xref: /openbmc/linux/fs/proc/page.c (revision 020c5260)
1 #include <linux/bootmem.h>
2 #include <linux/compiler.h>
3 #include <linux/fs.h>
4 #include <linux/init.h>
5 #include <linux/ksm.h>
6 #include <linux/mm.h>
7 #include <linux/mmzone.h>
8 #include <linux/huge_mm.h>
9 #include <linux/proc_fs.h>
10 #include <linux/seq_file.h>
11 #include <linux/hugetlb.h>
12 #include <linux/memcontrol.h>
13 #include <linux/mmu_notifier.h>
14 #include <linux/page_idle.h>
15 #include <linux/kernel-page-flags.h>
16 #include <linux/uaccess.h>
17 #include "internal.h"
18 
19 #define KPMSIZE sizeof(u64)
20 #define KPMMASK (KPMSIZE - 1)
21 #define KPMBITS (KPMSIZE * BITS_PER_BYTE)
22 
23 /* /proc/kpagecount - an array exposing page counts
24  *
25  * Each entry is a u64 representing the corresponding
26  * physical page count.
27  */
28 static ssize_t kpagecount_read(struct file *file, char __user *buf,
29 			     size_t count, loff_t *ppos)
30 {
31 	u64 __user *out = (u64 __user *)buf;
32 	struct page *ppage;
33 	unsigned long src = *ppos;
34 	unsigned long pfn;
35 	ssize_t ret = 0;
36 	u64 pcount;
37 
38 	pfn = src / KPMSIZE;
39 	count = min_t(size_t, count, (max_pfn * KPMSIZE) - src);
40 	if (src & KPMMASK || count & KPMMASK)
41 		return -EINVAL;
42 
43 	while (count > 0) {
44 		if (pfn_valid(pfn))
45 			ppage = pfn_to_page(pfn);
46 		else
47 			ppage = NULL;
48 		if (!ppage || PageSlab(ppage))
49 			pcount = 0;
50 		else
51 			pcount = page_mapcount(ppage);
52 
53 		if (put_user(pcount, out)) {
54 			ret = -EFAULT;
55 			break;
56 		}
57 
58 		pfn++;
59 		out++;
60 		count -= KPMSIZE;
61 
62 		cond_resched();
63 	}
64 
65 	*ppos += (char __user *)out - buf;
66 	if (!ret)
67 		ret = (char __user *)out - buf;
68 	return ret;
69 }
70 
71 static const struct file_operations proc_kpagecount_operations = {
72 	.llseek = mem_lseek,
73 	.read = kpagecount_read,
74 };
75 
76 /* /proc/kpageflags - an array exposing page flags
77  *
78  * Each entry is a u64 representing the corresponding
79  * physical page flags.
80  */
81 
82 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
83 {
84 	return ((kflags >> kbit) & 1) << ubit;
85 }
86 
87 u64 stable_page_flags(struct page *page)
88 {
89 	u64 k;
90 	u64 u;
91 
92 	/*
93 	 * pseudo flag: KPF_NOPAGE
94 	 * it differentiates a memory hole from a page with no flags
95 	 */
96 	if (!page)
97 		return 1 << KPF_NOPAGE;
98 
99 	k = page->flags;
100 	u = 0;
101 
102 	/*
103 	 * pseudo flags for the well known (anonymous) memory mapped pages
104 	 *
105 	 * Note that page->_mapcount is overloaded in SLOB/SLUB/SLQB, so the
106 	 * simple test in page_mapped() is not enough.
107 	 */
108 	if (!PageSlab(page) && page_mapped(page))
109 		u |= 1 << KPF_MMAP;
110 	if (PageAnon(page))
111 		u |= 1 << KPF_ANON;
112 	if (PageKsm(page))
113 		u |= 1 << KPF_KSM;
114 
115 	/*
116 	 * compound pages: export both head/tail info
117 	 * they together define a compound page's start/end pos and order
118 	 */
119 	if (PageHead(page))
120 		u |= 1 << KPF_COMPOUND_HEAD;
121 	if (PageTail(page))
122 		u |= 1 << KPF_COMPOUND_TAIL;
123 	if (PageHuge(page))
124 		u |= 1 << KPF_HUGE;
125 	/*
126 	 * PageTransCompound can be true for non-huge compound pages (slab
127 	 * pages or pages allocated by drivers with __GFP_COMP) because it
128 	 * just checks PG_head/PG_tail, so we need to check PageLRU/PageAnon
129 	 * to make sure a given page is a thp, not a non-huge compound page.
130 	 */
131 	else if (PageTransCompound(page)) {
132 		struct page *head = compound_head(page);
133 
134 		if (PageLRU(head) || PageAnon(head))
135 			u |= 1 << KPF_THP;
136 		else if (is_huge_zero_page(head)) {
137 			u |= 1 << KPF_ZERO_PAGE;
138 			u |= 1 << KPF_THP;
139 		}
140 	} else if (is_zero_pfn(page_to_pfn(page)))
141 		u |= 1 << KPF_ZERO_PAGE;
142 
143 
144 	/*
145 	 * Caveats on high order pages: page->_refcount will only be set
146 	 * -1 on the head page; SLUB/SLQB do the same for PG_slab;
147 	 * SLOB won't set PG_slab at all on compound pages.
148 	 */
149 	if (PageBuddy(page))
150 		u |= 1 << KPF_BUDDY;
151 	else if (page_count(page) == 0 && is_free_buddy_page(page))
152 		u |= 1 << KPF_BUDDY;
153 
154 	if (PageBalloon(page))
155 		u |= 1 << KPF_BALLOON;
156 
157 	if (page_is_idle(page))
158 		u |= 1 << KPF_IDLE;
159 
160 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
161 
162 	u |= kpf_copy_bit(k, KPF_SLAB,		PG_slab);
163 	if (PageTail(page) && PageSlab(compound_head(page)))
164 		u |= 1 << KPF_SLAB;
165 
166 	u |= kpf_copy_bit(k, KPF_ERROR,		PG_error);
167 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
168 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
169 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
170 
171 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
172 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
173 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
174 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
175 
176 	if (PageSwapCache(page))
177 		u |= 1 << KPF_SWAPCACHE;
178 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
179 
180 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
181 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
182 
183 #ifdef CONFIG_MEMORY_FAILURE
184 	u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
185 #endif
186 
187 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
188 	u |= kpf_copy_bit(k, KPF_UNCACHED,	PG_uncached);
189 #endif
190 
191 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
192 	u |= kpf_copy_bit(k, KPF_MAPPEDTODISK,	PG_mappedtodisk);
193 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
194 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
195 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
196 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
197 
198 	return u;
199 };
200 
201 static ssize_t kpageflags_read(struct file *file, char __user *buf,
202 			     size_t count, loff_t *ppos)
203 {
204 	u64 __user *out = (u64 __user *)buf;
205 	struct page *ppage;
206 	unsigned long src = *ppos;
207 	unsigned long pfn;
208 	ssize_t ret = 0;
209 
210 	pfn = src / KPMSIZE;
211 	count = min_t(unsigned long, count, (max_pfn * KPMSIZE) - src);
212 	if (src & KPMMASK || count & KPMMASK)
213 		return -EINVAL;
214 
215 	while (count > 0) {
216 		if (pfn_valid(pfn))
217 			ppage = pfn_to_page(pfn);
218 		else
219 			ppage = NULL;
220 
221 		if (put_user(stable_page_flags(ppage), out)) {
222 			ret = -EFAULT;
223 			break;
224 		}
225 
226 		pfn++;
227 		out++;
228 		count -= KPMSIZE;
229 
230 		cond_resched();
231 	}
232 
233 	*ppos += (char __user *)out - buf;
234 	if (!ret)
235 		ret = (char __user *)out - buf;
236 	return ret;
237 }
238 
239 static const struct file_operations proc_kpageflags_operations = {
240 	.llseek = mem_lseek,
241 	.read = kpageflags_read,
242 };
243 
244 #ifdef CONFIG_MEMCG
245 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
246 				size_t count, loff_t *ppos)
247 {
248 	u64 __user *out = (u64 __user *)buf;
249 	struct page *ppage;
250 	unsigned long src = *ppos;
251 	unsigned long pfn;
252 	ssize_t ret = 0;
253 	u64 ino;
254 
255 	pfn = src / KPMSIZE;
256 	count = min_t(unsigned long, count, (max_pfn * KPMSIZE) - src);
257 	if (src & KPMMASK || count & KPMMASK)
258 		return -EINVAL;
259 
260 	while (count > 0) {
261 		if (pfn_valid(pfn))
262 			ppage = pfn_to_page(pfn);
263 		else
264 			ppage = NULL;
265 
266 		if (ppage)
267 			ino = page_cgroup_ino(ppage);
268 		else
269 			ino = 0;
270 
271 		if (put_user(ino, out)) {
272 			ret = -EFAULT;
273 			break;
274 		}
275 
276 		pfn++;
277 		out++;
278 		count -= KPMSIZE;
279 
280 		cond_resched();
281 	}
282 
283 	*ppos += (char __user *)out - buf;
284 	if (!ret)
285 		ret = (char __user *)out - buf;
286 	return ret;
287 }
288 
289 static const struct file_operations proc_kpagecgroup_operations = {
290 	.llseek = mem_lseek,
291 	.read = kpagecgroup_read,
292 };
293 #endif /* CONFIG_MEMCG */
294 
295 static int __init proc_page_init(void)
296 {
297 	proc_create("kpagecount", S_IRUSR, NULL, &proc_kpagecount_operations);
298 	proc_create("kpageflags", S_IRUSR, NULL, &proc_kpageflags_operations);
299 #ifdef CONFIG_MEMCG
300 	proc_create("kpagecgroup", S_IRUSR, NULL, &proc_kpagecgroup_operations);
301 #endif
302 	return 0;
303 }
304 fs_initcall(proc_page_init);
305