blob: fe850b9e30995463e4f5be31eb53b93f4068130f [file] [log] [blame]
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001#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>
Vlastimil Babka7dd80b82016-03-15 14:56:12 -07008#include <linux/jump_label.h>
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07009#include <linux/migrate.h>
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070010#include <linux/stackdepot.h>
Joonsoo Kime2f612e2016-10-07 16:58:21 -070011#include <linux/seq_file.h>
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070012
Joonsoo Kim48c96a32014-12-12 16:56:01 -080013#include "internal.h"
14
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070015/*
16 * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
17 * to use off stack temporal storage
18 */
19#define PAGE_OWNER_STACK_DEPTH (16)
20
Joonsoo Kim9300d8d2016-10-07 16:58:30 -070021struct page_owner {
22 unsigned int order;
23 gfp_t gfp_mask;
24 int last_migrate_reason;
25 depot_stack_handle_t handle;
26};
27
Trilok Soni65cfa9f2016-04-20 18:12:51 -070028static bool page_owner_disabled =
29 !IS_ENABLED(CONFIG_PAGE_OWNER_ENABLE_DEFAULT);
Vlastimil Babka7dd80b82016-03-15 14:56:12 -070030DEFINE_STATIC_KEY_FALSE(page_owner_inited);
Joonsoo Kim48c96a32014-12-12 16:56:01 -080031
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070032static depot_stack_handle_t dummy_handle;
33static depot_stack_handle_t failure_handle;
34
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -080035static void init_early_allocated_pages(void);
36
Joonsoo Kim48c96a32014-12-12 16:56:01 -080037static int early_page_owner_param(char *buf)
38{
39 if (!buf)
40 return -EINVAL;
41
42 if (strcmp(buf, "on") == 0)
43 page_owner_disabled = false;
44
Trilok Soni65cfa9f2016-04-20 18:12:51 -070045 if (strcmp(buf, "off") == 0)
46 page_owner_disabled = true;
47
Joonsoo Kim48c96a32014-12-12 16:56:01 -080048 return 0;
49}
50early_param("page_owner", early_page_owner_param);
51
52static bool need_page_owner(void)
53{
54 if (page_owner_disabled)
55 return false;
56
57 return true;
58}
59
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070060static noinline void register_dummy_stack(void)
61{
62 unsigned long entries[4];
63 struct stack_trace dummy;
64
65 dummy.nr_entries = 0;
66 dummy.max_entries = ARRAY_SIZE(entries);
67 dummy.entries = &entries[0];
68 dummy.skip = 0;
69
70 save_stack_trace(&dummy);
71 dummy_handle = depot_save_stack(&dummy, GFP_KERNEL);
72}
73
74static noinline void register_failure_stack(void)
75{
76 unsigned long entries[4];
77 struct stack_trace failure;
78
79 failure.nr_entries = 0;
80 failure.max_entries = ARRAY_SIZE(entries);
81 failure.entries = &entries[0];
82 failure.skip = 0;
83
84 save_stack_trace(&failure);
85 failure_handle = depot_save_stack(&failure, GFP_KERNEL);
86}
87
Joonsoo Kim48c96a32014-12-12 16:56:01 -080088static void init_page_owner(void)
89{
90 if (page_owner_disabled)
91 return;
92
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -070093 register_dummy_stack();
94 register_failure_stack();
Vlastimil Babka7dd80b82016-03-15 14:56:12 -070095 static_branch_enable(&page_owner_inited);
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -080096 init_early_allocated_pages();
Joonsoo Kim48c96a32014-12-12 16:56:01 -080097}
98
99struct page_ext_operations page_owner_ops = {
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700100 .size = sizeof(struct page_owner),
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800101 .need = need_page_owner,
102 .init = init_page_owner,
103};
104
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700105static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
106{
107 return (void *)page_ext + page_owner_ops.offset;
108}
109
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800110void __reset_page_owner(struct page *page, unsigned int order)
111{
112 int i;
113 struct page_ext *page_ext;
114
115 for (i = 0; i < (1 << order); i++) {
116 page_ext = lookup_page_ext(page + i);
Yang Shif86e4272016-06-03 14:55:38 -0700117 if (unlikely(!page_ext))
118 continue;
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800119 __clear_bit(PAGE_EXT_OWNER, &page_ext->flags);
120 }
121}
122
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700123static inline bool check_recursive_alloc(struct stack_trace *trace,
124 unsigned long ip)
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800125{
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700126 int i, count;
Yang Shif86e4272016-06-03 14:55:38 -0700127
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700128 if (!trace->nr_entries)
129 return false;
130
131 for (i = 0, count = 0; i < trace->nr_entries; i++) {
132 if (trace->entries[i] == ip && ++count == 2)
133 return true;
134 }
135
136 return false;
137}
138
139static noinline depot_stack_handle_t save_stack(gfp_t flags)
140{
141 unsigned long entries[PAGE_OWNER_STACK_DEPTH];
Sergei Rogachev94f759d62015-02-11 15:28:34 -0800142 struct stack_trace trace = {
143 .nr_entries = 0,
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700144 .entries = entries,
145 .max_entries = PAGE_OWNER_STACK_DEPTH,
Prakash Gupta8f3c7cf2017-09-07 10:25:35 +1000146 .skip = 2
Sergei Rogachev94f759d62015-02-11 15:28:34 -0800147 };
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700148 depot_stack_handle_t handle;
149
150 save_stack_trace(&trace);
151 if (trace.nr_entries != 0 &&
152 trace.entries[trace.nr_entries-1] == ULONG_MAX)
153 trace.nr_entries--;
154
155 /*
156 * We need to check recursion here because our request to stackdepot
157 * could trigger memory allocation to save new entry. New memory
158 * allocation would reach here and call depot_save_stack() again
159 * if we don't catch it. There is still not enough memory in stackdepot
160 * so it would try to allocate memory again and loop forever.
161 */
162 if (check_recursive_alloc(&trace, _RET_IP_))
163 return dummy_handle;
164
165 handle = depot_save_stack(&trace, flags);
166 if (!handle)
167 handle = failure_handle;
168
169 return handle;
170}
171
172noinline void __set_page_owner(struct page *page, unsigned int order,
173 gfp_t gfp_mask)
174{
175 struct page_ext *page_ext = lookup_page_ext(page);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700176 struct page_owner *page_owner;
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800177
Yang Shif86e4272016-06-03 14:55:38 -0700178 if (unlikely(!page_ext))
179 return;
180
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700181 page_owner = get_page_owner(page_ext);
182 page_owner->handle = save_stack(gfp_mask);
183 page_owner->order = order;
184 page_owner->gfp_mask = gfp_mask;
185 page_owner->last_migrate_reason = -1;
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800186
187 __set_bit(PAGE_EXT_OWNER, &page_ext->flags);
188}
189
Vlastimil Babka7cd12b42016-03-15 14:56:18 -0700190void __set_page_owner_migrate_reason(struct page *page, int reason)
191{
192 struct page_ext *page_ext = lookup_page_ext(page);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700193 struct page_owner *page_owner;
194
Yang Shif86e4272016-06-03 14:55:38 -0700195 if (unlikely(!page_ext))
196 return;
Vlastimil Babka7cd12b42016-03-15 14:56:18 -0700197
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700198 page_owner = get_page_owner(page_ext);
199 page_owner->last_migrate_reason = reason;
Vlastimil Babka7cd12b42016-03-15 14:56:18 -0700200}
201
Joonsoo Kima9627bc2016-07-26 15:23:49 -0700202void __split_page_owner(struct page *page, unsigned int order)
Joonsoo Kime2cfc912015-07-17 16:24:18 -0700203{
Joonsoo Kima9627bc2016-07-26 15:23:49 -0700204 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -0700205 struct page_ext *page_ext = lookup_page_ext(page);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700206 struct page_owner *page_owner;
Joonsoo Kime2cfc912015-07-17 16:24:18 -0700207
Joonsoo Kima9627bc2016-07-26 15:23:49 -0700208 if (unlikely(!page_ext))
209 return;
210
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700211 page_owner = get_page_owner(page_ext);
212 page_owner->order = 0;
Joonsoo Kima9627bc2016-07-26 15:23:49 -0700213 for (i = 1; i < (1 << order); i++)
214 __copy_page_owner(page, page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -0700215}
216
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700217void __copy_page_owner(struct page *oldpage, struct page *newpage)
218{
219 struct page_ext *old_ext = lookup_page_ext(oldpage);
220 struct page_ext *new_ext = lookup_page_ext(newpage);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700221 struct page_owner *old_page_owner, *new_page_owner;
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700222
Yang Shif86e4272016-06-03 14:55:38 -0700223 if (unlikely(!old_ext || !new_ext))
224 return;
225
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700226 old_page_owner = get_page_owner(old_ext);
227 new_page_owner = get_page_owner(new_ext);
228 new_page_owner->order = old_page_owner->order;
229 new_page_owner->gfp_mask = old_page_owner->gfp_mask;
230 new_page_owner->last_migrate_reason =
231 old_page_owner->last_migrate_reason;
232 new_page_owner->handle = old_page_owner->handle;
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700233
234 /*
235 * We don't clear the bit on the oldpage as it's going to be freed
236 * after migration. Until then, the info can be useful in case of
237 * a bug, and the overal stats will be off a bit only temporarily.
238 * Also, migrate_misplaced_transhuge_page() can still fail the
239 * migration and then we want the oldpage to retain the info. But
240 * in that case we also don't need to explicitly clear the info from
241 * the new page, which will be freed.
242 */
243 __set_bit(PAGE_EXT_OWNER, &new_ext->flags);
244}
245
Joonsoo Kime2f612e2016-10-07 16:58:21 -0700246void pagetypeinfo_showmixedcount_print(struct seq_file *m,
247 pg_data_t *pgdat, struct zone *zone)
248{
249 struct page *page;
250 struct page_ext *page_ext;
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700251 struct page_owner *page_owner;
Joonsoo Kime2f612e2016-10-07 16:58:21 -0700252 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
253 unsigned long end_pfn = pfn + zone->spanned_pages;
254 unsigned long count[MIGRATE_TYPES] = { 0, };
255 int pageblock_mt, page_mt;
256 int i;
257
258 /* Scan block by block. First and last block may be incomplete */
259 pfn = zone->zone_start_pfn;
260
261 /*
262 * Walk the zone in pageblock_nr_pages steps. If a page block spans
263 * a zone boundary, it will be double counted between zones. This does
264 * not matter as the mixed block count will still be correct
265 */
266 for (; pfn < end_pfn; ) {
267 if (!pfn_valid(pfn)) {
268 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
269 continue;
270 }
271
272 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
273 block_end_pfn = min(block_end_pfn, end_pfn);
274
275 page = pfn_to_page(pfn);
276 pageblock_mt = get_pageblock_migratetype(page);
277
278 for (; pfn < block_end_pfn; pfn++) {
279 if (!pfn_valid_within(pfn))
280 continue;
281
282 page = pfn_to_page(pfn);
283
284 if (page_zone(page) != zone)
285 continue;
286
287 if (PageBuddy(page)) {
Vinayak Menon64958e22017-06-24 11:48:33 +1000288 unsigned long freepage_order;
289
290 freepage_order = page_order_unsafe(page);
291 if (freepage_order < MAX_ORDER)
292 pfn += (1UL << freepage_order) - 1;
Joonsoo Kime2f612e2016-10-07 16:58:21 -0700293 continue;
294 }
295
296 if (PageReserved(page))
297 continue;
298
299 page_ext = lookup_page_ext(page);
300 if (unlikely(!page_ext))
301 continue;
302
303 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
304 continue;
305
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700306 page_owner = get_page_owner(page_ext);
307 page_mt = gfpflags_to_migratetype(
308 page_owner->gfp_mask);
Joonsoo Kime2f612e2016-10-07 16:58:21 -0700309 if (pageblock_mt != page_mt) {
310 if (is_migrate_cma(pageblock_mt))
311 count[MIGRATE_MOVABLE]++;
312 else
313 count[pageblock_mt]++;
314
315 pfn = block_end_pfn;
316 break;
317 }
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700318 pfn += (1UL << page_owner->order) - 1;
Joonsoo Kime2f612e2016-10-07 16:58:21 -0700319 }
320 }
321
322 /* Print counts */
323 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
324 for (i = 0; i < MIGRATE_TYPES; i++)
325 seq_printf(m, "%12lu ", count[i]);
326 seq_putc(m, '\n');
327}
328
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800329static ssize_t
330print_page_owner(char __user *buf, size_t count, unsigned long pfn,
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700331 struct page *page, struct page_owner *page_owner,
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700332 depot_stack_handle_t handle)
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800333{
334 int ret;
335 int pageblock_mt, page_mt;
336 char *kbuf;
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700337 unsigned long entries[PAGE_OWNER_STACK_DEPTH];
Sergei Rogachev94f759d62015-02-11 15:28:34 -0800338 struct stack_trace trace = {
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700339 .nr_entries = 0,
340 .entries = entries,
341 .max_entries = PAGE_OWNER_STACK_DEPTH,
342 .skip = 0
Sergei Rogachev94f759d62015-02-11 15:28:34 -0800343 };
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800344
345 kbuf = kmalloc(count, GFP_KERNEL);
346 if (!kbuf)
347 return -ENOMEM;
348
349 ret = snprintf(kbuf, count,
Vlastimil Babka60f30352016-03-15 14:56:08 -0700350 "Page allocated via order %u, mask %#x(%pGg)\n",
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700351 page_owner->order, page_owner->gfp_mask,
352 &page_owner->gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800353
354 if (ret >= count)
355 goto err;
356
357 /* Print information relevant to grouping pages by mobility */
Mel Gorman0b423ca2016-05-19 17:14:27 -0700358 pageblock_mt = get_pageblock_migratetype(page);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700359 page_mt = gfpflags_to_migratetype(page_owner->gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800360 ret += snprintf(kbuf + ret, count - ret,
Vlastimil Babka60f30352016-03-15 14:56:08 -0700361 "PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n",
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800362 pfn,
Vlastimil Babka60f30352016-03-15 14:56:08 -0700363 migratetype_names[page_mt],
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800364 pfn >> pageblock_order,
Vlastimil Babka60f30352016-03-15 14:56:08 -0700365 migratetype_names[pageblock_mt],
366 page->flags, &page->flags);
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800367
368 if (ret >= count)
369 goto err;
370
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700371 depot_fetch_stack(handle, &trace);
Sergei Rogachev94f759d62015-02-11 15:28:34 -0800372 ret += snprint_stack_trace(kbuf + ret, count - ret, &trace, 0);
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800373 if (ret >= count)
374 goto err;
375
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700376 if (page_owner->last_migrate_reason != -1) {
Vlastimil Babka7cd12b42016-03-15 14:56:18 -0700377 ret += snprintf(kbuf + ret, count - ret,
378 "Page has been migrated, last migrate reason: %s\n",
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700379 migrate_reason_names[page_owner->last_migrate_reason]);
Vlastimil Babka7cd12b42016-03-15 14:56:18 -0700380 if (ret >= count)
381 goto err;
382 }
383
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800384 ret += snprintf(kbuf + ret, count - ret, "\n");
385 if (ret >= count)
386 goto err;
387
388 if (copy_to_user(buf, kbuf, ret))
389 ret = -EFAULT;
390
391 kfree(kbuf);
392 return ret;
393
394err:
395 kfree(kbuf);
396 return -ENOMEM;
397}
398
Vlastimil Babka4e462112016-03-15 14:56:21 -0700399void __dump_page_owner(struct page *page)
400{
401 struct page_ext *page_ext = lookup_page_ext(page);
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700402 struct page_owner *page_owner;
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700403 unsigned long entries[PAGE_OWNER_STACK_DEPTH];
Vlastimil Babka4e462112016-03-15 14:56:21 -0700404 struct stack_trace trace = {
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700405 .nr_entries = 0,
406 .entries = entries,
407 .max_entries = PAGE_OWNER_STACK_DEPTH,
408 .skip = 0
Vlastimil Babka4e462112016-03-15 14:56:21 -0700409 };
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700410 depot_stack_handle_t handle;
Sudip Mukherjee82850272016-06-24 14:50:24 -0700411 gfp_t gfp_mask;
412 int mt;
Vlastimil Babka4e462112016-03-15 14:56:21 -0700413
Yang Shif86e4272016-06-03 14:55:38 -0700414 if (unlikely(!page_ext)) {
415 pr_alert("There is not page extension available.\n");
416 return;
417 }
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700418
419 page_owner = get_page_owner(page_ext);
420 gfp_mask = page_owner->gfp_mask;
Sudip Mukherjee82850272016-06-24 14:50:24 -0700421 mt = gfpflags_to_migratetype(gfp_mask);
Yang Shif86e4272016-06-03 14:55:38 -0700422
Vlastimil Babka4e462112016-03-15 14:56:21 -0700423 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) {
424 pr_alert("page_owner info is not active (free page?)\n");
425 return;
426 }
427
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700428 handle = READ_ONCE(page_owner->handle);
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700429 if (!handle) {
430 pr_alert("page_owner info is not active (free page?)\n");
431 return;
432 }
433
434 depot_fetch_stack(handle, &trace);
Joe Perches756a025f02016-03-17 14:19:47 -0700435 pr_alert("page allocated via order %u, migratetype %s, gfp_mask %#x(%pGg)\n",
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700436 page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700437 print_stack_trace(&trace, 0);
438
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700439 if (page_owner->last_migrate_reason != -1)
Vlastimil Babka4e462112016-03-15 14:56:21 -0700440 pr_alert("page has been migrated, last migrate reason: %s\n",
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700441 migrate_reason_names[page_owner->last_migrate_reason]);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700442}
443
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800444static ssize_t
445read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos)
446{
447 unsigned long pfn;
448 struct page *page;
449 struct page_ext *page_ext;
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700450 struct page_owner *page_owner;
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700451 depot_stack_handle_t handle;
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800452
Vlastimil Babka7dd80b82016-03-15 14:56:12 -0700453 if (!static_branch_unlikely(&page_owner_inited))
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800454 return -EINVAL;
455
456 page = NULL;
457 pfn = min_low_pfn + *ppos;
458
459 /* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
460 while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0)
461 pfn++;
462
463 drain_all_pages(NULL);
464
465 /* Find an allocated page */
466 for (; pfn < max_pfn; pfn++) {
467 /*
468 * If the new page is in a new MAX_ORDER_NR_PAGES area,
469 * validate the area as existing, skip it if not
470 */
471 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) {
472 pfn += MAX_ORDER_NR_PAGES - 1;
473 continue;
474 }
475
476 /* Check for holes within a MAX_ORDER area */
477 if (!pfn_valid_within(pfn))
478 continue;
479
480 page = pfn_to_page(pfn);
481 if (PageBuddy(page)) {
482 unsigned long freepage_order = page_order_unsafe(page);
483
484 if (freepage_order < MAX_ORDER)
485 pfn += (1UL << freepage_order) - 1;
486 continue;
487 }
488
489 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700490 if (unlikely(!page_ext))
491 continue;
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800492
493 /*
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -0800494 * Some pages could be missed by concurrent allocation or free,
495 * because we don't hold the zone lock.
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800496 */
497 if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
498 continue;
499
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700500 page_owner = get_page_owner(page_ext);
501
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700502 /*
503 * Access to page_ext->handle isn't synchronous so we should
504 * be careful to access it.
505 */
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700506 handle = READ_ONCE(page_owner->handle);
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700507 if (!handle)
508 continue;
509
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800510 /* Record the next PFN to read in the file offset */
511 *ppos = (pfn - min_low_pfn) + 1;
512
Joonsoo Kimf2ca0b52016-07-26 15:23:55 -0700513 return print_page_owner(buf, count, pfn, page,
Joonsoo Kim9300d8d2016-10-07 16:58:30 -0700514 page_owner, handle);
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800515 }
516
517 return 0;
518}
519
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -0800520static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone)
521{
522 struct page *page;
523 struct page_ext *page_ext;
524 unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
525 unsigned long end_pfn = pfn + zone->spanned_pages;
526 unsigned long count = 0;
527
528 /* Scan block by block. First and last block may be incomplete */
529 pfn = zone->zone_start_pfn;
530
531 /*
532 * Walk the zone in pageblock_nr_pages steps. If a page block spans
533 * a zone boundary, it will be double counted between zones. This does
534 * not matter as the mixed block count will still be correct
535 */
536 for (; pfn < end_pfn; ) {
537 if (!pfn_valid(pfn)) {
538 pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
539 continue;
540 }
541
542 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
543 block_end_pfn = min(block_end_pfn, end_pfn);
544
545 page = pfn_to_page(pfn);
546
547 for (; pfn < block_end_pfn; pfn++) {
548 if (!pfn_valid_within(pfn))
549 continue;
550
551 page = pfn_to_page(pfn);
552
Joonsoo Kim9d43f5a2016-05-19 17:12:13 -0700553 if (page_zone(page) != zone)
554 continue;
555
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -0800556 /*
557 * We are safe to check buddy flag and order, because
558 * this is init stage and only single thread runs.
559 */
560 if (PageBuddy(page)) {
561 pfn += (1UL << page_order(page)) - 1;
562 continue;
563 }
564
565 if (PageReserved(page))
566 continue;
567
568 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700569 if (unlikely(!page_ext))
570 continue;
Joonsoo Kim61cf5fe2014-12-12 16:56:04 -0800571
572 /* Maybe overraping zone */
573 if (test_bit(PAGE_EXT_OWNER, &page_ext->flags))
574 continue;
575
576 /* Found early allocated page */
577 set_page_owner(page, 0, 0);
578 count++;
579 }
580 }
581
582 pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
583 pgdat->node_id, zone->name, count);
584}
585
586static void init_zones_in_node(pg_data_t *pgdat)
587{
588 struct zone *zone;
589 struct zone *node_zones = pgdat->node_zones;
590 unsigned long flags;
591
592 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
593 if (!populated_zone(zone))
594 continue;
595
596 spin_lock_irqsave(&zone->lock, flags);
597 init_pages_in_zone(pgdat, zone);
598 spin_unlock_irqrestore(&zone->lock, flags);
599 }
600}
601
602static void init_early_allocated_pages(void)
603{
604 pg_data_t *pgdat;
605
606 drain_all_pages(NULL);
607 for_each_online_pgdat(pgdat)
608 init_zones_in_node(pgdat);
609}
610
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800611static const struct file_operations proc_page_owner_operations = {
612 .read = read_page_owner,
613};
614
615static int __init pageowner_init(void)
616{
617 struct dentry *dentry;
618
Vlastimil Babka7dd80b82016-03-15 14:56:12 -0700619 if (!static_branch_unlikely(&page_owner_inited)) {
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800620 pr_info("page_owner is disabled\n");
621 return 0;
622 }
623
624 dentry = debugfs_create_file("page_owner", S_IRUSR, NULL,
625 NULL, &proc_page_owner_operations);
626 if (IS_ERR(dentry))
627 return PTR_ERR(dentry);
628
629 return 0;
630}
Paul Gortmaker44c5af92015-05-01 21:57:34 -0400631late_initcall(pageowner_init)