1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/mm/page_isolation.c
4  */
5 
6 #include <linux/mm.h>
7 #include <linux/page-isolation.h>
8 #include <linux/pageblock-flags.h>
9 #include <linux/memory.h>
10 #include <linux/hugetlb.h>
11 #include <linux/page_owner.h>
12 #include <linux/migrate.h>
13 #include "internal.h"
14 
15 #define CREATE_TRACE_POINTS
16 #include <trace/events/page_isolation.h>
17 
set_migratetype_isolate(struct page * page,int migratetype,int isol_flags)18 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
19 {
20 	struct zone *zone;
21 	unsigned long flags, pfn;
22 	struct memory_isolate_notify arg;
23 	int notifier_ret;
24 	int ret = -EBUSY;
25 
26 	zone = page_zone(page);
27 
28 	spin_lock_irqsave(&zone->lock, flags);
29 
30 	/*
31 	 * We assume the caller intended to SET migrate type to isolate.
32 	 * If it is already set, then someone else must have raced and
33 	 * set it before us.  Return -EBUSY
34 	 */
35 	if (is_migrate_isolate_page(page))
36 		goto out;
37 
38 	pfn = page_to_pfn(page);
39 	arg.start_pfn = pfn;
40 	arg.nr_pages = pageblock_nr_pages;
41 	arg.pages_found = 0;
42 
43 	/*
44 	 * It may be possible to isolate a pageblock even if the
45 	 * migratetype is not MIGRATE_MOVABLE. The memory isolation
46 	 * notifier chain is used by balloon drivers to return the
47 	 * number of pages in a range that are held by the balloon
48 	 * driver to shrink memory. If all the pages are accounted for
49 	 * by balloons, are free, or on the LRU, isolation can continue.
50 	 * Later, for example, when memory hotplug notifier runs, these
51 	 * pages reported as "can be isolated" should be isolated(freed)
52 	 * by the balloon driver through the memory notifier chain.
53 	 */
54 	notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
55 	notifier_ret = notifier_to_errno(notifier_ret);
56 	if (notifier_ret)
57 		goto out;
58 	/*
59 	 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
60 	 * We just check MOVABLE pages.
61 	 */
62 	if (!has_unmovable_pages(zone, page, arg.pages_found, migratetype,
63 				 isol_flags))
64 		ret = 0;
65 
66 	/*
67 	 * immobile means "not-on-lru" pages. If immobile is larger than
68 	 * removable-by-driver pages reported by notifier, we'll fail.
69 	 */
70 
71 out:
72 	if (!ret) {
73 		unsigned long nr_pages;
74 		int mt = get_pageblock_migratetype(page);
75 
76 		set_pageblock_migratetype(page, MIGRATE_ISOLATE);
77 		zone->nr_isolate_pageblock++;
78 		nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
79 									NULL);
80 
81 		__mod_zone_freepage_state(zone, -nr_pages, mt);
82 	}
83 
84 	spin_unlock_irqrestore(&zone->lock, flags);
85 	if (!ret)
86 		drain_all_pages(zone);
87 	return ret;
88 }
89 
unset_migratetype_isolate(struct page * page,unsigned migratetype)90 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
91 {
92 	struct zone *zone;
93 	unsigned long flags, nr_pages;
94 	bool isolated_page = false;
95 	unsigned int order;
96 	unsigned long pfn, buddy_pfn;
97 	struct page *buddy;
98 
99 	zone = page_zone(page);
100 	spin_lock_irqsave(&zone->lock, flags);
101 	if (!is_migrate_isolate_page(page))
102 		goto out;
103 
104 	/*
105 	 * Because freepage with more than pageblock_order on isolated
106 	 * pageblock is restricted to merge due to freepage counting problem,
107 	 * it is possible that there is free buddy page.
108 	 * move_freepages_block() doesn't care of merge so we need other
109 	 * approach in order to merge them. Isolation and free will make
110 	 * these pages to be merged.
111 	 */
112 	if (PageBuddy(page)) {
113 		order = page_order(page);
114 		if (order >= pageblock_order) {
115 			pfn = page_to_pfn(page);
116 			buddy_pfn = __find_buddy_pfn(pfn, order);
117 			buddy = page + (buddy_pfn - pfn);
118 
119 			if (pfn_valid_within(buddy_pfn) &&
120 			    !is_migrate_isolate_page(buddy)) {
121 				__isolate_free_page(page, order);
122 				isolated_page = true;
123 			}
124 		}
125 	}
126 
127 	/*
128 	 * If we isolate freepage with more than pageblock_order, there
129 	 * should be no freepage in the range, so we could avoid costly
130 	 * pageblock scanning for freepage moving.
131 	 */
132 	if (!isolated_page) {
133 		nr_pages = move_freepages_block(zone, page, migratetype, NULL);
134 		__mod_zone_freepage_state(zone, nr_pages, migratetype);
135 	}
136 	set_pageblock_migratetype(page, migratetype);
137 	zone->nr_isolate_pageblock--;
138 out:
139 	spin_unlock_irqrestore(&zone->lock, flags);
140 	if (isolated_page) {
141 		post_alloc_hook(page, order, __GFP_MOVABLE);
142 		__free_pages(page, order);
143 	}
144 }
145 
146 static inline struct page *
__first_valid_page(unsigned long pfn,unsigned long nr_pages)147 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
148 {
149 	int i;
150 
151 	for (i = 0; i < nr_pages; i++) {
152 		struct page *page;
153 
154 		page = pfn_to_online_page(pfn + i);
155 		if (!page)
156 			continue;
157 		return page;
158 	}
159 	return NULL;
160 }
161 
162 /**
163  * start_isolate_page_range() - make page-allocation-type of range of pages to
164  * be MIGRATE_ISOLATE.
165  * @start_pfn:		The lower PFN of the range to be isolated.
166  * @end_pfn:		The upper PFN of the range to be isolated.
167  *			start_pfn/end_pfn must be aligned to pageblock_order.
168  * @migratetype:	Migrate type to set in error recovery.
169  * @flags:		The following flags are allowed (they can be combined in
170  *			a bit mask)
171  *			SKIP_HWPOISON - ignore hwpoison pages
172  *			REPORT_FAILURE - report details about the failure to
173  *			isolate the range
174  *
175  * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
176  * the range will never be allocated. Any free pages and pages freed in the
177  * future will not be allocated again. If specified range includes migrate types
178  * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
179  * pages in the range finally, the caller have to free all pages in the range.
180  * test_page_isolated() can be used for test it.
181  *
182  * There is no high level synchronization mechanism that prevents two threads
183  * from trying to isolate overlapping ranges. If this happens, one thread
184  * will notice pageblocks in the overlapping range already set to isolate.
185  * This happens in set_migratetype_isolate, and set_migratetype_isolate
186  * returns an error. We then clean up by restoring the migration type on
187  * pageblocks we may have modified and return -EBUSY to caller. This
188  * prevents two threads from simultaneously working on overlapping ranges.
189  *
190  * Return: the number of isolated pageblocks on success and -EBUSY if any part
191  * of range cannot be isolated.
192  */
start_isolate_page_range(unsigned long start_pfn,unsigned long end_pfn,unsigned migratetype,int flags)193 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
194 			     unsigned migratetype, int flags)
195 {
196 	unsigned long pfn;
197 	unsigned long undo_pfn;
198 	struct page *page;
199 	int nr_isolate_pageblock = 0;
200 
201 	BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
202 	BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
203 
204 	for (pfn = start_pfn;
205 	     pfn < end_pfn;
206 	     pfn += pageblock_nr_pages) {
207 		page = __first_valid_page(pfn, pageblock_nr_pages);
208 		if (page) {
209 			if (set_migratetype_isolate(page, migratetype, flags)) {
210 				undo_pfn = pfn;
211 				goto undo;
212 			}
213 			nr_isolate_pageblock++;
214 		}
215 	}
216 	return nr_isolate_pageblock;
217 undo:
218 	for (pfn = start_pfn;
219 	     pfn < undo_pfn;
220 	     pfn += pageblock_nr_pages) {
221 		struct page *page = pfn_to_online_page(pfn);
222 		if (!page)
223 			continue;
224 		unset_migratetype_isolate(page, migratetype);
225 	}
226 
227 	return -EBUSY;
228 }
229 
230 /*
231  * Make isolated pages available again.
232  */
undo_isolate_page_range(unsigned long start_pfn,unsigned long end_pfn,unsigned migratetype)233 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
234 			    unsigned migratetype)
235 {
236 	unsigned long pfn;
237 	struct page *page;
238 
239 	BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
240 	BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
241 
242 	for (pfn = start_pfn;
243 	     pfn < end_pfn;
244 	     pfn += pageblock_nr_pages) {
245 		page = __first_valid_page(pfn, pageblock_nr_pages);
246 		if (!page || !is_migrate_isolate_page(page))
247 			continue;
248 		unset_migratetype_isolate(page, migratetype);
249 	}
250 }
251 /*
252  * Test all pages in the range is free(means isolated) or not.
253  * all pages in [start_pfn...end_pfn) must be in the same zone.
254  * zone->lock must be held before call this.
255  *
256  * Returns the last tested pfn.
257  */
258 static unsigned long
__test_page_isolated_in_pageblock(unsigned long pfn,unsigned long end_pfn,bool skip_hwpoisoned_pages)259 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
260 				  bool skip_hwpoisoned_pages)
261 {
262 	struct page *page;
263 
264 	while (pfn < end_pfn) {
265 		if (!pfn_valid_within(pfn)) {
266 			pfn++;
267 			continue;
268 		}
269 		page = pfn_to_page(pfn);
270 		if (PageBuddy(page))
271 			/*
272 			 * If the page is on a free list, it has to be on
273 			 * the correct MIGRATE_ISOLATE freelist. There is no
274 			 * simple way to verify that as VM_BUG_ON(), though.
275 			 */
276 			pfn += 1 << page_order(page);
277 		else if (skip_hwpoisoned_pages && PageHWPoison(page))
278 			/* A HWPoisoned page cannot be also PageBuddy */
279 			pfn++;
280 		else
281 			break;
282 	}
283 
284 	return pfn;
285 }
286 
287 /* Caller should ensure that requested range is in a single zone */
test_pages_isolated(unsigned long start_pfn,unsigned long end_pfn,bool skip_hwpoisoned_pages)288 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
289 			bool skip_hwpoisoned_pages)
290 {
291 	unsigned long pfn, flags;
292 	struct page *page;
293 	struct zone *zone;
294 
295 	/*
296 	 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
297 	 * are not aligned to pageblock_nr_pages.
298 	 * Then we just check migratetype first.
299 	 */
300 	for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
301 		page = __first_valid_page(pfn, pageblock_nr_pages);
302 		if (page && !is_migrate_isolate_page(page))
303 			break;
304 	}
305 	page = __first_valid_page(start_pfn, end_pfn - start_pfn);
306 	if ((pfn < end_pfn) || !page)
307 		return -EBUSY;
308 	/* Check all pages are free or marked as ISOLATED */
309 	zone = page_zone(page);
310 	spin_lock_irqsave(&zone->lock, flags);
311 	pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
312 						skip_hwpoisoned_pages);
313 	spin_unlock_irqrestore(&zone->lock, flags);
314 
315 	trace_test_pages_isolated(start_pfn, end_pfn, pfn);
316 
317 	return pfn < end_pfn ? -EBUSY : 0;
318 }
319 
alloc_migrate_target(struct page * page,unsigned long private)320 struct page *alloc_migrate_target(struct page *page, unsigned long private)
321 {
322 	return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);
323 }
324