]> git.ipfire.org Git - thirdparty/linux.git/blame - mm/page_isolation.c
ARM: ixp4xx: don't select SERIAL_OF_PLATFORM
[thirdparty/linux.git] / mm / page_isolation.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
a5d76b54
KH
2/*
3 * linux/mm/page_isolation.c
4 */
5
a5d76b54
KH
6#include <linux/mm.h>
7#include <linux/page-isolation.h>
8#include <linux/pageblock-flags.h>
ee6f509c 9#include <linux/memory.h>
c8721bbb 10#include <linux/hugetlb.h>
83358ece 11#include <linux/page_owner.h>
8b913238 12#include <linux/migrate.h>
a5d76b54
KH
13#include "internal.h"
14
0f0848e5
JK
15#define CREATE_TRACE_POINTS
16#include <trace/events/page_isolation.h>
17
d381c547 18static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
ee6f509c
MK
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
2c7452a0
MK
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
ee6f509c
MK
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 */
f5777bc2
QC
62 if (!has_unmovable_pages(zone, page, arg.pages_found, migratetype,
63 isol_flags))
ee6f509c
MK
64 ret = 0;
65
66 /*
ac34dcd2 67 * immobile means "not-on-lru" pages. If immobile is larger than
ee6f509c
MK
68 * removable-by-driver pages reported by notifier, we'll fail.
69 */
70
71out:
72 if (!ret) {
2139cbe6 73 unsigned long nr_pages;
4da2ce25 74 int mt = get_pageblock_migratetype(page);
2139cbe6 75
a458431e 76 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
ad53f92e 77 zone->nr_isolate_pageblock++;
02aa0cdd
VB
78 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
79 NULL);
2139cbe6 80
4da2ce25 81 __mod_zone_freepage_state(zone, -nr_pages, mt);
ee6f509c
MK
82 }
83
84 spin_unlock_irqrestore(&zone->lock, flags);
85 if (!ret)
ec25af84 86 drain_all_pages(zone);
ee6f509c
MK
87 return ret;
88}
89
c5b4e1b0 90static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
ee6f509c
MK
91{
92 struct zone *zone;
2139cbe6 93 unsigned long flags, nr_pages;
e3a2713c 94 bool isolated_page = false;
3c605096 95 unsigned int order;
76741e77 96 unsigned long pfn, buddy_pfn;
3c605096 97 struct page *buddy;
2139cbe6 98
ee6f509c
MK
99 zone = page_zone(page);
100 spin_lock_irqsave(&zone->lock, flags);
bbf9ce97 101 if (!is_migrate_isolate_page(page))
ee6f509c 102 goto out;
3c605096
JK
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) {
76741e77
VB
115 pfn = page_to_pfn(page);
116 buddy_pfn = __find_buddy_pfn(pfn, order);
117 buddy = page + (buddy_pfn - pfn);
3c605096 118
13ad59df 119 if (pfn_valid_within(buddy_pfn) &&
1ae7013d 120 !is_migrate_isolate_page(buddy)) {
3c605096 121 __isolate_free_page(page, order);
e3a2713c 122 isolated_page = true;
3c605096
JK
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) {
02aa0cdd 133 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
3c605096
JK
134 __mod_zone_freepage_state(zone, nr_pages, migratetype);
135 }
a458431e 136 set_pageblock_migratetype(page, migratetype);
ad53f92e 137 zone->nr_isolate_pageblock--;
ee6f509c
MK
138out:
139 spin_unlock_irqrestore(&zone->lock, flags);
83358ece 140 if (isolated_page) {
46f24fd8 141 post_alloc_hook(page, order, __GFP_MOVABLE);
e3a2713c 142 __free_pages(page, order);
83358ece 143 }
ee6f509c
MK
144}
145
a5d76b54
KH
146static inline struct page *
147__first_valid_page(unsigned long pfn, unsigned long nr_pages)
148{
149 int i;
2ce13640
MH
150
151 for (i = 0; i < nr_pages; i++) {
152 struct page *page;
153
2ce13640
MH
154 page = pfn_to_online_page(pfn + i);
155 if (!page)
156 continue;
157 return page;
158 }
159 return NULL;
a5d76b54
KH
160}
161
9b7ea46a
QC
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
a5d76b54
KH
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
9b7ea46a
QC
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.
2c7452a0
MK
181 *
182 * There is no high level synchronization mechanism that prevents two threads
9b7ea46a 183 * from trying to isolate overlapping ranges. If this happens, one thread
2c7452a0
MK
184 * will notice pageblocks in the overlapping range already set to isolate.
185 * This happens in set_migratetype_isolate, and set_migratetype_isolate
9b7ea46a
QC
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
2c7452a0 188 * prevents two threads from simultaneously working on overlapping ranges.
9b7ea46a
QC
189 *
190 * Return: the number of isolated pageblocks on success and -EBUSY if any part
191 * of range cannot be isolated.
a5d76b54 192 */
0815f3d8 193int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
d381c547 194 unsigned migratetype, int flags)
a5d76b54
KH
195{
196 unsigned long pfn;
197 unsigned long undo_pfn;
198 struct page *page;
9b7ea46a 199 int nr_isolate_pageblock = 0;
a5d76b54 200
fec174d6
NH
201 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
202 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
a5d76b54
KH
203
204 for (pfn = start_pfn;
205 pfn < end_pfn;
206 pfn += pageblock_nr_pages) {
207 page = __first_valid_page(pfn, pageblock_nr_pages);
9b7ea46a
QC
208 if (page) {
209 if (set_migratetype_isolate(page, migratetype, flags)) {
210 undo_pfn = pfn;
211 goto undo;
212 }
213 nr_isolate_pageblock++;
a5d76b54
KH
214 }
215 }
9b7ea46a 216 return nr_isolate_pageblock;
a5d76b54
KH
217undo:
218 for (pfn = start_pfn;
dbc0e4ce 219 pfn < undo_pfn;
2ce13640
MH
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 }
a5d76b54
KH
226
227 return -EBUSY;
228}
229
230/*
231 * Make isolated pages available again.
232 */
0815f3d8
MN
233int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
234 unsigned migratetype)
a5d76b54
KH
235{
236 unsigned long pfn;
237 struct page *page;
6f8d2b8a
WX
238
239 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
240 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
241
a5d76b54
KH
242 for (pfn = start_pfn;
243 pfn < end_pfn;
244 pfn += pageblock_nr_pages) {
245 page = __first_valid_page(pfn, pageblock_nr_pages);
bbf9ce97 246 if (!page || !is_migrate_isolate_page(page))
a5d76b54 247 continue;
0815f3d8 248 unset_migratetype_isolate(page, migratetype);
a5d76b54
KH
249 }
250 return 0;
251}
252/*
253 * Test all pages in the range is free(means isolated) or not.
254 * all pages in [start_pfn...end_pfn) must be in the same zone.
255 * zone->lock must be held before call this.
256 *
ec3b6882 257 * Returns the last tested pfn.
a5d76b54 258 */
fea85cff 259static unsigned long
b023f468
WC
260__test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
261 bool skip_hwpoisoned_pages)
a5d76b54
KH
262{
263 struct page *page;
264
265 while (pfn < end_pfn) {
266 if (!pfn_valid_within(pfn)) {
267 pfn++;
268 continue;
269 }
270 page = pfn_to_page(pfn);
aa016d14 271 if (PageBuddy(page))
435b405c 272 /*
aa016d14
VB
273 * If the page is on a free list, it has to be on
274 * the correct MIGRATE_ISOLATE freelist. There is no
275 * simple way to verify that as VM_BUG_ON(), though.
435b405c 276 */
a5d76b54 277 pfn += 1 << page_order(page);
aa016d14
VB
278 else if (skip_hwpoisoned_pages && PageHWPoison(page))
279 /* A HWPoisoned page cannot be also PageBuddy */
b023f468 280 pfn++;
a5d76b54
KH
281 else
282 break;
283 }
fea85cff
JK
284
285 return pfn;
a5d76b54
KH
286}
287
b9eb6319 288/* Caller should ensure that requested range is in a single zone */
b023f468
WC
289int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
290 bool skip_hwpoisoned_pages)
a5d76b54 291{
6c1b7f68 292 unsigned long pfn, flags;
a5d76b54 293 struct page *page;
6c1b7f68 294 struct zone *zone;
a5d76b54 295
a5d76b54 296 /*
85dbe706
TC
297 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
298 * are not aligned to pageblock_nr_pages.
299 * Then we just check migratetype first.
a5d76b54
KH
300 */
301 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
302 page = __first_valid_page(pfn, pageblock_nr_pages);
bbf9ce97 303 if (page && !is_migrate_isolate_page(page))
a5d76b54
KH
304 break;
305 }
a70dcb96
GS
306 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
307 if ((pfn < end_pfn) || !page)
a5d76b54 308 return -EBUSY;
85dbe706 309 /* Check all pages are free or marked as ISOLATED */
a70dcb96 310 zone = page_zone(page);
6c1b7f68 311 spin_lock_irqsave(&zone->lock, flags);
fea85cff 312 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
b023f468 313 skip_hwpoisoned_pages);
6c1b7f68 314 spin_unlock_irqrestore(&zone->lock, flags);
fea85cff 315
0f0848e5
JK
316 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
317
fea85cff 318 return pfn < end_pfn ? -EBUSY : 0;
a5d76b54 319}
723a0644 320
666feb21 321struct page *alloc_migrate_target(struct page *page, unsigned long private)
723a0644 322{
8b913238 323 return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);
723a0644 324}