Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
33c3fc71 | 2 | #include <linux/init.h> |
57c8a661 | 3 | #include <linux/memblock.h> |
33c3fc71 VD |
4 | #include <linux/fs.h> |
5 | #include <linux/sysfs.h> | |
6 | #include <linux/kobject.h> | |
92fb1db2 | 7 | #include <linux/memory_hotplug.h> |
33c3fc71 VD |
8 | #include <linux/mm.h> |
9 | #include <linux/mmzone.h> | |
10 | #include <linux/pagemap.h> | |
11 | #include <linux/rmap.h> | |
12 | #include <linux/mmu_notifier.h> | |
13 | #include <linux/page_ext.h> | |
14 | #include <linux/page_idle.h> | |
15 | ||
4aed23a2 MWO |
16 | #include "internal.h" |
17 | ||
33c3fc71 VD |
18 | #define BITMAP_CHUNK_SIZE sizeof(u64) |
19 | #define BITMAP_CHUNK_BITS (BITMAP_CHUNK_SIZE * BITS_PER_BYTE) | |
20 | ||
21 | /* | |
22 | * Idle page tracking only considers user memory pages, for other types of | |
23 | * pages the idle flag is always unset and an attempt to set it is silently | |
24 | * ignored. | |
25 | * | |
26 | * We treat a page as a user memory page if it is on an LRU list, because it is | |
27 | * always safe to pass such a page to rmap_walk(), which is essential for idle | |
28 | * page tracking. With such an indicator of user pages we can skip isolated | |
29 | * pages, but since there are not usually many of them, it will hardly affect | |
30 | * the overall result. | |
31 | * | |
32 | * This function tries to get a user memory page by pfn as described above. | |
33 | */ | |
5acc17fd | 34 | static struct folio *page_idle_get_folio(unsigned long pfn) |
33c3fc71 | 35 | { |
92fb1db2 | 36 | struct page *page = pfn_to_online_page(pfn); |
5acc17fd | 37 | struct folio *folio; |
33c3fc71 | 38 | |
5acc17fd | 39 | if (!page || PageTail(page)) |
33c3fc71 VD |
40 | return NULL; |
41 | ||
5acc17fd KW |
42 | folio = page_folio(page); |
43 | if (!folio_test_lru(folio) || !folio_try_get(folio)) | |
44 | return NULL; | |
45 | if (unlikely(page_folio(page) != folio || !folio_test_lru(folio))) { | |
46 | folio_put(folio); | |
47 | folio = NULL; | |
33c3fc71 | 48 | } |
5acc17fd | 49 | return folio; |
33c3fc71 VD |
50 | } |
51 | ||
2f031c6f | 52 | static bool page_idle_clear_pte_refs_one(struct folio *folio, |
33c3fc71 VD |
53 | struct vm_area_struct *vma, |
54 | unsigned long addr, void *arg) | |
55 | { | |
4aed23a2 | 56 | DEFINE_FOLIO_VMA_WALK(pvmw, folio, vma, addr, 0); |
33c3fc71 VD |
57 | bool referenced = false; |
58 | ||
699fa216 KS |
59 | while (page_vma_mapped_walk(&pvmw)) { |
60 | addr = pvmw.address; | |
61 | if (pvmw.pte) { | |
f0849ac0 YS |
62 | /* |
63 | * For PTE-mapped THP, one sub page is referenced, | |
64 | * the whole THP is referenced. | |
65 | */ | |
66 | if (ptep_clear_young_notify(vma, addr, pvmw.pte)) | |
67 | referenced = true; | |
699fa216 | 68 | } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { |
f0849ac0 YS |
69 | if (pmdp_clear_young_notify(vma, addr, pvmw.pmd)) |
70 | referenced = true; | |
699fa216 KS |
71 | } else { |
72 | /* unexpected pmd-mapped page? */ | |
73 | WARN_ON_ONCE(1); | |
74 | } | |
33c3fc71 | 75 | } |
b20ce5e0 | 76 | |
33c3fc71 | 77 | if (referenced) { |
4aed23a2 | 78 | folio_clear_idle(folio); |
33c3fc71 VD |
79 | /* |
80 | * We cleared the referenced bit in a mapping to this page. To | |
81 | * avoid interference with page reclaim, mark it young so that | |
b3ac0413 | 82 | * folio_referenced() will return > 0. |
33c3fc71 | 83 | */ |
4aed23a2 | 84 | folio_set_young(folio); |
33c3fc71 | 85 | } |
e4b82222 | 86 | return true; |
33c3fc71 VD |
87 | } |
88 | ||
5acc17fd | 89 | static void page_idle_clear_pte_refs(struct folio *folio) |
33c3fc71 VD |
90 | { |
91 | /* | |
6d4675e6 MK |
92 | * Since rwc.try_lock is unused, rwc is effectively immutable, so we |
93 | * can make it static to save some cycles and stack. | |
33c3fc71 | 94 | */ |
6d4675e6 | 95 | static struct rmap_walk_control rwc = { |
33c3fc71 | 96 | .rmap_one = page_idle_clear_pte_refs_one, |
2f031c6f | 97 | .anon_lock = folio_lock_anon_vma_read, |
33c3fc71 VD |
98 | }; |
99 | bool need_lock; | |
100 | ||
4aed23a2 | 101 | if (!folio_mapped(folio) || !folio_raw_mapping(folio)) |
33c3fc71 VD |
102 | return; |
103 | ||
4aed23a2 MWO |
104 | need_lock = !folio_test_anon(folio) || folio_test_ksm(folio); |
105 | if (need_lock && !folio_trylock(folio)) | |
33c3fc71 VD |
106 | return; |
107 | ||
84fbbe21 | 108 | rmap_walk(folio, &rwc); |
33c3fc71 VD |
109 | |
110 | if (need_lock) | |
4aed23a2 | 111 | folio_unlock(folio); |
33c3fc71 VD |
112 | } |
113 | ||
114 | static ssize_t page_idle_bitmap_read(struct file *file, struct kobject *kobj, | |
115 | struct bin_attribute *attr, char *buf, | |
116 | loff_t pos, size_t count) | |
117 | { | |
118 | u64 *out = (u64 *)buf; | |
5acc17fd | 119 | struct folio *folio; |
33c3fc71 VD |
120 | unsigned long pfn, end_pfn; |
121 | int bit; | |
122 | ||
123 | if (pos % BITMAP_CHUNK_SIZE || count % BITMAP_CHUNK_SIZE) | |
124 | return -EINVAL; | |
125 | ||
126 | pfn = pos * BITS_PER_BYTE; | |
127 | if (pfn >= max_pfn) | |
128 | return 0; | |
129 | ||
130 | end_pfn = pfn + count * BITS_PER_BYTE; | |
131 | if (end_pfn > max_pfn) | |
7298e3b0 | 132 | end_pfn = max_pfn; |
33c3fc71 VD |
133 | |
134 | for (; pfn < end_pfn; pfn++) { | |
135 | bit = pfn % BITMAP_CHUNK_BITS; | |
136 | if (!bit) | |
137 | *out = 0ULL; | |
5acc17fd KW |
138 | folio = page_idle_get_folio(pfn); |
139 | if (folio) { | |
140 | if (folio_test_idle(folio)) { | |
33c3fc71 VD |
141 | /* |
142 | * The page might have been referenced via a | |
143 | * pte, in which case it is not idle. Clear | |
144 | * refs and recheck. | |
145 | */ | |
5acc17fd KW |
146 | page_idle_clear_pte_refs(folio); |
147 | if (folio_test_idle(folio)) | |
33c3fc71 VD |
148 | *out |= 1ULL << bit; |
149 | } | |
5acc17fd | 150 | folio_put(folio); |
33c3fc71 VD |
151 | } |
152 | if (bit == BITMAP_CHUNK_BITS - 1) | |
153 | out++; | |
154 | cond_resched(); | |
155 | } | |
156 | return (char *)out - buf; | |
157 | } | |
158 | ||
159 | static ssize_t page_idle_bitmap_write(struct file *file, struct kobject *kobj, | |
160 | struct bin_attribute *attr, char *buf, | |
161 | loff_t pos, size_t count) | |
162 | { | |
163 | const u64 *in = (u64 *)buf; | |
5acc17fd | 164 | struct folio *folio; |
33c3fc71 VD |
165 | unsigned long pfn, end_pfn; |
166 | int bit; | |
167 | ||
168 | if (pos % BITMAP_CHUNK_SIZE || count % BITMAP_CHUNK_SIZE) | |
169 | return -EINVAL; | |
170 | ||
171 | pfn = pos * BITS_PER_BYTE; | |
172 | if (pfn >= max_pfn) | |
173 | return -ENXIO; | |
174 | ||
175 | end_pfn = pfn + count * BITS_PER_BYTE; | |
176 | if (end_pfn > max_pfn) | |
7298e3b0 | 177 | end_pfn = max_pfn; |
33c3fc71 VD |
178 | |
179 | for (; pfn < end_pfn; pfn++) { | |
180 | bit = pfn % BITMAP_CHUNK_BITS; | |
181 | if ((*in >> bit) & 1) { | |
5acc17fd KW |
182 | folio = page_idle_get_folio(pfn); |
183 | if (folio) { | |
184 | page_idle_clear_pte_refs(folio); | |
185 | folio_set_idle(folio); | |
186 | folio_put(folio); | |
33c3fc71 VD |
187 | } |
188 | } | |
189 | if (bit == BITMAP_CHUNK_BITS - 1) | |
190 | in++; | |
191 | cond_resched(); | |
192 | } | |
193 | return (char *)in - buf; | |
194 | } | |
195 | ||
196 | static struct bin_attribute page_idle_bitmap_attr = | |
0825a6f9 | 197 | __BIN_ATTR(bitmap, 0600, |
33c3fc71 VD |
198 | page_idle_bitmap_read, page_idle_bitmap_write, 0); |
199 | ||
200 | static struct bin_attribute *page_idle_bin_attrs[] = { | |
201 | &page_idle_bitmap_attr, | |
202 | NULL, | |
203 | }; | |
204 | ||
fd147cbb | 205 | static const struct attribute_group page_idle_attr_group = { |
33c3fc71 VD |
206 | .bin_attrs = page_idle_bin_attrs, |
207 | .name = "page_idle", | |
208 | }; | |
209 | ||
33c3fc71 VD |
210 | static int __init page_idle_init(void) |
211 | { | |
212 | int err; | |
213 | ||
214 | err = sysfs_create_group(mm_kobj, &page_idle_attr_group); | |
215 | if (err) { | |
216 | pr_err("page_idle: register sysfs failed\n"); | |
217 | return err; | |
218 | } | |
219 | return 0; | |
220 | } | |
221 | subsys_initcall(page_idle_init); |