1#include <linux/debugfs.h>
2#include <linux/mm.h>
3#include <linux/slab.h>
4#include <linux/uaccess.h>
5#include <linux/bootmem.h>
6#include <linux/stacktrace.h>
7#include <linux/page_owner.h>
8#include "internal.h"
9
10static bool page_owner_disabled = true;
11bool page_owner_inited __read_mostly;
12
13static void init_early_allocated_pages(void);
14
15static int early_page_owner_param(char *buf)
16{
17	if (!buf)
18		return -EINVAL;
19
20	if (strcmp(buf, "on") == 0)
21		page_owner_disabled = false;
22
23	return 0;
24}
25early_param("page_owner", early_page_owner_param);
26
27static bool need_page_owner(void)
28{
29	if (page_owner_disabled)
30		return false;
31
32	return true;
33}
34
35static void init_page_owner(void)
36{
37	if (page_owner_disabled)
38		return;
39
40	page_owner_inited = true;
41	init_early_allocated_pages();
42}
43
44struct page_ext_operations page_owner_ops = {
45	.need = need_page_owner,
46	.init = init_page_owner,
47};
48
49void __reset_page_owner(struct page *page, unsigned int order)
50{
51	int i;
52	struct page_ext *page_ext;
53
54	for (i = 0; i < (1 << order); i++) {
55		page_ext = lookup_page_ext(page + i);
56		__clear_bit(PAGE_EXT_OWNER, &page_ext->flags);
57	}
58}
59
60void __set_page_owner(struct page *page, unsigned int order, gfp_t gfp_mask)
61{
62	struct page_ext *page_ext = lookup_page_ext(page);
63	struct stack_trace trace = {
64		.nr_entries = 0,
65		.max_entries = ARRAY_SIZE(page_ext->trace_entries),
66		.entries = &page_ext->trace_entries[0],
67		.skip = 3,
68	};
69
70	save_stack_trace(&trace);
71
72	page_ext->order = order;
73	page_ext->gfp_mask = gfp_mask;
74	page_ext->nr_entries = trace.nr_entries;
75
76	__set_bit(PAGE_EXT_OWNER, &page_ext->flags);
77}
78
79static ssize_t
80print_page_owner(char __user *buf, size_t count, unsigned long pfn,
81		struct page *page, struct page_ext *page_ext)
82{
83	int ret;
84	int pageblock_mt, page_mt;
85	char *kbuf;
86	struct stack_trace trace = {
87		.nr_entries = page_ext->nr_entries,
88		.entries = &page_ext->trace_entries[0],
89	};
90
91	kbuf = kmalloc(count, GFP_KERNEL);
92	if (!kbuf)
93		return -ENOMEM;
94
95	ret = snprintf(kbuf, count,
96			"Page allocated via order %u, mask 0x%x\n",
97			page_ext->order, page_ext->gfp_mask);
98
99	if (ret >= count)
100		goto err;
101
102	/* Print information relevant to grouping pages by mobility */
103	pageblock_mt = get_pfnblock_migratetype(page, pfn);
104	page_mt  = gfpflags_to_migratetype(page_ext->gfp_mask);
105	ret += snprintf(kbuf + ret, count - ret,
106			"PFN %lu Block %lu type %d %s Flags %s%s%s%s%s%s%s%s%s%s%s%s\n",
107			pfn,
108			pfn >> pageblock_order,
109			pageblock_mt,
110			pageblock_mt != page_mt ? "Fallback" : "        ",
111			PageLocked(page)	? "K" : " ",
112			PageError(page)		? "E" : " ",
113			PageReferenced(page)	? "R" : " ",
114			PageUptodate(page)	? "U" : " ",
115			PageDirty(page)		? "D" : " ",
116			PageLRU(page)		? "L" : " ",
117			PageActive(page)	? "A" : " ",
118			PageSlab(page)		? "S" : " ",
119			PageWriteback(page)	? "W" : " ",
120			PageCompound(page)	? "C" : " ",
121			PageSwapCache(page)	? "B" : " ",
122			PageMappedToDisk(page)	? "M" : " ");
123
124	if (ret >= count)
125		goto err;
126
127	ret += snprint_stack_trace(kbuf + ret, count - ret, &trace, 0);
128	if (ret >= count)
129		goto err;
130
131	ret += snprintf(kbuf + ret, count - ret, "\n");
132	if (ret >= count)
133		goto err;
134
135	if (copy_to_user(buf, kbuf, ret))
136		ret = -EFAULT;
137
138	kfree(kbuf);
139	return ret;
140
141err:
142	kfree(kbuf);
143	return -ENOMEM;
144}
145
146static ssize_t
147read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos)
148{
149	unsigned long pfn;
150	struct page *page;
151	struct page_ext *page_ext;
152
153	if (!page_owner_inited)
154		return -EINVAL;
155
156	page = NULL;
157	pfn = min_low_pfn + *ppos;
158
159	/* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
160	while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0)
161		pfn++;
162
163	drain_all_pages(NULL);
164
165	/* Find an allocated page */
166	for (; pfn < max_pfn; pfn++) {
167		/*
168		 * If the new page is in a new MAX_ORDER_NR_PAGES area,
169		 * validate the area as existing, skip it if not
170		 */
171		if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) {
172			pfn += MAX_ORDER_NR_PAGES - 1;
173			continue;
174		}
175
176		/* Check for holes within a MAX_ORDER area */
177		if (!pfn_valid_within(pfn))
178			continue;
179
180		page = pfn_to_page(pfn);
181		if (PageBuddy(page)) {
182			unsigned long freepage_order = page_order_unsafe(page);
183
184			if (freepage_order < MAX_ORDER)
185				pfn += (1UL << freepage_order) - 1;
186			continue;
187		}
188
189		page_ext = lookup_page_ext(page);
190
191		/*
192		 * Some pages could be missed by concurrent allocation or free,
193		 * because we don't hold the zone lock.
194		 */
195		if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
196			continue;
197
198		/* Record the next PFN to read in the file offset */
199		*ppos = (pfn - min_low_pfn) + 1;
200
201		return print_page_owner(buf, count, pfn, page, page_ext);
202	}
203
204	return 0;
205}
206
207static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone)
208{
209	struct page *page;
210	struct page_ext *page_ext;
211	unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
212	unsigned long end_pfn = pfn + zone->spanned_pages;
213	unsigned long count = 0;
214
215	/* Scan block by block. First and last block may be incomplete */
216	pfn = zone->zone_start_pfn;
217
218	/*
219	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
220	 * a zone boundary, it will be double counted between zones. This does
221	 * not matter as the mixed block count will still be correct
222	 */
223	for (; pfn < end_pfn; ) {
224		if (!pfn_valid(pfn)) {
225			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
226			continue;
227		}
228
229		block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
230		block_end_pfn = min(block_end_pfn, end_pfn);
231
232		page = pfn_to_page(pfn);
233
234		for (; pfn < block_end_pfn; pfn++) {
235			if (!pfn_valid_within(pfn))
236				continue;
237
238			page = pfn_to_page(pfn);
239
240			/*
241			 * We are safe to check buddy flag and order, because
242			 * this is init stage and only single thread runs.
243			 */
244			if (PageBuddy(page)) {
245				pfn += (1UL << page_order(page)) - 1;
246				continue;
247			}
248
249			if (PageReserved(page))
250				continue;
251
252			page_ext = lookup_page_ext(page);
253
254			/* Maybe overraping zone */
255			if (test_bit(PAGE_EXT_OWNER, &page_ext->flags))
256				continue;
257
258			/* Found early allocated page */
259			set_page_owner(page, 0, 0);
260			count++;
261		}
262	}
263
264	pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
265		pgdat->node_id, zone->name, count);
266}
267
268static void init_zones_in_node(pg_data_t *pgdat)
269{
270	struct zone *zone;
271	struct zone *node_zones = pgdat->node_zones;
272	unsigned long flags;
273
274	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
275		if (!populated_zone(zone))
276			continue;
277
278		spin_lock_irqsave(&zone->lock, flags);
279		init_pages_in_zone(pgdat, zone);
280		spin_unlock_irqrestore(&zone->lock, flags);
281	}
282}
283
284static void init_early_allocated_pages(void)
285{
286	pg_data_t *pgdat;
287
288	drain_all_pages(NULL);
289	for_each_online_pgdat(pgdat)
290		init_zones_in_node(pgdat);
291}
292
293static const struct file_operations proc_page_owner_operations = {
294	.read		= read_page_owner,
295};
296
297static int __init pageowner_init(void)
298{
299	struct dentry *dentry;
300
301	if (!page_owner_inited) {
302		pr_info("page_owner is disabled\n");
303		return 0;
304	}
305
306	dentry = debugfs_create_file("page_owner", S_IRUSR, NULL,
307			NULL, &proc_page_owner_operations);
308	if (IS_ERR(dentry))
309		return PTR_ERR(dentry);
310
311	return 0;
312}
313module_init(pageowner_init)
314