projects/linux·blob master · refs

289 lines · 6.9 KiB

1// SPDX-License-Identifier: GPL-2.0
2#include <linux/memblock.h>
3#include <linux/compiler.h>
4#include <linux/fs.h>
5#include <linux/init.h>
6#include <linux/ksm.h>
7#include <linux/mm.h>
8#include <linux/mmzone.h>
9#include <linux/huge_mm.h>
10#include <linux/proc_fs.h>
11#include <linux/seq_file.h>
12#include <linux/hugetlb.h>
13#include <linux/memremap.h>
14#include <linux/memcontrol.h>
15#include <linux/mmu_notifier.h>
16#include <linux/page_idle.h>
17#include <linux/kernel-page-flags.h>
18#include <linux/uaccess.h>
19#include "internal.h"
20
21#define KPMSIZE sizeof(u64)
22#define KPMMASK (KPMSIZE - 1)
23
24enum kpage_operation {
25 KPAGE_FLAGS,
26 KPAGE_COUNT,
27 KPAGE_CGROUP,
28};
29
30static inline unsigned long get_max_dump_pfn(void)
31{
32#ifdef CONFIG_SPARSEMEM
33 /*
34 * The memmap of early sections is completely populated and marked
35 * online even if max_pfn does not fall on a section boundary -
36 * pfn_to_online_page() will succeed on all pages. Allow inspecting
37 * these memmaps.
38 */
39 return round_up(max_pfn, PAGES_PER_SECTION);
40#else
41 return max_pfn;
42#endif
43}
44
45static u64 get_kpage_count(const struct page *page)
46{
47 struct page_snapshot ps;
48 u64 ret;
49
50 snapshot_page(&ps, page);
51
52 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
53 ret = folio_precise_page_mapcount(&ps.folio_snapshot,
54 &ps.page_snapshot);
55 else
56 ret = folio_average_page_mapcount(&ps.folio_snapshot);
57
58 return ret;
59}
60
61static ssize_t kpage_read(struct file *file, char __user *buf,
62 size_t count, loff_t *ppos,
63 enum kpage_operation op)
64{
65 const unsigned long max_dump_pfn = get_max_dump_pfn();
66 u64 __user *out = (u64 __user *)buf;
67 struct page *page;
68 unsigned long src = *ppos;
69 unsigned long pfn;
70 ssize_t ret = 0;
71 u64 info;
72
73 pfn = src / KPMSIZE;
74 if (src & KPMMASK || count & KPMMASK)
75 return -EINVAL;
76 if (src >= max_dump_pfn * KPMSIZE)
77 return 0;
78 count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
79
80 while (count > 0) {
81 /*
82 * TODO: ZONE_DEVICE support requires to identify
83 * memmaps that were actually initialized.
84 */
85 page = pfn_to_online_page(pfn);
86
87 if (page) {
88 switch (op) {
89 case KPAGE_FLAGS:
90 info = stable_page_flags(page);
91 break;
92 case KPAGE_COUNT:
93 info = get_kpage_count(page);
94 break;
95 case KPAGE_CGROUP:
96 info = page_cgroup_ino(page);
97 break;
98 default:
99 info = 0;
100 break;
101 }
102 } else
103 info = 0;
104
105 if (put_user(info, out)) {
106 ret = -EFAULT;
107 break;
108 }
109
110 pfn++;
111 out++;
112 count -= KPMSIZE;
113
114 cond_resched();
115 }
116
117 *ppos += (char __user *)out - buf;
118 if (!ret)
119 ret = (char __user *)out - buf;
120 return ret;
121}
122
123/* /proc/kpagecount - an array exposing page mapcounts
124 *
125 * Each entry is a u64 representing the corresponding
126 * physical page mapcount.
127 */
128static ssize_t kpagecount_read(struct file *file, char __user *buf,
129 size_t count, loff_t *ppos)
130{
131 return kpage_read(file, buf, count, ppos, KPAGE_COUNT);
132}
133
134static const struct proc_ops kpagecount_proc_ops = {
135 .proc_flags = PROC_ENTRY_PERMANENT,
136 .proc_lseek = mem_lseek,
137 .proc_read = kpagecount_read,
138};
139
140
141static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
142{
143 return ((kflags >> kbit) & 1) << ubit;
144}
145
146u64 stable_page_flags(const struct page *page)
147{
148 const struct folio *folio;
149 struct page_snapshot ps;
150 unsigned long k;
151 u64 u = 0;
152
153 /*
154 * pseudo flag: KPF_NOPAGE
155 * it differentiates a memory hole from a page with no flags
156 */
157 if (!page)
158 return BIT_ULL(KPF_NOPAGE);
159
160 snapshot_page(&ps, page);
161 folio = &ps.folio_snapshot;
162 k = folio->flags.f;
163
164 /*
165 * pseudo flags for the well known (anonymous) memory mapped pages
166 */
167 if (folio_mapped(folio))
168 u |= BIT_ULL(KPF_MMAP);
169 if (folio_test_anon(folio)) {
170 u |= BIT_ULL(KPF_ANON);
171 if (folio_test_ksm(folio))
172 u |= BIT_ULL(KPF_KSM);
173 }
174
175 /*
176 * compound pages: export both head/tail info
177 * they together define a compound page's start/end pos and order
178 */
179 if (ps.idx == 0)
180 u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
181 else
182 u |= BIT_ULL(KPF_COMPOUND_TAIL);
183 if (folio_test_hugetlb(folio))
184 u |= BIT_ULL(KPF_HUGE);
185 else if (folio_test_large(folio) &&
186 folio_test_large_rmappable(folio)) {
187 /* Note: we indicate any THPs here, not just PMD-sized ones */
188 u |= BIT_ULL(KPF_THP);
189 } else if (is_huge_zero_pfn(ps.pfn)) {
190 u |= BIT_ULL(KPF_ZERO_PAGE);
191 u |= BIT_ULL(KPF_THP);
192 } else if (is_zero_pfn(ps.pfn)) {
193 u |= BIT_ULL(KPF_ZERO_PAGE);
194 }
195
196 if (ps.flags & PAGE_SNAPSHOT_PG_BUDDY)
197 u |= BIT_ULL(KPF_BUDDY);
198
199 if (ps.flags & PAGE_SNAPSHOT_PG_IDLE)
200 u |= BIT_ULL(KPF_IDLE);
201
202 if (folio_test_offline(folio))
203 u |= BIT_ULL(KPF_OFFLINE);
204 if (folio_test_pgtable(folio))
205 u |= BIT_ULL(KPF_PGTABLE);
206 if (folio_test_slab(folio))
207 u |= BIT_ULL(KPF_SLAB);
208
209 u |= kpf_copy_bit(k, KPF_LOCKED, PG_locked);
210 u |= kpf_copy_bit(k, KPF_DIRTY, PG_dirty);
211 u |= kpf_copy_bit(k, KPF_UPTODATE, PG_uptodate);
212 u |= kpf_copy_bit(k, KPF_WRITEBACK, PG_writeback);
213
214 u |= kpf_copy_bit(k, KPF_LRU, PG_lru);
215 u |= kpf_copy_bit(k, KPF_REFERENCED, PG_referenced);
216 u |= kpf_copy_bit(k, KPF_ACTIVE, PG_active);
217 u |= kpf_copy_bit(k, KPF_RECLAIM, PG_reclaim);
218
219 if (folio_test_swapcache(folio))
220 u |= BIT_ULL(KPF_SWAPCACHE);
221
222 u |= kpf_copy_bit(k, KPF_SWAPBACKED, PG_swapbacked);
223 u |= kpf_copy_bit(k, KPF_UNEVICTABLE, PG_unevictable);
224 u |= kpf_copy_bit(k, KPF_MLOCKED, PG_mlocked);
225
226#ifdef CONFIG_MEMORY_FAILURE
227 if (u & BIT_ULL(KPF_HUGE))
228 u |= kpf_copy_bit(k, KPF_HWPOISON, PG_hwpoison);
229 else
230 u |= kpf_copy_bit(ps.page_snapshot.flags.f, KPF_HWPOISON, PG_hwpoison);
231#endif
232
233 u |= kpf_copy_bit(k, KPF_RESERVED, PG_reserved);
234 u |= kpf_copy_bit(k, KPF_OWNER_2, PG_owner_2);
235 u |= kpf_copy_bit(k, KPF_PRIVATE, PG_private);
236 u |= kpf_copy_bit(k, KPF_PRIVATE_2, PG_private_2);
237 u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE, PG_owner_priv_1);
238 u |= kpf_copy_bit(k, KPF_ARCH, PG_arch_1);
239#ifdef CONFIG_ARCH_USES_PG_ARCH_2
240 u |= kpf_copy_bit(k, KPF_ARCH_2, PG_arch_2);
241#endif
242#ifdef CONFIG_ARCH_USES_PG_ARCH_3
243 u |= kpf_copy_bit(k, KPF_ARCH_3, PG_arch_3);
244#endif
245
246 return u;
247}
248EXPORT_SYMBOL_GPL(stable_page_flags);
249
250/* /proc/kpageflags - an array exposing page flags
251 *
252 * Each entry is a u64 representing the corresponding
253 * physical page flags.
254 */
255static ssize_t kpageflags_read(struct file *file, char __user *buf,
256 size_t count, loff_t *ppos)
257{
258 return kpage_read(file, buf, count, ppos, KPAGE_FLAGS);
259}
260
261static const struct proc_ops kpageflags_proc_ops = {
262 .proc_flags = PROC_ENTRY_PERMANENT,
263 .proc_lseek = mem_lseek,
264 .proc_read = kpageflags_read,
265};
266
267#ifdef CONFIG_MEMCG
268static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
269 size_t count, loff_t *ppos)
270{
271 return kpage_read(file, buf, count, ppos, KPAGE_CGROUP);
272}
273static const struct proc_ops kpagecgroup_proc_ops = {
274 .proc_flags = PROC_ENTRY_PERMANENT,
275 .proc_lseek = mem_lseek,
276 .proc_read = kpagecgroup_read,
277};
278#endif /* CONFIG_MEMCG */
279
280static int __init proc_page_init(void)
281{
282 proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
283 proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
284#ifdef CONFIG_MEMCG
285 proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
286#endif
287 return 0;
288}
289fs_initcall(proc_page_init);