xref: /openbmc/linux/fs/proc/page.c (revision e8ec0493)
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/memcontrol.h>
14 #include <linux/mmu_notifier.h>
15 #include <linux/page_idle.h>
16 #include <linux/kernel-page-flags.h>
17 #include <linux/uaccess.h>
18 #include "internal.h"
19 
20 #define KPMSIZE sizeof(u64)
21 #define KPMMASK (KPMSIZE - 1)
22 #define KPMBITS (KPMSIZE * BITS_PER_BYTE)
23 
24 static inline unsigned long get_max_dump_pfn(void)
25 {
26 #ifdef CONFIG_SPARSEMEM
27 	/*
28 	 * The memmap of early sections is completely populated and marked
29 	 * online even if max_pfn does not fall on a section boundary -
30 	 * pfn_to_online_page() will succeed on all pages. Allow inspecting
31 	 * these memmaps.
32 	 */
33 	return round_up(max_pfn, PAGES_PER_SECTION);
34 #else
35 	return max_pfn;
36 #endif
37 }
38 
39 /* /proc/kpagecount - an array exposing page counts
40  *
41  * Each entry is a u64 representing the corresponding
42  * physical page count.
43  */
44 static ssize_t kpagecount_read(struct file *file, char __user *buf,
45 			     size_t count, loff_t *ppos)
46 {
47 	const unsigned long max_dump_pfn = get_max_dump_pfn();
48 	u64 __user *out = (u64 __user *)buf;
49 	struct page *ppage;
50 	unsigned long src = *ppos;
51 	unsigned long pfn;
52 	ssize_t ret = 0;
53 	u64 pcount;
54 
55 	pfn = src / KPMSIZE;
56 	if (src & KPMMASK || count & KPMMASK)
57 		return -EINVAL;
58 	if (src >= max_dump_pfn * KPMSIZE)
59 		return 0;
60 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
61 
62 	while (count > 0) {
63 		/*
64 		 * TODO: ZONE_DEVICE support requires to identify
65 		 * memmaps that were actually initialized.
66 		 */
67 		ppage = pfn_to_online_page(pfn);
68 
69 		if (!ppage || PageSlab(ppage) || page_has_type(ppage))
70 			pcount = 0;
71 		else
72 			pcount = page_mapcount(ppage);
73 
74 		if (put_user(pcount, out)) {
75 			ret = -EFAULT;
76 			break;
77 		}
78 
79 		pfn++;
80 		out++;
81 		count -= KPMSIZE;
82 
83 		cond_resched();
84 	}
85 
86 	*ppos += (char __user *)out - buf;
87 	if (!ret)
88 		ret = (char __user *)out - buf;
89 	return ret;
90 }
91 
92 static const struct proc_ops kpagecount_proc_ops = {
93 	.proc_lseek	= mem_lseek,
94 	.proc_read	= kpagecount_read,
95 };
96 
97 /* /proc/kpageflags - an array exposing page flags
98  *
99  * Each entry is a u64 representing the corresponding
100  * physical page flags.
101  */
102 
103 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
104 {
105 	return ((kflags >> kbit) & 1) << ubit;
106 }
107 
108 u64 stable_page_flags(struct page *page)
109 {
110 	u64 k;
111 	u64 u;
112 
113 	/*
114 	 * pseudo flag: KPF_NOPAGE
115 	 * it differentiates a memory hole from a page with no flags
116 	 */
117 	if (!page)
118 		return 1 << KPF_NOPAGE;
119 
120 	k = page->flags;
121 	u = 0;
122 
123 	/*
124 	 * pseudo flags for the well known (anonymous) memory mapped pages
125 	 *
126 	 * Note that page->_mapcount is overloaded in SLOB/SLUB/SLQB, so the
127 	 * simple test in page_mapped() is not enough.
128 	 */
129 	if (!PageSlab(page) && page_mapped(page))
130 		u |= 1 << KPF_MMAP;
131 	if (PageAnon(page))
132 		u |= 1 << KPF_ANON;
133 	if (PageKsm(page))
134 		u |= 1 << KPF_KSM;
135 
136 	/*
137 	 * compound pages: export both head/tail info
138 	 * they together define a compound page's start/end pos and order
139 	 */
140 	if (PageHead(page))
141 		u |= 1 << KPF_COMPOUND_HEAD;
142 	if (PageTail(page))
143 		u |= 1 << KPF_COMPOUND_TAIL;
144 	if (PageHuge(page))
145 		u |= 1 << KPF_HUGE;
146 	/*
147 	 * PageTransCompound can be true for non-huge compound pages (slab
148 	 * pages or pages allocated by drivers with __GFP_COMP) because it
149 	 * just checks PG_head/PG_tail, so we need to check PageLRU/PageAnon
150 	 * to make sure a given page is a thp, not a non-huge compound page.
151 	 */
152 	else if (PageTransCompound(page)) {
153 		struct page *head = compound_head(page);
154 
155 		if (PageLRU(head) || PageAnon(head))
156 			u |= 1 << KPF_THP;
157 		else if (is_huge_zero_page(head)) {
158 			u |= 1 << KPF_ZERO_PAGE;
159 			u |= 1 << KPF_THP;
160 		}
161 	} else if (is_zero_pfn(page_to_pfn(page)))
162 		u |= 1 << KPF_ZERO_PAGE;
163 
164 
165 	/*
166 	 * Caveats on high order pages: page->_refcount will only be set
167 	 * -1 on the head page; SLUB/SLQB do the same for PG_slab;
168 	 * SLOB won't set PG_slab at all on compound pages.
169 	 */
170 	if (PageBuddy(page))
171 		u |= 1 << KPF_BUDDY;
172 	else if (page_count(page) == 0 && is_free_buddy_page(page))
173 		u |= 1 << KPF_BUDDY;
174 
175 	if (PageOffline(page))
176 		u |= 1 << KPF_OFFLINE;
177 	if (PageTable(page))
178 		u |= 1 << KPF_PGTABLE;
179 
180 	if (page_is_idle(page))
181 		u |= 1 << KPF_IDLE;
182 
183 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
184 
185 	u |= kpf_copy_bit(k, KPF_SLAB,		PG_slab);
186 	if (PageTail(page) && PageSlab(compound_head(page)))
187 		u |= 1 << KPF_SLAB;
188 
189 	u |= kpf_copy_bit(k, KPF_ERROR,		PG_error);
190 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
191 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
192 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
193 
194 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
195 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
196 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
197 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
198 
199 	if (PageSwapCache(page))
200 		u |= 1 << KPF_SWAPCACHE;
201 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
202 
203 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
204 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
205 
206 #ifdef CONFIG_MEMORY_FAILURE
207 	u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
208 #endif
209 
210 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
211 	u |= kpf_copy_bit(k, KPF_UNCACHED,	PG_uncached);
212 #endif
213 
214 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
215 	u |= kpf_copy_bit(k, KPF_MAPPEDTODISK,	PG_mappedtodisk);
216 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
217 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
218 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
219 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
220 
221 	return u;
222 };
223 
224 static ssize_t kpageflags_read(struct file *file, char __user *buf,
225 			     size_t count, loff_t *ppos)
226 {
227 	const unsigned long max_dump_pfn = get_max_dump_pfn();
228 	u64 __user *out = (u64 __user *)buf;
229 	struct page *ppage;
230 	unsigned long src = *ppos;
231 	unsigned long pfn;
232 	ssize_t ret = 0;
233 
234 	pfn = src / KPMSIZE;
235 	if (src & KPMMASK || count & KPMMASK)
236 		return -EINVAL;
237 	if (src >= max_dump_pfn * KPMSIZE)
238 		return 0;
239 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
240 
241 	while (count > 0) {
242 		/*
243 		 * TODO: ZONE_DEVICE support requires to identify
244 		 * memmaps that were actually initialized.
245 		 */
246 		ppage = pfn_to_online_page(pfn);
247 
248 		if (put_user(stable_page_flags(ppage), out)) {
249 			ret = -EFAULT;
250 			break;
251 		}
252 
253 		pfn++;
254 		out++;
255 		count -= KPMSIZE;
256 
257 		cond_resched();
258 	}
259 
260 	*ppos += (char __user *)out - buf;
261 	if (!ret)
262 		ret = (char __user *)out - buf;
263 	return ret;
264 }
265 
266 static const struct proc_ops kpageflags_proc_ops = {
267 	.proc_lseek	= mem_lseek,
268 	.proc_read	= kpageflags_read,
269 };
270 
271 #ifdef CONFIG_MEMCG
272 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
273 				size_t count, loff_t *ppos)
274 {
275 	const unsigned long max_dump_pfn = get_max_dump_pfn();
276 	u64 __user *out = (u64 __user *)buf;
277 	struct page *ppage;
278 	unsigned long src = *ppos;
279 	unsigned long pfn;
280 	ssize_t ret = 0;
281 	u64 ino;
282 
283 	pfn = src / KPMSIZE;
284 	if (src & KPMMASK || count & KPMMASK)
285 		return -EINVAL;
286 	if (src >= max_dump_pfn * KPMSIZE)
287 		return 0;
288 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
289 
290 	while (count > 0) {
291 		/*
292 		 * TODO: ZONE_DEVICE support requires to identify
293 		 * memmaps that were actually initialized.
294 		 */
295 		ppage = pfn_to_online_page(pfn);
296 
297 		if (ppage)
298 			ino = page_cgroup_ino(ppage);
299 		else
300 			ino = 0;
301 
302 		if (put_user(ino, out)) {
303 			ret = -EFAULT;
304 			break;
305 		}
306 
307 		pfn++;
308 		out++;
309 		count -= KPMSIZE;
310 
311 		cond_resched();
312 	}
313 
314 	*ppos += (char __user *)out - buf;
315 	if (!ret)
316 		ret = (char __user *)out - buf;
317 	return ret;
318 }
319 
320 static const struct proc_ops kpagecgroup_proc_ops = {
321 	.proc_lseek	= mem_lseek,
322 	.proc_read	= kpagecgroup_read,
323 };
324 #endif /* CONFIG_MEMCG */
325 
326 static int __init proc_page_init(void)
327 {
328 	proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
329 	proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
330 #ifdef CONFIG_MEMCG
331 	proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
332 #endif
333 	return 0;
334 }
335 fs_initcall(proc_page_init);
336