1 #include <linux/debugfs.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 <linux/jump_label.h>
11 static bool page_owner_disabled
= true;
12 DEFINE_STATIC_KEY_FALSE(page_owner_inited
);
14 static void init_early_allocated_pages(void);
16 static int early_page_owner_param(char *buf
)
21 if (strcmp(buf
, "on") == 0)
22 page_owner_disabled
= false;
26 early_param("page_owner", early_page_owner_param
);
28 static bool need_page_owner(void)
30 if (page_owner_disabled
)
36 static void init_page_owner(void)
38 if (page_owner_disabled
)
41 static_branch_enable(&page_owner_inited
);
42 init_early_allocated_pages();
45 struct page_ext_operations page_owner_ops
= {
46 .need
= need_page_owner
,
47 .init
= init_page_owner
,
50 void __reset_page_owner(struct page
*page
, unsigned int order
)
53 struct page_ext
*page_ext
;
55 for (i
= 0; i
< (1 << order
); i
++) {
56 page_ext
= lookup_page_ext(page
+ i
);
57 __clear_bit(PAGE_EXT_OWNER
, &page_ext
->flags
);
61 void __set_page_owner(struct page
*page
, unsigned int order
, gfp_t gfp_mask
)
63 struct page_ext
*page_ext
= lookup_page_ext(page
);
64 struct stack_trace trace
= {
66 .max_entries
= ARRAY_SIZE(page_ext
->trace_entries
),
67 .entries
= &page_ext
->trace_entries
[0],
71 save_stack_trace(&trace
);
73 page_ext
->order
= order
;
74 page_ext
->gfp_mask
= gfp_mask
;
75 page_ext
->nr_entries
= trace
.nr_entries
;
77 __set_bit(PAGE_EXT_OWNER
, &page_ext
->flags
);
80 gfp_t
__get_page_owner_gfp(struct page
*page
)
82 struct page_ext
*page_ext
= lookup_page_ext(page
);
84 return page_ext
->gfp_mask
;
88 print_page_owner(char __user
*buf
, size_t count
, unsigned long pfn
,
89 struct page
*page
, struct page_ext
*page_ext
)
92 int pageblock_mt
, page_mt
;
94 struct stack_trace trace
= {
95 .nr_entries
= page_ext
->nr_entries
,
96 .entries
= &page_ext
->trace_entries
[0],
99 kbuf
= kmalloc(count
, GFP_KERNEL
);
103 ret
= snprintf(kbuf
, count
,
104 "Page allocated via order %u, mask %#x(%pGg)\n",
105 page_ext
->order
, page_ext
->gfp_mask
,
106 &page_ext
->gfp_mask
);
111 /* Print information relevant to grouping pages by mobility */
112 pageblock_mt
= get_pfnblock_migratetype(page
, pfn
);
113 page_mt
= gfpflags_to_migratetype(page_ext
->gfp_mask
);
114 ret
+= snprintf(kbuf
+ ret
, count
- ret
,
115 "PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n",
117 migratetype_names
[page_mt
],
118 pfn
>> pageblock_order
,
119 migratetype_names
[pageblock_mt
],
120 page
->flags
, &page
->flags
);
125 ret
+= snprint_stack_trace(kbuf
+ ret
, count
- ret
, &trace
, 0);
129 ret
+= snprintf(kbuf
+ ret
, count
- ret
, "\n");
133 if (copy_to_user(buf
, kbuf
, ret
))
145 read_page_owner(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*ppos
)
149 struct page_ext
*page_ext
;
151 if (!static_branch_unlikely(&page_owner_inited
))
155 pfn
= min_low_pfn
+ *ppos
;
157 /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
158 while (!pfn_valid(pfn
) && (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0)
161 drain_all_pages(NULL
);
163 /* Find an allocated page */
164 for (; pfn
< max_pfn
; pfn
++) {
166 * If the new page is in a new MAX_ORDER_NR_PAGES area,
167 * validate the area as existing, skip it if not
169 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0 && !pfn_valid(pfn
)) {
170 pfn
+= MAX_ORDER_NR_PAGES
- 1;
174 /* Check for holes within a MAX_ORDER area */
175 if (!pfn_valid_within(pfn
))
178 page
= pfn_to_page(pfn
);
179 if (PageBuddy(page
)) {
180 unsigned long freepage_order
= page_order_unsafe(page
);
182 if (freepage_order
< MAX_ORDER
)
183 pfn
+= (1UL << freepage_order
) - 1;
187 page_ext
= lookup_page_ext(page
);
190 * Some pages could be missed by concurrent allocation or free,
191 * because we don't hold the zone lock.
193 if (!test_bit(PAGE_EXT_OWNER
, &page_ext
->flags
))
196 /* Record the next PFN to read in the file offset */
197 *ppos
= (pfn
- min_low_pfn
) + 1;
199 return print_page_owner(buf
, count
, pfn
, page
, page_ext
);
205 static void init_pages_in_zone(pg_data_t
*pgdat
, struct zone
*zone
)
208 struct page_ext
*page_ext
;
209 unsigned long pfn
= zone
->zone_start_pfn
, block_end_pfn
;
210 unsigned long end_pfn
= pfn
+ zone
->spanned_pages
;
211 unsigned long count
= 0;
213 /* Scan block by block. First and last block may be incomplete */
214 pfn
= zone
->zone_start_pfn
;
217 * Walk the zone in pageblock_nr_pages steps. If a page block spans
218 * a zone boundary, it will be double counted between zones. This does
219 * not matter as the mixed block count will still be correct
221 for (; pfn
< end_pfn
; ) {
222 if (!pfn_valid(pfn
)) {
223 pfn
= ALIGN(pfn
+ 1, MAX_ORDER_NR_PAGES
);
227 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
228 block_end_pfn
= min(block_end_pfn
, end_pfn
);
230 page
= pfn_to_page(pfn
);
232 for (; pfn
< block_end_pfn
; pfn
++) {
233 if (!pfn_valid_within(pfn
))
236 page
= pfn_to_page(pfn
);
239 * We are safe to check buddy flag and order, because
240 * this is init stage and only single thread runs.
242 if (PageBuddy(page
)) {
243 pfn
+= (1UL << page_order(page
)) - 1;
247 if (PageReserved(page
))
250 page_ext
= lookup_page_ext(page
);
252 /* Maybe overraping zone */
253 if (test_bit(PAGE_EXT_OWNER
, &page_ext
->flags
))
256 /* Found early allocated page */
257 set_page_owner(page
, 0, 0);
262 pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
263 pgdat
->node_id
, zone
->name
, count
);
266 static void init_zones_in_node(pg_data_t
*pgdat
)
269 struct zone
*node_zones
= pgdat
->node_zones
;
272 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
273 if (!populated_zone(zone
))
276 spin_lock_irqsave(&zone
->lock
, flags
);
277 init_pages_in_zone(pgdat
, zone
);
278 spin_unlock_irqrestore(&zone
->lock
, flags
);
282 static void init_early_allocated_pages(void)
286 drain_all_pages(NULL
);
287 for_each_online_pgdat(pgdat
)
288 init_zones_in_node(pgdat
);
291 static const struct file_operations proc_page_owner_operations
= {
292 .read
= read_page_owner
,
295 static int __init
pageowner_init(void)
297 struct dentry
*dentry
;
299 if (!static_branch_unlikely(&page_owner_inited
)) {
300 pr_info("page_owner is disabled\n");
304 dentry
= debugfs_create_file("page_owner", S_IRUSR
, NULL
,
305 NULL
, &proc_page_owner_operations
);
307 return PTR_ERR(dentry
);
311 late_initcall(pageowner_init
)