Merge tag 'locking-urgent-2020-12-27' of git://git.kernel.org/pub/scm/linux/kernel...
[linux/fpc-iii.git] / mm / page_idle.c
blob64e5344a992ca9d3bc52838130b3e26dd781f2eb
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/memblock.h>
4 #include <linux/fs.h>
5 #include <linux/sysfs.h>
6 #include <linux/kobject.h>
7 #include <linux/memory_hotplug.h>
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>
16 #define BITMAP_CHUNK_SIZE sizeof(u64)
17 #define BITMAP_CHUNK_BITS (BITMAP_CHUNK_SIZE * BITS_PER_BYTE)
20 * Idle page tracking only considers user memory pages, for other types of
21 * pages the idle flag is always unset and an attempt to set it is silently
22 * ignored.
24 * We treat a page as a user memory page if it is on an LRU list, because it is
25 * always safe to pass such a page to rmap_walk(), which is essential for idle
26 * page tracking. With such an indicator of user pages we can skip isolated
27 * pages, but since there are not usually many of them, it will hardly affect
28 * the overall result.
30 * This function tries to get a user memory page by pfn as described above.
32 static struct page *page_idle_get_page(unsigned long pfn)
34 struct page *page = pfn_to_online_page(pfn);
36 if (!page || !PageLRU(page) ||
37 !get_page_unless_zero(page))
38 return NULL;
40 if (unlikely(!PageLRU(page))) {
41 put_page(page);
42 page = NULL;
44 return page;
47 static bool page_idle_clear_pte_refs_one(struct page *page,
48 struct vm_area_struct *vma,
49 unsigned long addr, void *arg)
51 struct page_vma_mapped_walk pvmw = {
52 .page = page,
53 .vma = vma,
54 .address = addr,
56 bool referenced = false;
58 while (page_vma_mapped_walk(&pvmw)) {
59 addr = pvmw.address;
60 if (pvmw.pte) {
62 * For PTE-mapped THP, one sub page is referenced,
63 * the whole THP is referenced.
65 if (ptep_clear_young_notify(vma, addr, pvmw.pte))
66 referenced = true;
67 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
68 if (pmdp_clear_young_notify(vma, addr, pvmw.pmd))
69 referenced = true;
70 } else {
71 /* unexpected pmd-mapped page? */
72 WARN_ON_ONCE(1);
76 if (referenced) {
77 clear_page_idle(page);
79 * We cleared the referenced bit in a mapping to this page. To
80 * avoid interference with page reclaim, mark it young so that
81 * page_referenced() will return > 0.
83 set_page_young(page);
85 return true;
88 static void page_idle_clear_pte_refs(struct page *page)
91 * Since rwc.arg is unused, rwc is effectively immutable, so we
92 * can make it static const to save some cycles and stack.
94 static const struct rmap_walk_control rwc = {
95 .rmap_one = page_idle_clear_pte_refs_one,
96 .anon_lock = page_lock_anon_vma_read,
98 bool need_lock;
100 if (!page_mapped(page) ||
101 !page_rmapping(page))
102 return;
104 need_lock = !PageAnon(page) || PageKsm(page);
105 if (need_lock && !trylock_page(page))
106 return;
108 rmap_walk(page, (struct rmap_walk_control *)&rwc);
110 if (need_lock)
111 unlock_page(page);
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)
118 u64 *out = (u64 *)buf;
119 struct page *page;
120 unsigned long pfn, end_pfn;
121 int bit;
123 if (pos % BITMAP_CHUNK_SIZE || count % BITMAP_CHUNK_SIZE)
124 return -EINVAL;
126 pfn = pos * BITS_PER_BYTE;
127 if (pfn >= max_pfn)
128 return 0;
130 end_pfn = pfn + count * BITS_PER_BYTE;
131 if (end_pfn > max_pfn)
132 end_pfn = max_pfn;
134 for (; pfn < end_pfn; pfn++) {
135 bit = pfn % BITMAP_CHUNK_BITS;
136 if (!bit)
137 *out = 0ULL;
138 page = page_idle_get_page(pfn);
139 if (page) {
140 if (page_is_idle(page)) {
142 * The page might have been referenced via a
143 * pte, in which case it is not idle. Clear
144 * refs and recheck.
146 page_idle_clear_pte_refs(page);
147 if (page_is_idle(page))
148 *out |= 1ULL << bit;
150 put_page(page);
152 if (bit == BITMAP_CHUNK_BITS - 1)
153 out++;
154 cond_resched();
156 return (char *)out - buf;
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)
163 const u64 *in = (u64 *)buf;
164 struct page *page;
165 unsigned long pfn, end_pfn;
166 int bit;
168 if (pos % BITMAP_CHUNK_SIZE || count % BITMAP_CHUNK_SIZE)
169 return -EINVAL;
171 pfn = pos * BITS_PER_BYTE;
172 if (pfn >= max_pfn)
173 return -ENXIO;
175 end_pfn = pfn + count * BITS_PER_BYTE;
176 if (end_pfn > max_pfn)
177 end_pfn = max_pfn;
179 for (; pfn < end_pfn; pfn++) {
180 bit = pfn % BITMAP_CHUNK_BITS;
181 if ((*in >> bit) & 1) {
182 page = page_idle_get_page(pfn);
183 if (page) {
184 page_idle_clear_pte_refs(page);
185 set_page_idle(page);
186 put_page(page);
189 if (bit == BITMAP_CHUNK_BITS - 1)
190 in++;
191 cond_resched();
193 return (char *)in - buf;
196 static struct bin_attribute page_idle_bitmap_attr =
197 __BIN_ATTR(bitmap, 0600,
198 page_idle_bitmap_read, page_idle_bitmap_write, 0);
200 static struct bin_attribute *page_idle_bin_attrs[] = {
201 &page_idle_bitmap_attr,
202 NULL,
205 static const struct attribute_group page_idle_attr_group = {
206 .bin_attrs = page_idle_bin_attrs,
207 .name = "page_idle",
210 #ifndef CONFIG_64BIT
211 static bool need_page_idle(void)
213 return true;
215 struct page_ext_operations page_idle_ops = {
216 .need = need_page_idle,
218 #endif
220 static int __init page_idle_init(void)
222 int err;
224 err = sysfs_create_group(mm_kobj, &page_idle_attr_group);
225 if (err) {
226 pr_err("page_idle: register sysfs failed\n");
227 return err;
229 return 0;
231 subsys_initcall(page_idle_init);