xref: /OK3568_Linux_fs/kernel/mm/page_owner.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/debugfs.h>
3 #include <linux/mm.h>
4 #include <linux/slab.h>
5 #include <linux/uaccess.h>
6 #include <linux/memblock.h>
7 #include <linux/stacktrace.h>
8 #include <linux/page_owner.h>
9 #include <linux/jump_label.h>
10 #include <linux/migrate.h>
11 #include <linux/stackdepot.h>
12 #include <linux/seq_file.h>
13 #include <linux/sched/clock.h>
14 
15 #include "internal.h"
16 
17 /*
18  * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
19  * to use off stack temporal storage
20  */
21 #define PAGE_OWNER_STACK_DEPTH (16)
22 
23 struct page_owner {
24 	unsigned short order;
25 	short last_migrate_reason;
26 	gfp_t gfp_mask;
27 	depot_stack_handle_t handle;
28 	depot_stack_handle_t free_handle;
29 	u64 ts_nsec;
30 	u64 free_ts_nsec;
31 	pid_t pid;
32 };
33 
34 bool page_owner_enabled;
35 DEFINE_STATIC_KEY_FALSE(page_owner_inited);
36 
37 static depot_stack_handle_t dummy_handle;
38 static depot_stack_handle_t failure_handle;
39 static depot_stack_handle_t early_handle;
40 
41 static void init_early_allocated_pages(void);
42 
early_page_owner_param(char * buf)43 static int __init early_page_owner_param(char *buf)
44 {
45 	if (!buf)
46 		return -EINVAL;
47 
48 	if (strcmp(buf, "on") == 0)
49 		page_owner_enabled = true;
50 
51 	return 0;
52 }
53 early_param("page_owner", early_page_owner_param);
54 
need_page_owner(void)55 static bool need_page_owner(void)
56 {
57 	return page_owner_enabled;
58 }
59 
create_dummy_stack(void)60 static __always_inline depot_stack_handle_t create_dummy_stack(void)
61 {
62 	unsigned long entries[4];
63 	unsigned int nr_entries;
64 
65 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
66 	return stack_depot_save(entries, nr_entries, GFP_KERNEL);
67 }
68 
register_dummy_stack(void)69 static noinline void register_dummy_stack(void)
70 {
71 	dummy_handle = create_dummy_stack();
72 }
73 
register_failure_stack(void)74 static noinline void register_failure_stack(void)
75 {
76 	failure_handle = create_dummy_stack();
77 }
78 
register_early_stack(void)79 static noinline void register_early_stack(void)
80 {
81 	early_handle = create_dummy_stack();
82 }
83 
init_page_owner(void)84 static void init_page_owner(void)
85 {
86 	if (!page_owner_enabled)
87 		return;
88 
89 	register_dummy_stack();
90 	register_failure_stack();
91 	register_early_stack();
92 	static_branch_enable(&page_owner_inited);
93 	init_early_allocated_pages();
94 }
95 
96 struct page_ext_operations page_owner_ops = {
97 	.size = sizeof(struct page_owner),
98 	.need = need_page_owner,
99 	.init = init_page_owner,
100 };
101 
get_page_owner(struct page_ext * page_ext)102 struct page_owner *get_page_owner(struct page_ext *page_ext)
103 {
104 	return (void *)page_ext + page_owner_ops.offset;
105 }
106 EXPORT_SYMBOL_GPL(get_page_owner);
107 
get_page_owner_handle(struct page_ext * page_ext,unsigned long pfn)108 depot_stack_handle_t get_page_owner_handle(struct page_ext *page_ext, unsigned long pfn)
109 {
110 	struct page_owner *page_owner;
111 	depot_stack_handle_t handle;
112 
113 	if (!page_owner_enabled)
114 		return 0;
115 
116 	page_owner = get_page_owner(page_ext);
117 
118 	/* skip handle for tail pages of higher order allocations */
119 	if (!IS_ALIGNED(pfn, 1 << page_owner->order))
120 		return 0;
121 
122 	handle = READ_ONCE(page_owner->handle);
123 	return handle;
124 }
125 EXPORT_SYMBOL_GPL(get_page_owner_handle);
126 
check_recursive_alloc(unsigned long * entries,unsigned int nr_entries,unsigned long ip)127 static inline bool check_recursive_alloc(unsigned long *entries,
128 					 unsigned int nr_entries,
129 					 unsigned long ip)
130 {
131 	unsigned int i;
132 
133 	for (i = 0; i < nr_entries; i++) {
134 		if (entries[i] == ip)
135 			return true;
136 	}
137 	return false;
138 }
139 
save_stack(gfp_t flags)140 static noinline depot_stack_handle_t save_stack(gfp_t flags)
141 {
142 	unsigned long entries[PAGE_OWNER_STACK_DEPTH];
143 	depot_stack_handle_t handle;
144 	unsigned int nr_entries;
145 
146 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
147 
148 	/*
149 	 * We need to check recursion here because our request to
150 	 * stackdepot could trigger memory allocation to save new
151 	 * entry. New memory allocation would reach here and call
152 	 * stack_depot_save_entries() again if we don't catch it. There is
153 	 * still not enough memory in stackdepot so it would try to
154 	 * allocate memory again and loop forever.
155 	 */
156 	if (check_recursive_alloc(entries, nr_entries, _RET_IP_))
157 		return dummy_handle;
158 
159 	handle = stack_depot_save(entries, nr_entries, flags);
160 	if (!handle)
161 		handle = failure_handle;
162 
163 	return handle;
164 }
165 
__reset_page_owner(struct page * page,unsigned int order)166 void __reset_page_owner(struct page *page, unsigned int order)
167 {
168 	int i;
169 	struct page_ext *page_ext;
170 	depot_stack_handle_t handle = 0;
171 	struct page_owner *page_owner;
172 	u64 free_ts_nsec = local_clock();
173 
174 	handle = save_stack(GFP_NOWAIT | __GFP_NOWARN);
175 
176 	page_ext = page_ext_get(page);
177 	if (unlikely(!page_ext))
178 		return;
179 	for (i = 0; i < (1 << order); i++) {
180 		__clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
181 		page_owner = get_page_owner(page_ext);
182 		page_owner->free_handle = handle;
183 		page_owner->free_ts_nsec = free_ts_nsec;
184 		page_ext = page_ext_next(page_ext);
185 	}
186 	page_ext_put(page_ext);
187 }
188 
__set_page_owner_handle(struct page * page,struct page_ext * page_ext,depot_stack_handle_t handle,unsigned int order,gfp_t gfp_mask)189 static inline void __set_page_owner_handle(struct page *page,
190 	struct page_ext *page_ext, depot_stack_handle_t handle,
191 	unsigned int order, gfp_t gfp_mask)
192 {
193 	struct page_owner *page_owner;
194 	int i;
195 
196 	for (i = 0; i < (1 << order); i++) {
197 		page_owner = get_page_owner(page_ext);
198 		page_owner->handle = handle;
199 		page_owner->order = order;
200 		page_owner->gfp_mask = gfp_mask;
201 		page_owner->last_migrate_reason = -1;
202 		page_owner->pid = current->pid;
203 		page_owner->ts_nsec = local_clock();
204 		__set_bit(PAGE_EXT_OWNER, &page_ext->flags);
205 		__set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
206 
207 		page_ext = page_ext_next(page_ext);
208 	}
209 }
210 
__set_page_owner(struct page * page,unsigned int order,gfp_t gfp_mask)211 noinline void __set_page_owner(struct page *page, unsigned int order,
212 					gfp_t gfp_mask)
213 {
214 	struct page_ext *page_ext;
215 	depot_stack_handle_t handle;
216 
217 	handle = save_stack(gfp_mask);
218 
219 	page_ext = page_ext_get(page);
220 	if (unlikely(!page_ext))
221 		return;
222 	__set_page_owner_handle(page, page_ext, handle, order, gfp_mask);
223 	page_ext_put(page_ext);
224 }
225 
__set_page_owner_migrate_reason(struct page * page,int reason)226 void __set_page_owner_migrate_reason(struct page *page, int reason)
227 {
228 	struct page_ext *page_ext = page_ext_get(page);
229 	struct page_owner *page_owner;
230 
231 	if (unlikely(!page_ext))
232 		return;
233 
234 	page_owner = get_page_owner(page_ext);
235 	page_owner->last_migrate_reason = reason;
236 	page_ext_put(page_ext);
237 }
238 
__split_page_owner(struct page * page,unsigned int nr)239 void __split_page_owner(struct page *page, unsigned int nr)
240 {
241 	int i;
242 	struct page_ext *page_ext = page_ext_get(page);
243 	struct page_owner *page_owner;
244 
245 	if (unlikely(!page_ext))
246 		return;
247 
248 	for (i = 0; i < nr; i++) {
249 		page_owner = get_page_owner(page_ext);
250 		page_owner->order = 0;
251 		page_ext = page_ext_next(page_ext);
252 	}
253 	page_ext_put(page_ext);
254 }
255 
__copy_page_owner(struct page * oldpage,struct page * newpage)256 void __copy_page_owner(struct page *oldpage, struct page *newpage)
257 {
258 	struct page_ext *old_ext;
259 	struct page_ext *new_ext;
260 	struct page_owner *old_page_owner, *new_page_owner;
261 
262 	old_ext = page_ext_get(oldpage);
263 	if (unlikely(!old_ext))
264 		return;
265 
266 	new_ext = page_ext_get(newpage);
267 	if (unlikely(!new_ext)) {
268 		page_ext_put(old_ext);
269 		return;
270 	}
271 
272 	old_page_owner = get_page_owner(old_ext);
273 	new_page_owner = get_page_owner(new_ext);
274 	new_page_owner->order = old_page_owner->order;
275 	new_page_owner->gfp_mask = old_page_owner->gfp_mask;
276 	new_page_owner->last_migrate_reason =
277 		old_page_owner->last_migrate_reason;
278 	new_page_owner->handle = old_page_owner->handle;
279 	new_page_owner->pid = old_page_owner->pid;
280 	new_page_owner->ts_nsec = old_page_owner->ts_nsec;
281 	new_page_owner->free_ts_nsec = old_page_owner->ts_nsec;
282 
283 	/*
284 	 * We don't clear the bit on the oldpage as it's going to be freed
285 	 * after migration. Until then, the info can be useful in case of
286 	 * a bug, and the overal stats will be off a bit only temporarily.
287 	 * Also, migrate_misplaced_transhuge_page() can still fail the
288 	 * migration and then we want the oldpage to retain the info. But
289 	 * in that case we also don't need to explicitly clear the info from
290 	 * the new page, which will be freed.
291 	 */
292 	__set_bit(PAGE_EXT_OWNER, &new_ext->flags);
293 	__set_bit(PAGE_EXT_OWNER_ALLOCATED, &new_ext->flags);
294 	page_ext_put(new_ext);
295 	page_ext_put(old_ext);
296 }
297 
pagetypeinfo_showmixedcount_print(struct seq_file * m,pg_data_t * pgdat,struct zone * zone)298 void pagetypeinfo_showmixedcount_print(struct seq_file *m,
299 				       pg_data_t *pgdat, struct zone *zone)
300 {
301 	struct page *page;
302 	struct page_ext *page_ext;
303 	struct page_owner *page_owner;
304 	unsigned long pfn = zone->zone_start_pfn, block_end_pfn;
305 	unsigned long end_pfn = pfn + zone->spanned_pages;
306 	unsigned long count[MIGRATE_TYPES] = { 0, };
307 	int pageblock_mt, page_mt;
308 	int i;
309 
310 	/* Scan block by block. First and last block may be incomplete */
311 	pfn = zone->zone_start_pfn;
312 
313 	/*
314 	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
315 	 * a zone boundary, it will be double counted between zones. This does
316 	 * not matter as the mixed block count will still be correct
317 	 */
318 	for (; pfn < end_pfn; ) {
319 		page = pfn_to_online_page(pfn);
320 		if (!page) {
321 			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
322 			continue;
323 		}
324 
325 		block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
326 		block_end_pfn = min(block_end_pfn, end_pfn);
327 
328 		pageblock_mt = get_pageblock_migratetype(page);
329 
330 		for (; pfn < block_end_pfn; pfn++) {
331 			if (!pfn_valid_within(pfn))
332 				continue;
333 
334 			/* The pageblock is online, no need to recheck. */
335 			page = pfn_to_page(pfn);
336 
337 			if (page_zone(page) != zone)
338 				continue;
339 
340 			if (PageBuddy(page)) {
341 				unsigned long freepage_order;
342 
343 				freepage_order = buddy_order_unsafe(page);
344 				if (freepage_order < MAX_ORDER)
345 					pfn += (1UL << freepage_order) - 1;
346 				continue;
347 			}
348 
349 			if (PageReserved(page))
350 				continue;
351 
352 			page_ext = page_ext_get(page);
353 			if (unlikely(!page_ext))
354 				continue;
355 
356 			if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
357 				goto ext_put_continue;
358 
359 			page_owner = get_page_owner(page_ext);
360 			page_mt = gfp_migratetype(page_owner->gfp_mask);
361 			if (pageblock_mt != page_mt) {
362 				if (is_migrate_cma(pageblock_mt))
363 					count[MIGRATE_MOVABLE]++;
364 				else
365 					count[pageblock_mt]++;
366 
367 				pfn = block_end_pfn;
368 				page_ext_put(page_ext);
369 				break;
370 			}
371 			pfn += (1UL << page_owner->order) - 1;
372 ext_put_continue:
373 			page_ext_put(page_ext);
374 		}
375 	}
376 
377 	/* Print counts */
378 	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
379 	for (i = 0; i < MIGRATE_TYPES; i++)
380 		seq_printf(m, "%12lu ", count[i]);
381 	seq_putc(m, '\n');
382 }
383 
384 static ssize_t
print_page_owner(char __user * buf,size_t count,unsigned long pfn,struct page * page,struct page_owner * page_owner,depot_stack_handle_t handle)385 print_page_owner(char __user *buf, size_t count, unsigned long pfn,
386 		struct page *page, struct page_owner *page_owner,
387 		depot_stack_handle_t handle)
388 {
389 	int ret, pageblock_mt, page_mt;
390 	unsigned long *entries;
391 	unsigned int nr_entries;
392 	char *kbuf;
393 
394 	count = min_t(size_t, count, PAGE_SIZE);
395 	kbuf = kmalloc(count, GFP_KERNEL);
396 	if (!kbuf)
397 		return -ENOMEM;
398 
399 	ret = snprintf(kbuf, count,
400 			"Page allocated via order %u, mask %#x(%pGg), pid %d, ts %llu ns, free_ts %llu ns\n",
401 			page_owner->order, page_owner->gfp_mask,
402 			&page_owner->gfp_mask, page_owner->pid,
403 			page_owner->ts_nsec, page_owner->free_ts_nsec);
404 
405 	if (ret >= count)
406 		goto err;
407 
408 	/* Print information relevant to grouping pages by mobility */
409 	pageblock_mt = get_pageblock_migratetype(page);
410 	page_mt  = gfp_migratetype(page_owner->gfp_mask);
411 	ret += snprintf(kbuf + ret, count - ret,
412 			"PFN %lu type %s Block %lu type %s Flags %#lx(%pGp)\n",
413 			pfn,
414 			migratetype_names[page_mt],
415 			pfn >> pageblock_order,
416 			migratetype_names[pageblock_mt],
417 			page->flags, &page->flags);
418 
419 	if (ret >= count)
420 		goto err;
421 
422 	nr_entries = stack_depot_fetch(handle, &entries);
423 	ret += stack_trace_snprint(kbuf + ret, count - ret, entries, nr_entries, 0);
424 	if (ret >= count)
425 		goto err;
426 
427 	if (page_owner->last_migrate_reason != -1) {
428 		ret += snprintf(kbuf + ret, count - ret,
429 			"Page has been migrated, last migrate reason: %s\n",
430 			migrate_reason_names[page_owner->last_migrate_reason]);
431 		if (ret >= count)
432 			goto err;
433 	}
434 
435 	ret += snprintf(kbuf + ret, count - ret, "\n");
436 	if (ret >= count)
437 		goto err;
438 
439 	if (copy_to_user(buf, kbuf, ret))
440 		ret = -EFAULT;
441 
442 	kfree(kbuf);
443 	return ret;
444 
445 err:
446 	kfree(kbuf);
447 	return -ENOMEM;
448 }
449 
__dump_page_owner(struct page * page)450 void __dump_page_owner(struct page *page)
451 {
452 	struct page_ext *page_ext = page_ext_get((void *)page);
453 	struct page_owner *page_owner;
454 	depot_stack_handle_t handle;
455 	unsigned long *entries;
456 	unsigned int nr_entries;
457 	gfp_t gfp_mask;
458 	int mt;
459 
460 	if (unlikely(!page_ext)) {
461 		pr_alert("There is not page extension available.\n");
462 		return;
463 	}
464 
465 	page_owner = get_page_owner(page_ext);
466 	gfp_mask = page_owner->gfp_mask;
467 	mt = gfp_migratetype(gfp_mask);
468 
469 	if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) {
470 		pr_alert("page_owner info is not present (never set?)\n");
471 		page_ext_put(page_ext);
472 		return;
473 	}
474 
475 	if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
476 		pr_alert("page_owner tracks the page as allocated\n");
477 	else
478 		pr_alert("page_owner tracks the page as freed\n");
479 
480 	pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, ts %llu, free_ts %llu\n",
481 		 page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask,
482 		 page_owner->pid, page_owner->ts_nsec, page_owner->free_ts_nsec);
483 
484 	handle = READ_ONCE(page_owner->handle);
485 	if (!handle) {
486 		pr_alert("page_owner allocation stack trace missing\n");
487 	} else {
488 		nr_entries = stack_depot_fetch(handle, &entries);
489 		stack_trace_print(entries, nr_entries, 0);
490 	}
491 
492 	handle = READ_ONCE(page_owner->free_handle);
493 	if (!handle) {
494 		pr_alert("page_owner free stack trace missing\n");
495 	} else {
496 		nr_entries = stack_depot_fetch(handle, &entries);
497 		pr_alert("page last free stack trace:\n");
498 		stack_trace_print(entries, nr_entries, 0);
499 	}
500 
501 	if (page_owner->last_migrate_reason != -1)
502 		pr_alert("page has been migrated, last migrate reason: %s\n",
503 			migrate_reason_names[page_owner->last_migrate_reason]);
504 	page_ext_put(page_ext);
505 }
506 
507 static ssize_t
read_page_owner(struct file * file,char __user * buf,size_t count,loff_t * ppos)508 read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos)
509 {
510 	unsigned long pfn;
511 	struct page *page;
512 	struct page_ext *page_ext;
513 	struct page_owner *page_owner;
514 	depot_stack_handle_t handle;
515 
516 	if (!static_branch_unlikely(&page_owner_inited))
517 		return -EINVAL;
518 
519 	page = NULL;
520 	pfn = min_low_pfn + *ppos;
521 
522 	/* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
523 	while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0)
524 		pfn++;
525 
526 	drain_all_pages(NULL);
527 
528 	/* Find an allocated page */
529 	for (; pfn < max_pfn; pfn++) {
530 		/*
531 		 * This temporary page_owner is required so
532 		 * that we can avoid the context switches while holding
533 		 * the rcu lock and copying the page owner information to
534 		 * user through copy_to_user() or GFP_KERNEL allocations.
535 		 */
536 		struct page_owner page_owner_tmp;
537 
538 		/*
539 		 * If the new page is in a new MAX_ORDER_NR_PAGES area,
540 		 * validate the area as existing, skip it if not
541 		 */
542 		if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) {
543 			pfn += MAX_ORDER_NR_PAGES - 1;
544 			continue;
545 		}
546 
547 		/* Check for holes within a MAX_ORDER area */
548 		if (!pfn_valid_within(pfn))
549 			continue;
550 
551 		page = pfn_to_page(pfn);
552 		if (PageBuddy(page)) {
553 			unsigned long freepage_order = buddy_order_unsafe(page);
554 
555 			if (freepage_order < MAX_ORDER)
556 				pfn += (1UL << freepage_order) - 1;
557 			continue;
558 		}
559 
560 		page_ext = page_ext_get(page);
561 		if (unlikely(!page_ext))
562 			continue;
563 
564 		/*
565 		 * Some pages could be missed by concurrent allocation or free,
566 		 * because we don't hold the zone lock.
567 		 */
568 		if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
569 			goto ext_put_continue;
570 
571 		/*
572 		 * Although we do have the info about past allocation of free
573 		 * pages, it's not relevant for current memory usage.
574 		 */
575 		if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
576 			goto ext_put_continue;
577 
578 		page_owner = get_page_owner(page_ext);
579 
580 		/*
581 		 * Don't print "tail" pages of high-order allocations as that
582 		 * would inflate the stats.
583 		 */
584 		if (!IS_ALIGNED(pfn, 1 << page_owner->order))
585 			goto ext_put_continue;
586 
587 		/*
588 		 * Access to page_ext->handle isn't synchronous so we should
589 		 * be careful to access it.
590 		 */
591 		handle = READ_ONCE(page_owner->handle);
592 		if (!handle)
593 			goto ext_put_continue;
594 
595 		/* Record the next PFN to read in the file offset */
596 		*ppos = (pfn - min_low_pfn) + 1;
597 
598 		page_owner_tmp = *page_owner;
599 		page_ext_put(page_ext);
600 		return print_page_owner(buf, count, pfn, page,
601 				&page_owner_tmp, handle);
602 ext_put_continue:
603 		page_ext_put(page_ext);
604 	}
605 
606 	return 0;
607 }
608 
init_pages_in_zone(pg_data_t * pgdat,struct zone * zone)609 static void init_pages_in_zone(pg_data_t *pgdat, struct zone *zone)
610 {
611 	unsigned long pfn = zone->zone_start_pfn;
612 	unsigned long end_pfn = zone_end_pfn(zone);
613 	unsigned long count = 0;
614 
615 	/*
616 	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
617 	 * a zone boundary, it will be double counted between zones. This does
618 	 * not matter as the mixed block count will still be correct
619 	 */
620 	for (; pfn < end_pfn; ) {
621 		unsigned long block_end_pfn;
622 
623 		if (!pfn_valid(pfn)) {
624 			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
625 			continue;
626 		}
627 
628 		block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
629 		block_end_pfn = min(block_end_pfn, end_pfn);
630 
631 		for (; pfn < block_end_pfn; pfn++) {
632 			struct page *page;
633 			struct page_ext *page_ext;
634 
635 			if (!pfn_valid_within(pfn))
636 				continue;
637 
638 			page = pfn_to_page(pfn);
639 
640 			if (page_zone(page) != zone)
641 				continue;
642 
643 			/*
644 			 * To avoid having to grab zone->lock, be a little
645 			 * careful when reading buddy page order. The only
646 			 * danger is that we skip too much and potentially miss
647 			 * some early allocated pages, which is better than
648 			 * heavy lock contention.
649 			 */
650 			if (PageBuddy(page)) {
651 				unsigned long order = buddy_order_unsafe(page);
652 
653 				if (order > 0 && order < MAX_ORDER)
654 					pfn += (1UL << order) - 1;
655 				continue;
656 			}
657 
658 			if (PageReserved(page))
659 				continue;
660 
661 			page_ext = page_ext_get(page);
662 			if (unlikely(!page_ext))
663 				continue;
664 
665 			/* Maybe overlapping zone */
666 			if (test_bit(PAGE_EXT_OWNER, &page_ext->flags))
667 				goto ext_put_continue;
668 
669 			/* Found early allocated page */
670 			__set_page_owner_handle(page, page_ext, early_handle,
671 						0, 0);
672 			count++;
673 ext_put_continue:
674 			page_ext_put(page_ext);
675 		}
676 		cond_resched();
677 	}
678 
679 	pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
680 		pgdat->node_id, zone->name, count);
681 }
682 
init_zones_in_node(pg_data_t * pgdat)683 static void init_zones_in_node(pg_data_t *pgdat)
684 {
685 	struct zone *zone;
686 	struct zone *node_zones = pgdat->node_zones;
687 
688 	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
689 		if (!populated_zone(zone))
690 			continue;
691 
692 		init_pages_in_zone(pgdat, zone);
693 	}
694 }
695 
init_early_allocated_pages(void)696 static void init_early_allocated_pages(void)
697 {
698 	pg_data_t *pgdat;
699 
700 	for_each_online_pgdat(pgdat)
701 		init_zones_in_node(pgdat);
702 }
703 
704 static const struct file_operations proc_page_owner_operations = {
705 	.read		= read_page_owner,
706 };
707 
pageowner_init(void)708 static int __init pageowner_init(void)
709 {
710 	if (!static_branch_unlikely(&page_owner_inited)) {
711 		pr_info("page_owner is disabled\n");
712 		return 0;
713 	}
714 
715 	debugfs_create_file("page_owner", 0400, NULL, NULL,
716 			    &proc_page_owner_operations);
717 
718 	return 0;
719 }
720 late_initcall(pageowner_init)
721