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>
9 #include <linux/migrate.h>
12 static bool page_owner_disabled
= true;
13 DEFINE_STATIC_KEY_FALSE(page_owner_inited
);
15 static void init_early_allocated_pages(void);
17 static int early_page_owner_param(char *buf
)
22 if (strcmp(buf
, "on") == 0)
23 page_owner_disabled
= false;
27 early_param("page_owner", early_page_owner_param
);
29 static bool need_page_owner(void)
31 if (page_owner_disabled
)
37 static void init_page_owner(void)
39 if (page_owner_disabled
)
42 static_branch_enable(&page_owner_inited
);
43 init_early_allocated_pages();
46 struct page_ext_operations page_owner_ops
= {
47 .need
= need_page_owner
,
48 .init
= init_page_owner
,
51 void __reset_page_owner(struct page
*page
, unsigned int order
)
54 struct page_ext
*page_ext
;
56 for (i
= 0; i
< (1 << order
); i
++) {
57 page_ext
= lookup_page_ext(page
+ i
);
58 if (unlikely(!page_ext
))
60 __clear_bit(PAGE_EXT_OWNER
, &page_ext
->flags
);
64 void __set_page_owner(struct page
*page
, unsigned int order
, gfp_t gfp_mask
)
66 struct page_ext
*page_ext
= lookup_page_ext(page
);
68 struct stack_trace trace
= {
70 .max_entries
= ARRAY_SIZE(page_ext
->trace_entries
),
71 .entries
= &page_ext
->trace_entries
[0],
75 if (unlikely(!page_ext
))
78 save_stack_trace(&trace
);
80 page_ext
->order
= order
;
81 page_ext
->gfp_mask
= gfp_mask
;
82 page_ext
->nr_entries
= trace
.nr_entries
;
83 page_ext
->last_migrate_reason
= -1;
85 __set_bit(PAGE_EXT_OWNER
, &page_ext
->flags
);
88 void __set_page_owner_migrate_reason(struct page
*page
, int reason
)
90 struct page_ext
*page_ext
= lookup_page_ext(page
);
91 if (unlikely(!page_ext
))
94 page_ext
->last_migrate_reason
= reason
;
97 void __split_page_owner(struct page
*page
, unsigned int order
)
100 struct page_ext
*page_ext
= lookup_page_ext(page
);
102 if (unlikely(!page_ext
))
106 for (i
= 1; i
< (1 << order
); i
++)
107 __copy_page_owner(page
, page
+ i
);
110 void __copy_page_owner(struct page
*oldpage
, struct page
*newpage
)
112 struct page_ext
*old_ext
= lookup_page_ext(oldpage
);
113 struct page_ext
*new_ext
= lookup_page_ext(newpage
);
116 if (unlikely(!old_ext
|| !new_ext
))
119 new_ext
->order
= old_ext
->order
;
120 new_ext
->gfp_mask
= old_ext
->gfp_mask
;
121 new_ext
->last_migrate_reason
= old_ext
->last_migrate_reason
;
122 new_ext
->nr_entries
= old_ext
->nr_entries
;
124 for (i
= 0; i
< ARRAY_SIZE(new_ext
->trace_entries
); i
++)
125 new_ext
->trace_entries
[i
] = old_ext
->trace_entries
[i
];
128 * We don't clear the bit on the oldpage as it's going to be freed
129 * after migration. Until then, the info can be useful in case of
130 * a bug, and the overal stats will be off a bit only temporarily.
131 * Also, migrate_misplaced_transhuge_page() can still fail the
132 * migration and then we want the oldpage to retain the info. But
133 * in that case we also don't need to explicitly clear the info from
134 * the new page, which will be freed.
136 __set_bit(PAGE_EXT_OWNER
, &new_ext
->flags
);
140 print_page_owner(char __user
*buf
, size_t count
, unsigned long pfn
,
141 struct page
*page
, struct page_ext
*page_ext
)
144 int pageblock_mt
, page_mt
;
146 struct stack_trace trace
= {
147 .nr_entries
= page_ext
->nr_entries
,
148 .entries
= &page_ext
->trace_entries
[0],
151 kbuf
= kmalloc(count
, GFP_KERNEL
);
155 ret
= snprintf(kbuf
, count
,
156 "Page allocated via order %u, mask %#x(%pGg)\n",
157 page_ext
->order
, page_ext
->gfp_mask
,
158 &page_ext
->gfp_mask
);
163 /* Print information relevant to grouping pages by mobility */
164 pageblock_mt
= get_pageblock_migratetype(page
);
165 page_mt
= gfpflags_to_migratetype(page_ext
->gfp_mask
);
166 ret
+= snprintf(kbuf
+ ret
, count
- ret
,
167 "PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n",
169 migratetype_names
[page_mt
],
170 pfn
>> pageblock_order
,
171 migratetype_names
[pageblock_mt
],
172 page
->flags
, &page
->flags
);
177 ret
+= snprint_stack_trace(kbuf
+ ret
, count
- ret
, &trace
, 0);
181 if (page_ext
->last_migrate_reason
!= -1) {
182 ret
+= snprintf(kbuf
+ ret
, count
- ret
,
183 "Page has been migrated, last migrate reason: %s\n",
184 migrate_reason_names
[page_ext
->last_migrate_reason
]);
189 ret
+= snprintf(kbuf
+ ret
, count
- ret
, "\n");
193 if (copy_to_user(buf
, kbuf
, ret
))
204 void __dump_page_owner(struct page
*page
)
206 struct page_ext
*page_ext
= lookup_page_ext(page
);
207 struct stack_trace trace
= {
208 .nr_entries
= page_ext
->nr_entries
,
209 .entries
= &page_ext
->trace_entries
[0],
214 if (unlikely(!page_ext
)) {
215 pr_alert("There is not page extension available.\n");
218 gfp_mask
= page_ext
->gfp_mask
;
219 mt
= gfpflags_to_migratetype(gfp_mask
);
221 if (!test_bit(PAGE_EXT_OWNER
, &page_ext
->flags
)) {
222 pr_alert("page_owner info is not active (free page?)\n");
226 pr_alert("page allocated via order %u, migratetype %s, gfp_mask %#x(%pGg)\n",
227 page_ext
->order
, migratetype_names
[mt
], gfp_mask
, &gfp_mask
);
228 print_stack_trace(&trace
, 0);
230 if (page_ext
->last_migrate_reason
!= -1)
231 pr_alert("page has been migrated, last migrate reason: %s\n",
232 migrate_reason_names
[page_ext
->last_migrate_reason
]);
236 read_page_owner(struct file
*file
, char __user
*buf
, size_t count
, loff_t
*ppos
)
240 struct page_ext
*page_ext
;
242 if (!static_branch_unlikely(&page_owner_inited
))
246 pfn
= min_low_pfn
+ *ppos
;
248 /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
249 while (!pfn_valid(pfn
) && (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0)
252 drain_all_pages(NULL
);
254 /* Find an allocated page */
255 for (; pfn
< max_pfn
; pfn
++) {
257 * If the new page is in a new MAX_ORDER_NR_PAGES area,
258 * validate the area as existing, skip it if not
260 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0 && !pfn_valid(pfn
)) {
261 pfn
+= MAX_ORDER_NR_PAGES
- 1;
265 /* Check for holes within a MAX_ORDER area */
266 if (!pfn_valid_within(pfn
))
269 page
= pfn_to_page(pfn
);
270 if (PageBuddy(page
)) {
271 unsigned long freepage_order
= page_order_unsafe(page
);
273 if (freepage_order
< MAX_ORDER
)
274 pfn
+= (1UL << freepage_order
) - 1;
278 page_ext
= lookup_page_ext(page
);
279 if (unlikely(!page_ext
))
283 * Some pages could be missed by concurrent allocation or free,
284 * because we don't hold the zone lock.
286 if (!test_bit(PAGE_EXT_OWNER
, &page_ext
->flags
))
289 /* Record the next PFN to read in the file offset */
290 *ppos
= (pfn
- min_low_pfn
) + 1;
292 return print_page_owner(buf
, count
, pfn
, page
, page_ext
);
298 static void init_pages_in_zone(pg_data_t
*pgdat
, struct zone
*zone
)
301 struct page_ext
*page_ext
;
302 unsigned long pfn
= zone
->zone_start_pfn
, block_end_pfn
;
303 unsigned long end_pfn
= pfn
+ zone
->spanned_pages
;
304 unsigned long count
= 0;
306 /* Scan block by block. First and last block may be incomplete */
307 pfn
= zone
->zone_start_pfn
;
310 * Walk the zone in pageblock_nr_pages steps. If a page block spans
311 * a zone boundary, it will be double counted between zones. This does
312 * not matter as the mixed block count will still be correct
314 for (; pfn
< end_pfn
; ) {
315 if (!pfn_valid(pfn
)) {
316 pfn
= ALIGN(pfn
+ 1, MAX_ORDER_NR_PAGES
);
320 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
321 block_end_pfn
= min(block_end_pfn
, end_pfn
);
323 page
= pfn_to_page(pfn
);
325 for (; pfn
< block_end_pfn
; pfn
++) {
326 if (!pfn_valid_within(pfn
))
329 page
= pfn_to_page(pfn
);
331 if (page_zone(page
) != zone
)
335 * We are safe to check buddy flag and order, because
336 * this is init stage and only single thread runs.
338 if (PageBuddy(page
)) {
339 pfn
+= (1UL << page_order(page
)) - 1;
343 if (PageReserved(page
))
346 page_ext
= lookup_page_ext(page
);
347 if (unlikely(!page_ext
))
350 /* Maybe overraping zone */
351 if (test_bit(PAGE_EXT_OWNER
, &page_ext
->flags
))
354 /* Found early allocated page */
355 set_page_owner(page
, 0, 0);
360 pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
361 pgdat
->node_id
, zone
->name
, count
);
364 static void init_zones_in_node(pg_data_t
*pgdat
)
367 struct zone
*node_zones
= pgdat
->node_zones
;
370 for (zone
= node_zones
; zone
- node_zones
< MAX_NR_ZONES
; ++zone
) {
371 if (!populated_zone(zone
))
374 spin_lock_irqsave(&zone
->lock
, flags
);
375 init_pages_in_zone(pgdat
, zone
);
376 spin_unlock_irqrestore(&zone
->lock
, flags
);
380 static void init_early_allocated_pages(void)
384 drain_all_pages(NULL
);
385 for_each_online_pgdat(pgdat
)
386 init_zones_in_node(pgdat
);
389 static const struct file_operations proc_page_owner_operations
= {
390 .read
= read_page_owner
,
393 static int __init
pageowner_init(void)
395 struct dentry
*dentry
;
397 if (!static_branch_unlikely(&page_owner_inited
)) {
398 pr_info("page_owner is disabled\n");
402 dentry
= debugfs_create_file("page_owner", S_IRUSR
, NULL
,
403 NULL
, &proc_page_owner_operations
);
405 return PTR_ERR(dentry
);
409 late_initcall(pageowner_init
)