xref: /openbmc/linux/fs/proc/page.c (revision 92abee9c)
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 #define KPMBITS (KPMSIZE * BITS_PER_BYTE)
24 
get_max_dump_pfn(void)25 static inline unsigned long get_max_dump_pfn(void)
26 {
27 #ifdef CONFIG_SPARSEMEM
28 	/*
29 	 * The memmap of early sections is completely populated and marked
30 	 * online even if max_pfn does not fall on a section boundary -
31 	 * pfn_to_online_page() will succeed on all pages. Allow inspecting
32 	 * these memmaps.
33 	 */
34 	return round_up(max_pfn, PAGES_PER_SECTION);
35 #else
36 	return max_pfn;
37 #endif
38 }
39 
40 /* /proc/kpagecount - an array exposing page counts
41  *
42  * Each entry is a u64 representing the corresponding
43  * physical page count.
44  */
kpagecount_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)45 static ssize_t kpagecount_read(struct file *file, char __user *buf,
46 			     size_t count, loff_t *ppos)
47 {
48 	const unsigned long max_dump_pfn = get_max_dump_pfn();
49 	u64 __user *out = (u64 __user *)buf;
50 	struct page *ppage;
51 	unsigned long src = *ppos;
52 	unsigned long pfn;
53 	ssize_t ret = 0;
54 	u64 pcount;
55 
56 	pfn = src / KPMSIZE;
57 	if (src & KPMMASK || count & KPMMASK)
58 		return -EINVAL;
59 	if (src >= max_dump_pfn * KPMSIZE)
60 		return 0;
61 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
62 
63 	while (count > 0) {
64 		/*
65 		 * TODO: ZONE_DEVICE support requires to identify
66 		 * memmaps that were actually initialized.
67 		 */
68 		ppage = pfn_to_online_page(pfn);
69 
70 		if (!ppage)
71 			pcount = 0;
72 		else
73 			pcount = page_mapcount(ppage);
74 
75 		if (put_user(pcount, out)) {
76 			ret = -EFAULT;
77 			break;
78 		}
79 
80 		pfn++;
81 		out++;
82 		count -= KPMSIZE;
83 
84 		cond_resched();
85 	}
86 
87 	*ppos += (char __user *)out - buf;
88 	if (!ret)
89 		ret = (char __user *)out - buf;
90 	return ret;
91 }
92 
93 static const struct proc_ops kpagecount_proc_ops = {
94 	.proc_flags	= PROC_ENTRY_PERMANENT,
95 	.proc_lseek	= mem_lseek,
96 	.proc_read	= kpagecount_read,
97 };
98 
99 /* /proc/kpageflags - an array exposing page flags
100  *
101  * Each entry is a u64 representing the corresponding
102  * physical page flags.
103  */
104 
kpf_copy_bit(u64 kflags,int ubit,int kbit)105 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
106 {
107 	return ((kflags >> kbit) & 1) << ubit;
108 }
109 
stable_page_flags(struct page * page)110 u64 stable_page_flags(struct page *page)
111 {
112 	u64 k;
113 	u64 u;
114 
115 	/*
116 	 * pseudo flag: KPF_NOPAGE
117 	 * it differentiates a memory hole from a page with no flags
118 	 */
119 	if (!page)
120 		return 1 << KPF_NOPAGE;
121 
122 	k = page->flags;
123 	u = 0;
124 
125 	/*
126 	 * pseudo flags for the well known (anonymous) memory mapped pages
127 	 */
128 	if (page_mapped(page))
129 		u |= 1 << KPF_MMAP;
130 	if (PageAnon(page))
131 		u |= 1 << KPF_ANON;
132 	if (PageKsm(page))
133 		u |= 1 << KPF_KSM;
134 
135 	/*
136 	 * compound pages: export both head/tail info
137 	 * they together define a compound page's start/end pos and order
138 	 */
139 	if (PageHead(page))
140 		u |= 1 << KPF_COMPOUND_HEAD;
141 	if (PageTail(page))
142 		u |= 1 << KPF_COMPOUND_TAIL;
143 	if (PageHuge(page))
144 		u |= 1 << KPF_HUGE;
145 	/*
146 	 * PageTransCompound can be true for non-huge compound pages (slab
147 	 * pages or pages allocated by drivers with __GFP_COMP) because it
148 	 * just checks PG_head/PG_tail, so we need to check PageLRU/PageAnon
149 	 * to make sure a given page is a thp, not a non-huge compound page.
150 	 */
151 	else if (PageTransCompound(page)) {
152 		struct page *head = compound_head(page);
153 
154 		if (PageLRU(head) || PageAnon(head))
155 			u |= 1 << KPF_THP;
156 		else if (is_huge_zero_page(head)) {
157 			u |= 1 << KPF_ZERO_PAGE;
158 			u |= 1 << KPF_THP;
159 		}
160 	} else if (is_zero_pfn(page_to_pfn(page)))
161 		u |= 1 << KPF_ZERO_PAGE;
162 
163 
164 	/*
165 	 * Caveats on high order pages: PG_buddy and PG_slab will only be set
166 	 * on the head page.
167 	 */
168 	if (PageBuddy(page))
169 		u |= 1 << KPF_BUDDY;
170 	else if (page_count(page) == 0 && is_free_buddy_page(page))
171 		u |= 1 << KPF_BUDDY;
172 
173 	if (PageOffline(page))
174 		u |= 1 << KPF_OFFLINE;
175 	if (PageTable(page))
176 		u |= 1 << KPF_PGTABLE;
177 
178 	if (page_is_idle(page))
179 		u |= 1 << KPF_IDLE;
180 
181 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
182 
183 	u |= kpf_copy_bit(k, KPF_SLAB,		PG_slab);
184 	if (PageTail(page) && PageSlab(page))
185 		u |= 1 << KPF_SLAB;
186 
187 	u |= kpf_copy_bit(k, KPF_ERROR,		PG_error);
188 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
189 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
190 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
191 
192 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
193 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
194 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
195 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
196 
197 	if (PageSwapCache(page))
198 		u |= 1 << KPF_SWAPCACHE;
199 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
200 
201 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
202 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
203 
204 #ifdef CONFIG_MEMORY_FAILURE
205 	u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
206 #endif
207 
208 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
209 	u |= kpf_copy_bit(k, KPF_UNCACHED,	PG_uncached);
210 #endif
211 
212 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
213 	u |= kpf_copy_bit(k, KPF_MAPPEDTODISK,	PG_mappedtodisk);
214 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
215 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
216 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
217 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
218 #ifdef CONFIG_ARCH_USES_PG_ARCH_X
219 	u |= kpf_copy_bit(k, KPF_ARCH_2,	PG_arch_2);
220 	u |= kpf_copy_bit(k, KPF_ARCH_3,	PG_arch_3);
221 #endif
222 
223 	return u;
224 };
225 
kpageflags_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)226 static ssize_t kpageflags_read(struct file *file, char __user *buf,
227 			     size_t count, loff_t *ppos)
228 {
229 	const unsigned long max_dump_pfn = get_max_dump_pfn();
230 	u64 __user *out = (u64 __user *)buf;
231 	struct page *ppage;
232 	unsigned long src = *ppos;
233 	unsigned long pfn;
234 	ssize_t ret = 0;
235 
236 	pfn = src / KPMSIZE;
237 	if (src & KPMMASK || count & KPMMASK)
238 		return -EINVAL;
239 	if (src >= max_dump_pfn * KPMSIZE)
240 		return 0;
241 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
242 
243 	while (count > 0) {
244 		/*
245 		 * TODO: ZONE_DEVICE support requires to identify
246 		 * memmaps that were actually initialized.
247 		 */
248 		ppage = pfn_to_online_page(pfn);
249 
250 		if (put_user(stable_page_flags(ppage), out)) {
251 			ret = -EFAULT;
252 			break;
253 		}
254 
255 		pfn++;
256 		out++;
257 		count -= KPMSIZE;
258 
259 		cond_resched();
260 	}
261 
262 	*ppos += (char __user *)out - buf;
263 	if (!ret)
264 		ret = (char __user *)out - buf;
265 	return ret;
266 }
267 
268 static const struct proc_ops kpageflags_proc_ops = {
269 	.proc_flags	= PROC_ENTRY_PERMANENT,
270 	.proc_lseek	= mem_lseek,
271 	.proc_read	= kpageflags_read,
272 };
273 
274 #ifdef CONFIG_MEMCG
kpagecgroup_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)275 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
276 				size_t count, loff_t *ppos)
277 {
278 	const unsigned long max_dump_pfn = get_max_dump_pfn();
279 	u64 __user *out = (u64 __user *)buf;
280 	struct page *ppage;
281 	unsigned long src = *ppos;
282 	unsigned long pfn;
283 	ssize_t ret = 0;
284 	u64 ino;
285 
286 	pfn = src / KPMSIZE;
287 	if (src & KPMMASK || count & KPMMASK)
288 		return -EINVAL;
289 	if (src >= max_dump_pfn * KPMSIZE)
290 		return 0;
291 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
292 
293 	while (count > 0) {
294 		/*
295 		 * TODO: ZONE_DEVICE support requires to identify
296 		 * memmaps that were actually initialized.
297 		 */
298 		ppage = pfn_to_online_page(pfn);
299 
300 		if (ppage)
301 			ino = page_cgroup_ino(ppage);
302 		else
303 			ino = 0;
304 
305 		if (put_user(ino, out)) {
306 			ret = -EFAULT;
307 			break;
308 		}
309 
310 		pfn++;
311 		out++;
312 		count -= KPMSIZE;
313 
314 		cond_resched();
315 	}
316 
317 	*ppos += (char __user *)out - buf;
318 	if (!ret)
319 		ret = (char __user *)out - buf;
320 	return ret;
321 }
322 
323 static const struct proc_ops kpagecgroup_proc_ops = {
324 	.proc_flags	= PROC_ENTRY_PERMANENT,
325 	.proc_lseek	= mem_lseek,
326 	.proc_read	= kpagecgroup_read,
327 };
328 #endif /* CONFIG_MEMCG */
329 
proc_page_init(void)330 static int __init proc_page_init(void)
331 {
332 	proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
333 	proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
334 #ifdef CONFIG_MEMCG
335 	proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
336 #endif
337 	return 0;
338 }
339 fs_initcall(proc_page_init);
340