { 195 if (isol_flags & REPORT_FAILURE) {
195 /* 196 /*
196 * printk() with zone->lock he 197 * printk() with zone->lock held will likely trigger a
197 * lockdep splat, so defer it 198 * lockdep splat, so defer it here.
198 */ 199 */
199 dump_page(unmovable, "unmovabl 200 dump_page(unmovable, "unmovable page");
200 } 201 }
201 202
202 return -EBUSY; 203 return -EBUSY;
203 } 204 }
204 205
205 static void unset_migratetype_isolate(struct p 206 static void unset_migratetype_isolate(struct page *page, int migratetype)
206 { 207 {
207 struct zone *zone; 208 struct zone *zone;
208 unsigned long flags; !! 209 unsigned long flags, nr_pages;
209 bool isolated_page = false; 210 bool isolated_page = false;
210 unsigned int order; 211 unsigned int order;
211 struct page *buddy; 212 struct page *buddy;
212 213
213 zone = page_zone(page); 214 zone = page_zone(page);
214 spin_lock_irqsave(&zone->lock, flags); 215 spin_lock_irqsave(&zone->lock, flags);
215 if (!is_migrate_isolate_page(page)) 216 if (!is_migrate_isolate_page(page))
216 goto out; 217 goto out;
217 218
218 /* 219 /*
219 * Because freepage with more than pag 220 * Because freepage with more than pageblock_order on isolated
220 * pageblock is restricted to merge du 221 * pageblock is restricted to merge due to freepage counting problem,
221 * it is possible that there is free b 222 * it is possible that there is free buddy page.
222 * move_freepages_block() doesn't care 223 * move_freepages_block() doesn't care of merge so we need other
223 * approach in order to merge them. Is 224 * approach in order to merge them. Isolation and free will make
224 * these pages to be merged. 225 * these pages to be merged.
225 */ 226 */
226 if (PageBuddy(page)) { 227 if (PageBuddy(page)) {
227 order = buddy_order(page); 228 order = buddy_order(page);
228 if (order >= pageblock_order & !! 229 if (order >= pageblock_order && order < MAX_ORDER - 1) {
229 buddy = find_buddy_pag 230 buddy = find_buddy_page_pfn(page, page_to_pfn(page),
230 231 order, NULL);
231 if (buddy && !is_migra 232 if (buddy && !is_migrate_isolate_page(buddy)) {
232 isolated_page 233 isolated_page = !!__isolate_free_page(page, order);
233 /* 234 /*
234 * Isolating a 235 * Isolating a free page in an isolated pageblock
235 * is expected 236 * is expected to always work as watermarks don't
236 * apply here. 237 * apply here.
237 */ 238 */
238 VM_WARN_ON(!is 239 VM_WARN_ON(!isolated_page);
239 } 240 }
240 } 241 }
241 } 242 }
242 243
243 /* 244 /*
244 * If we isolate freepage with more th 245 * If we isolate freepage with more than pageblock_order, there
245 * should be no freepage in the range, 246 * should be no freepage in the range, so we could avoid costly
246 * pageblock scanning for freepage mov 247 * pageblock scanning for freepage moving.
247 * 248 *
248 * We didn't actually touch any of the 249 * We didn't actually touch any of the isolated pages, so place them
249 * to the tail of the freelist. This i 250 * to the tail of the freelist. This is an optimization for memory
250 * onlining - just onlined memory won' 251 * onlining - just onlined memory won't immediately be considered for
251 * allocation. 252 * allocation.
252 */ 253 */
253 if (!isolated_page) { 254 if (!isolated_page) {
254 /* !! 255 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
255 * Isolating this block alread !! 256 __mod_zone_freepage_state(zone, nr_pages, migratetype);
256 * should not fail on zone bou <<
257 */ <<
258 WARN_ON_ONCE(!move_freepages_b <<
259 } else { <<
260 set_pageblock_migratetype(page <<
261 __putback_isolated_page(page, <<
262 } 257 }
>> 258 set_pageblock_migratetype(page, migratetype);
>> 259 if (isolated_page)
>> 260 __putback_isolated_page(page, order, migratetype);
263 zone->nr_isolate_pageblock--; 261 zone->nr_isolate_pageblock--;
264 out: 262 out:
265 spin_unlock_irqrestore(&zone->lock, fl 263 spin_unlock_irqrestore(&zone->lock, flags);
266 } 264 }
267 265
268 static inline struct page * 266 static inline struct page *
269 __first_valid_page(unsigned long pfn, unsigned 267 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
270 { 268 {
271 int i; 269 int i;
272 270
273 for (i = 0; i < nr_pages; i++) { 271 for (i = 0; i < nr_pages; i++) {
274 struct page *page; 272 struct page *page;
275 273
276 page = pfn_to_online_page(pfn 274 page = pfn_to_online_page(pfn + i);
277 if (!page) 275 if (!page)
278 continue; 276 continue;
279 return page; 277 return page;
280 } 278 }
281 return NULL; 279 return NULL;
282 } 280 }
283 281
284 /** 282 /**
285 * isolate_single_pageblock() -- tries to isol 283 * isolate_single_pageblock() -- tries to isolate a pageblock that might be
286 * within a free or in-use page. 284 * within a free or in-use page.
287 * @boundary_pfn: pageblock-alig 285 * @boundary_pfn: pageblock-aligned pfn that a page might cross
288 * @flags: isolation flag 286 * @flags: isolation flags
289 * @gfp_flags: GFP flags used 287 * @gfp_flags: GFP flags used for migrating pages
290 * @isolate_before: isolate the pageblock 288 * @isolate_before: isolate the pageblock before the boundary_pfn
291 * @skip_isolation: the flag to skip the p 289 * @skip_isolation: the flag to skip the pageblock isolation in second
292 * isolate_single_pageblo 290 * isolate_single_pageblock()
293 * @migratetype: migrate type to set in 291 * @migratetype: migrate type to set in error recovery.
294 * 292 *
295 * Free and in-use pages can be as big as MAX_ !! 293 * Free and in-use pages can be as big as MAX_ORDER-1 and contain more than one
296 * pageblock. When not all pageblocks within a 294 * pageblock. When not all pageblocks within a page are isolated at the same
297 * time, free page accounting can go wrong. Fo 295 * time, free page accounting can go wrong. For example, in the case of
298 * MAX_PAGE_ORDER = pageblock_order + 1, a MAX !! 296 * MAX_ORDER-1 = pageblock_order + 1, a MAX_ORDER-1 page has two pagelbocks.
299 * pagelbocks. !! 297 * [ MAX_ORDER-1 ]
300 * [ MAX_PAGE_ORDER ] <<
301 * [ pageblock0 | pageblock1 ] 298 * [ pageblock0 | pageblock1 ]
302 * When either pageblock is isolated, if it is 299 * When either pageblock is isolated, if it is a free page, the page is not
303 * split into separate migratetype lists, whic 300 * split into separate migratetype lists, which is supposed to; if it is an
304 * in-use page and freed later, __free_one_pag 301 * in-use page and freed later, __free_one_page() does not split the free page
305 * either. The function handles this by splitt 302 * either. The function handles this by splitting the free page or migrating
306 * the in-use page then splitting the free pag 303 * the in-use page then splitting the free page.
307 */ 304 */
308 static int isolate_single_pageblock(unsigned l 305 static int isolate_single_pageblock(unsigned long boundary_pfn, int flags,
309 gfp_t gfp_flags, bool 306 gfp_t gfp_flags, bool isolate_before, bool skip_isolation,
310 int migratetype) 307 int migratetype)
311 { 308 {
312 unsigned long start_pfn; 309 unsigned long start_pfn;
313 unsigned long isolate_pageblock; 310 unsigned long isolate_pageblock;
314 unsigned long pfn; 311 unsigned long pfn;
315 struct zone *zone; 312 struct zone *zone;
316 int ret; 313 int ret;
317 314
318 VM_BUG_ON(!pageblock_aligned(boundary_ !! 315 VM_BUG_ON(!IS_ALIGNED(boundary_pfn, pageblock_nr_pages));
319 316
320 if (isolate_before) 317 if (isolate_before)
321 isolate_pageblock = boundary_p 318 isolate_pageblock = boundary_pfn - pageblock_nr_pages;
322 else 319 else
323 isolate_pageblock = boundary_p 320 isolate_pageblock = boundary_pfn;
324 321
325 /* 322 /*
326 * scan at the beginning of MAX_ORDER_ 323 * scan at the beginning of MAX_ORDER_NR_PAGES aligned range to avoid
327 * only isolating a subset of pagebloc 324 * only isolating a subset of pageblocks from a bigger than pageblock
328 * free or in-use page. Also make sure 325 * free or in-use page. Also make sure all to-be-isolated pageblocks
329 * are within the same zone. 326 * are within the same zone.
330 */ 327 */
331 zone = page_zone(pfn_to_page(isolate_ 328 zone = page_zone(pfn_to_page(isolate_pageblock));
332 start_pfn = max(ALIGN_DOWN(isolate_pa 329 start_pfn = max(ALIGN_DOWN(isolate_pageblock, MAX_ORDER_NR_PAGES),
333 zone->zo 330 zone->zone_start_pfn);
334 331
335 if (skip_isolation) { 332 if (skip_isolation) {
336 int mt __maybe_unused = get_pa !! 333 int mt = get_pageblock_migratetype(pfn_to_page(isolate_pageblock));
337 334
338 VM_BUG_ON(!is_migrate_isolate( 335 VM_BUG_ON(!is_migrate_isolate(mt));
339 } else { 336 } else {
340 ret = set_migratetype_isolate( 337 ret = set_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype,
341 flags, isolate 338 flags, isolate_pageblock, isolate_pageblock + pageblock_nr_pages);
342 339
343 if (ret) 340 if (ret)
344 return ret; 341 return ret;
345 } 342 }
346 343
347 /* 344 /*
348 * Bail out early when the to-be-isola 345 * Bail out early when the to-be-isolated pageblock does not form
349 * a free or in-use page across bounda 346 * a free or in-use page across boundary_pfn:
350 * 347 *
351 * 1. isolate before boundary_pfn: the 348 * 1. isolate before boundary_pfn: the page after is not online
352 * 2. isolate after boundary_pfn: the 349 * 2. isolate after boundary_pfn: the page before is not online
353 * 350 *
354 * This also ensures correctness. With 351 * This also ensures correctness. Without it, when isolate after
355 * boundary_pfn and [start_pfn, bounda 352 * boundary_pfn and [start_pfn, boundary_pfn) are not online,
356 * __first_valid_page() will return un 353 * __first_valid_page() will return unexpected NULL in the for loop
357 * below. 354 * below.
358 */ 355 */
359 if (isolate_before) { 356 if (isolate_before) {
360 if (!pfn_to_online_page(bounda 357 if (!pfn_to_online_page(boundary_pfn))
361 return 0; 358 return 0;
362 } else { 359 } else {
363 if (!pfn_to_online_page(bounda 360 if (!pfn_to_online_page(boundary_pfn - 1))
364 return 0; 361 return 0;
365 } 362 }
366 363
367 for (pfn = start_pfn; pfn < boundary_p 364 for (pfn = start_pfn; pfn < boundary_pfn;) {
368 struct page *page = __first_va 365 struct page *page = __first_valid_page(pfn, boundary_pfn - pfn);
369 366
370 VM_BUG_ON(!page); 367 VM_BUG_ON(!page);
371 pfn = page_to_pfn(page); 368 pfn = page_to_pfn(page);
372 !! 369 /*
373 if (PageUnaccepted(page)) { !! 370 * start_pfn is MAX_ORDER_NR_PAGES aligned, if there is any
374 pfn += MAX_ORDER_NR_PA !! 371 * free pages in [start_pfn, boundary_pfn), its head page will
375 continue; !! 372 * always be in the range.
376 } !! 373 */
377 <<
378 if (PageBuddy(page)) { 374 if (PageBuddy(page)) {
379 int order = buddy_orde 375 int order = buddy_order(page);
380 376
381 /* move_freepages_bloc !! 377 if (pfn + (1UL << order) > boundary_pfn) {
382 VM_WARN_ON_ONCE(pfn + !! 378 /* free page changed before split, check it again */
>> 379 if (split_free_page(page, order, boundary_pfn - pfn))
>> 380 continue;
>> 381 }
383 382
384 pfn += 1UL << order; 383 pfn += 1UL << order;
385 continue; 384 continue;
386 } 385 }
387 <<
388 /* 386 /*
389 * If a compound page is strad !! 387 * migrate compound pages then let the free page handling code
390 * to migrate it out of the wa !! 388 * above do the rest. If migration is not possible, just fail.
391 * <<
392 * We don't have to worry abou <<
393 * free page that straddles in <<
394 * pages are freed as order-0 <<
395 * (currently) do not exceed p <<
396 * <<
397 * The block of interest has a <<
398 * MIGRATE_ISOLATE above, so w <<
399 * will free its pages onto th <<
400 */ 389 */
401 if (PageCompound(page)) { 390 if (PageCompound(page)) {
402 struct page *head = co 391 struct page *head = compound_head(page);
403 unsigned long head_pfn 392 unsigned long head_pfn = page_to_pfn(head);
404 unsigned long nr_pages 393 unsigned long nr_pages = compound_nr(head);
405 394
406 if (head_pfn + nr_page !! 395 if (head_pfn + nr_pages <= boundary_pfn) {
407 PageHuge(page)) { <<
408 pfn = head_pfn 396 pfn = head_pfn + nr_pages;
409 continue; 397 continue;
410 } 398 }
411 !! 399 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
412 /* 400 /*
413 * These pages are mov !! 401 * hugetlb, lru compound (THP), and movable compound pages
414 * not expected to exc !! 402 * can be migrated. Otherwise, fail the isolation.
415 * <<
416 * Let us know when th <<
417 * proper free and spl <<
418 */ 403 */
419 VM_WARN_ON_ONCE_PAGE(P !! 404 if (PageHuge(page) || PageLRU(page) || __PageMovable(page)) {
420 VM_WARN_ON_ONCE_PAGE(_ !! 405 int order;
>> 406 unsigned long outer_pfn;
>> 407 int page_mt = get_pageblock_migratetype(page);
>> 408 bool isolate_page = !is_migrate_isolate_page(page);
>> 409 struct compact_control cc = {
>> 410 .nr_migratepages = 0,
>> 411 .order = -1,
>> 412 .zone = page_zone(pfn_to_page(head_pfn)),
>> 413 .mode = MIGRATE_SYNC,
>> 414 .ignore_skip_hint = true,
>> 415 .no_set_skip_hint = true,
>> 416 .gfp_mask = gfp_flags,
>> 417 .alloc_contig = true,
>> 418 };
>> 419 INIT_LIST_HEAD(&cc.migratepages);
421 420
422 goto failed; !! 421 /*
>> 422 * XXX: mark the page as MIGRATE_ISOLATE so that
>> 423 * no one else can grab the freed page after migration.
>> 424 * Ideally, the page should be freed as two separate
>> 425 * pages to be added into separate migratetype free
>> 426 * lists.
>> 427 */
>> 428 if (isolate_page) {
>> 429 ret = set_migratetype_isolate(page, page_mt,
>> 430 flags, head_pfn, head_pfn + nr_pages);
>> 431 if (ret)
>> 432 goto failed;
>> 433 }
>> 434
>> 435 ret = __alloc_contig_migrate_range(&cc, head_pfn,
>> 436 head_pfn + nr_pages);
>> 437
>> 438 /*
>> 439 * restore the page's migratetype so that it can
>> 440 * be split into separate migratetype free lists
>> 441 * later.
>> 442 */
>> 443 if (isolate_page)
>> 444 unset_migratetype_isolate(page, page_mt);
>> 445
>> 446 if (ret)
>> 447 goto failed;
>> 448 /*
>> 449 * reset pfn to the head of the free page, so
>> 450 * that the free page handling code above can split
>> 451 * the free page to the right migratetype list.
>> 452 *
>> 453 * head_pfn is not used here as a hugetlb page order
>> 454 * can be bigger than MAX_ORDER-1, but after it is
>> 455 * freed, the free page order is not. Use pfn within
>> 456 * the range to find the head of the free page.
>> 457 */
>> 458 order = 0;
>> 459 outer_pfn = pfn;
>> 460 while (!PageBuddy(pfn_to_page(outer_pfn))) {
>> 461 /* stop if we cannot find the free page */
>> 462 if (++order >= MAX_ORDER)
>> 463 goto failed;
>> 464 outer_pfn &= ~0UL << order;
>> 465 }
>> 466 pfn = outer_pfn;
>> 467 continue;
>> 468 } else
>> 469 #endif
>> 470 goto failed;
423 } 471 }
424 472
425 pfn++; 473 pfn++;
426 } 474 }
427 return 0; 475 return 0;
428 failed: 476 failed:
429 /* restore the original migratetype */ 477 /* restore the original migratetype */
430 if (!skip_isolation) 478 if (!skip_isolation)
431 unset_migratetype_isolate(pfn_ 479 unset_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype);
432 return -EBUSY; 480 return -EBUSY;
433 } 481 }
434 482
435 /** 483 /**
436 * start_isolate_page_range() - mark page rang !! 484 * start_isolate_page_range() - make page-allocation-type of range of pages to
437 * @start_pfn: The first PFN of the r !! 485 * be MIGRATE_ISOLATE.
438 * @end_pfn: The last PFN of the ra !! 486 * @start_pfn: The lower PFN of the range to be isolated.
>> 487 * @end_pfn: The upper PFN of the range to be isolated.
439 * @migratetype: Migrate type to set in 488 * @migratetype: Migrate type to set in error recovery.
440 * @flags: The following flags ar 489 * @flags: The following flags are allowed (they can be combined in
441 * a bit mask) 490 * a bit mask)
442 * MEMORY_OFFLINE - isola 491 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
443 * e.g., 492 * e.g., skip over PageHWPoison() pages
444 * and P 493 * and PageOffline() pages.
445 * REPORT_FAILURE - repor 494 * REPORT_FAILURE - report details about the failure to
446 * isolate the range 495 * isolate the range
447 * @gfp_flags: GFP flags used for mig 496 * @gfp_flags: GFP flags used for migrating pages that sit across the
448 * range boundaries. 497 * range boundaries.
449 * 498 *
450 * Making page-allocation-type to be MIGRATE_I 499 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
451 * the range will never be allocated. Any free 500 * the range will never be allocated. Any free pages and pages freed in the
452 * future will not be allocated again. If spec 501 * future will not be allocated again. If specified range includes migrate types
453 * other than MOVABLE or CMA, this will fail w 502 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
454 * pages in the range finally, the caller have 503 * pages in the range finally, the caller have to free all pages in the range.
455 * test_page_isolated() can be used for test i 504 * test_page_isolated() can be used for test it.
456 * 505 *
457 * The function first tries to isolate the pag 506 * The function first tries to isolate the pageblocks at the beginning and end
458 * of the range, since there might be pages ac 507 * of the range, since there might be pages across the range boundaries.
459 * Afterwards, it isolates the rest of the ran 508 * Afterwards, it isolates the rest of the range.
460 * 509 *
461 * There is no high level synchronization mech 510 * There is no high level synchronization mechanism that prevents two threads
462 * from trying to isolate overlapping ranges. 511 * from trying to isolate overlapping ranges. If this happens, one thread
463 * will notice pageblocks in the overlapping r 512 * will notice pageblocks in the overlapping range already set to isolate.
464 * This happens in set_migratetype_isolate, an 513 * This happens in set_migratetype_isolate, and set_migratetype_isolate
465 * returns an error. We then clean up by resto 514 * returns an error. We then clean up by restoring the migration type on
466 * pageblocks we may have modified and return 515 * pageblocks we may have modified and return -EBUSY to caller. This
467 * prevents two threads from simultaneously wo 516 * prevents two threads from simultaneously working on overlapping ranges.
468 * 517 *
469 * Please note that there is no strong synchro 518 * Please note that there is no strong synchronization with the page allocator
470 * either. Pages might be freed while their pa 519 * either. Pages might be freed while their page blocks are marked ISOLATED.
471 * A call to drain_all_pages() after isolation 520 * A call to drain_all_pages() after isolation can flush most of them. However
472 * in some cases pages might still end up on p 521 * in some cases pages might still end up on pcp lists and that would allow
473 * for their allocation even when they are in 522 * for their allocation even when they are in fact isolated already. Depending
474 * on how strong of a guarantee the caller nee 523 * on how strong of a guarantee the caller needs, zone_pcp_disable/enable()
475 * might be used to flush and disable pcplist 524 * might be used to flush and disable pcplist before isolation and enable after
476 * unisolation. 525 * unisolation.
477 * 526 *
478 * Return: 0 on success and -EBUSY if any part 527 * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
479 */ 528 */
480 int start_isolate_page_range(unsigned long sta 529 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
481 int migratetype, 530 int migratetype, int flags, gfp_t gfp_flags)
482 { 531 {
483 unsigned long pfn; 532 unsigned long pfn;
484 struct page *page; 533 struct page *page;
485 /* isolation is done at page block gra 534 /* isolation is done at page block granularity */
486 unsigned long isolate_start = pagebloc !! 535 unsigned long isolate_start = ALIGN_DOWN(start_pfn, pageblock_nr_pages);
487 unsigned long isolate_end = pageblock_ !! 536 unsigned long isolate_end = ALIGN(end_pfn, pageblock_nr_pages);
488 int ret; 537 int ret;
489 bool skip_isolation = false; 538 bool skip_isolation = false;
490 539
491 /* isolate [isolate_start, isolate_sta 540 /* isolate [isolate_start, isolate_start + pageblock_nr_pages) pageblock */
492 ret = isolate_single_pageblock(isolate 541 ret = isolate_single_pageblock(isolate_start, flags, gfp_flags, false,
493 skip_isolation, migrat 542 skip_isolation, migratetype);
494 if (ret) 543 if (ret)
495 return ret; 544 return ret;
496 545
497 if (isolate_start == isolate_end - pag 546 if (isolate_start == isolate_end - pageblock_nr_pages)
498 skip_isolation = true; 547 skip_isolation = true;
499 548
500 /* isolate [isolate_end - pageblock_nr 549 /* isolate [isolate_end - pageblock_nr_pages, isolate_end) pageblock */
501 ret = isolate_single_pageblock(isolate 550 ret = isolate_single_pageblock(isolate_end, flags, gfp_flags, true,
502 skip_isolation, migrat 551 skip_isolation, migratetype);
503 if (ret) { 552 if (ret) {
504 unset_migratetype_isolate(pfn_ 553 unset_migratetype_isolate(pfn_to_page(isolate_start), migratetype);
505 return ret; 554 return ret;
506 } 555 }
507 556
508 /* skip isolated pageblocks at the beg 557 /* skip isolated pageblocks at the beginning and end */
509 for (pfn = isolate_start + pageblock_n 558 for (pfn = isolate_start + pageblock_nr_pages;
510 pfn < isolate_end - pageblock_nr_ 559 pfn < isolate_end - pageblock_nr_pages;
511 pfn += pageblock_nr_pages) { 560 pfn += pageblock_nr_pages) {
512 page = __first_valid_page(pfn, 561 page = __first_valid_page(pfn, pageblock_nr_pages);
513 if (page && set_migratetype_is 562 if (page && set_migratetype_isolate(page, migratetype, flags,
514 start_ 563 start_pfn, end_pfn)) {
515 undo_isolate_page_rang 564 undo_isolate_page_range(isolate_start, pfn, migratetype);
516 unset_migratetype_isol 565 unset_migratetype_isolate(
517 pfn_to_page(is 566 pfn_to_page(isolate_end - pageblock_nr_pages),
518 migratetype); 567 migratetype);
519 return -EBUSY; 568 return -EBUSY;
520 } 569 }
521 } 570 }
522 return 0; 571 return 0;
523 } 572 }
524 573
525 /** !! 574 /*
526 * undo_isolate_page_range - undo effects of s !! 575 * Make isolated pages available again.
527 * @start_pfn: The first PFN of the i <<
528 * @end_pfn: The last PFN of the is <<
529 * @migratetype: New migrate type to se <<
530 * <<
531 * This finds every MIGRATE_ISOLATE page block <<
532 * and switches it to @migratetype. <<
533 */ 576 */
534 void undo_isolate_page_range(unsigned long sta 577 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
535 int migratetype) 578 int migratetype)
536 { 579 {
537 unsigned long pfn; 580 unsigned long pfn;
538 struct page *page; 581 struct page *page;
539 unsigned long isolate_start = pagebloc !! 582 unsigned long isolate_start = ALIGN_DOWN(start_pfn, pageblock_nr_pages);
540 unsigned long isolate_end = pageblock_ !! 583 unsigned long isolate_end = ALIGN(end_pfn, pageblock_nr_pages);
>> 584
541 585
542 for (pfn = isolate_start; 586 for (pfn = isolate_start;
543 pfn < isolate_end; 587 pfn < isolate_end;
544 pfn += pageblock_nr_pages) { 588 pfn += pageblock_nr_pages) {
545 page = __first_valid_page(pfn, 589 page = __first_valid_page(pfn, pageblock_nr_pages);
546 if (!page || !is_migrate_isola 590 if (!page || !is_migrate_isolate_page(page))
547 continue; 591 continue;
548 unset_migratetype_isolate(page 592 unset_migratetype_isolate(page, migratetype);
549 } 593 }
550 } 594 }
551 /* 595 /*
552 * Test all pages in the range is free(means i 596 * Test all pages in the range is free(means isolated) or not.
553 * all pages in [start_pfn...end_pfn) must be 597 * all pages in [start_pfn...end_pfn) must be in the same zone.
554 * zone->lock must be held before call this. 598 * zone->lock must be held before call this.
555 * 599 *
556 * Returns the last tested pfn. 600 * Returns the last tested pfn.
557 */ 601 */
558 static unsigned long 602 static unsigned long
559 __test_page_isolated_in_pageblock(unsigned lon 603 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
560 int flags) 604 int flags)
561 { 605 {
562 struct page *page; 606 struct page *page;
563 607
564 while (pfn < end_pfn) { 608 while (pfn < end_pfn) {
565 page = pfn_to_page(pfn); 609 page = pfn_to_page(pfn);
566 if (PageBuddy(page)) 610 if (PageBuddy(page))
567 /* 611 /*
568 * If the page is on a 612 * If the page is on a free list, it has to be on
569 * the correct MIGRATE 613 * the correct MIGRATE_ISOLATE freelist. There is no
570 * simple way to verif 614 * simple way to verify that as VM_BUG_ON(), though.
571 */ 615 */
572 pfn += 1 << buddy_orde 616 pfn += 1 << buddy_order(page);
573 else if ((flags & MEMORY_OFFLI 617 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
574 /* A HWPoisoned page c 618 /* A HWPoisoned page cannot be also PageBuddy */
575 pfn++; 619 pfn++;
576 else if ((flags & MEMORY_OFFLI 620 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
577 !page_count(page)) 621 !page_count(page))
578 /* 622 /*
579 * The responsible dri 623 * The responsible driver agreed to skip PageOffline()
580 * pages when offlinin 624 * pages when offlining memory by dropping its
581 * reference in MEM_GO 625 * reference in MEM_GOING_OFFLINE.
582 */ 626 */
583 pfn++; 627 pfn++;
584 else 628 else
585 break; 629 break;
586 } 630 }
587 631
588 return pfn; 632 return pfn;
589 } 633 }
590 634
591 /** !! 635 /* Caller should ensure that requested range is in a single zone */
592 * test_pages_isolated - check if pageblocks i <<
593 * @start_pfn: The first PFN of the i <<
594 * @end_pfn: The first PFN *after* <<
595 * @isol_flags: Testing mode flags <<
596 * <<
597 * This tests if all in the specified range ar <<
598 * <<
599 * If %MEMORY_OFFLINE is specified in @flags, <<
600 * poisoned and offlined pages free as well. <<
601 * <<
602 * Caller must ensure the requested range does <<
603 * <<
604 * Returns 0 if true, -EBUSY if one or more pa <<
605 */ <<
606 int test_pages_isolated(unsigned long start_pf 636 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
607 int isol_flags) 637 int isol_flags)
608 { 638 {
609 unsigned long pfn, flags; 639 unsigned long pfn, flags;
610 struct page *page; 640 struct page *page;
611 struct zone *zone; 641 struct zone *zone;
612 int ret; 642 int ret;
613 643
614 /* 644 /*
615 * Note: pageblock_nr_pages != MAX_PAG !! 645 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
616 * pages are not aligned to pageblock_ !! 646 * are not aligned to pageblock_nr_pages.
617 * Then we just check migratetype firs 647 * Then we just check migratetype first.
618 */ 648 */
619 for (pfn = start_pfn; pfn < end_pfn; p 649 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
620 page = __first_valid_page(pfn, 650 page = __first_valid_page(pfn, pageblock_nr_pages);
621 if (page && !is_migrate_isolat 651 if (page && !is_migrate_isolate_page(page))
622 break; 652 break;
623 } 653 }
624 page = __first_valid_page(start_pfn, e 654 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
625 if ((pfn < end_pfn) || !page) { 655 if ((pfn < end_pfn) || !page) {
626 ret = -EBUSY; 656 ret = -EBUSY;
627 goto out; 657 goto out;
628 } 658 }
629 659
630 /* Check all pages are free or marked 660 /* Check all pages are free or marked as ISOLATED */
631 zone = page_zone(page); 661 zone = page_zone(page);
632 spin_lock_irqsave(&zone->lock, flags); 662 spin_lock_irqsave(&zone->lock, flags);
633 pfn = __test_page_isolated_in_pagebloc 663 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
634 spin_unlock_irqrestore(&zone->lock, fl 664 spin_unlock_irqrestore(&zone->lock, flags);
635 665
636 ret = pfn < end_pfn ? -EBUSY : 0; 666 ret = pfn < end_pfn ? -EBUSY : 0;
637 667
638 out: 668 out:
639 trace_test_pages_isolated(start_pfn, e 669 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
640 670
641 return ret; 671 return ret;
642 } 672 }
643 673
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.