page_owner.c

   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/memcontrol.h>
  14#include <linux/sched/clock.h>
  15
  16#include "internal.h"
  17
  18/*
  19 * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
  20 * to use off stack temporal storage
  21 */
  22#define PAGE_OWNER_STACK_DEPTH (16)
  23
  24struct page_owner {
  25	unsigned short order;
  26	short last_migrate_reason;
  27	gfp_t gfp_mask;
  28	depot_stack_handle_t handle;
  29	depot_stack_handle_t free_handle;
  30	u64 ts_nsec;
  31	u64 free_ts_nsec;
  32	char comm[TASK_COMM_LEN];
  33	pid_t pid;
  34	pid_t tgid;
  35	pid_t free_pid;
  36	pid_t free_tgid;
  37};
  38
  39struct stack {
  40	struct stack_record *stack_record;
  41	struct stack *next;
  42};
  43static struct stack dummy_stack;
  44static struct stack failure_stack;
  45static struct stack *stack_list;
  46static DEFINE_SPINLOCK(stack_list_lock);
  47
  48#define STACK_PRINT_FLAG_STACK		0x1
  49#define STACK_PRINT_FLAG_PAGES		0x2
  50#define STACK_PRINT_FLAG_HANDLE		0x4
  51
  52struct stack_print_ctx {
  53	struct stack *stack;
  54	u8 flags;
  55};
  56
  57static bool page_owner_enabled __initdata;
  58DEFINE_STATIC_KEY_FALSE(page_owner_inited);
  59
  60static depot_stack_handle_t dummy_handle;
  61static depot_stack_handle_t failure_handle;
  62static depot_stack_handle_t early_handle;
  63
  64static void init_early_allocated_pages(void);
  65
  66static inline void set_current_in_page_owner(void)
  67{
  68	/*
  69	 * Avoid recursion.
  70	 *
  71	 * We might need to allocate more memory from page_owner code, so make
  72	 * sure to signal it in order to avoid recursion.
  73	 */
  74	current->in_page_owner = 1;
  75}
  76
  77static inline void unset_current_in_page_owner(void)
  78{
  79	current->in_page_owner = 0;
  80}
  81
  82static int __init early_page_owner_param(char *buf)
  83{
  84	int ret = kstrtobool(buf, &page_owner_enabled);
  85
  86	if (page_owner_enabled)
  87		stack_depot_request_early_init();
  88
  89	return ret;
  90}
  91early_param("page_owner", early_page_owner_param);
  92
  93static __init bool need_page_owner(void)
  94{
  95	return page_owner_enabled;
  96}
  97
  98static __always_inline depot_stack_handle_t create_dummy_stack(void)
  99{
 100	unsigned long entries[4];
 101	unsigned int nr_entries;
 102
 103	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
 104	return stack_depot_save(entries, nr_entries, GFP_KERNEL);
 105}
 106
 107static noinline void register_dummy_stack(void)
 108{
 109	dummy_handle = create_dummy_stack();
 110}
 111
 112static noinline void register_failure_stack(void)
 113{
 114	failure_handle = create_dummy_stack();
 115}
 116
 117static noinline void register_early_stack(void)
 118{
 119	early_handle = create_dummy_stack();
 120}
 121
 122static __init void init_page_owner(void)
 123{
 124	if (!page_owner_enabled)
 125		return;
 126
 127	register_dummy_stack();
 128	register_failure_stack();
 129	register_early_stack();
 130	init_early_allocated_pages();
 131	/* Initialize dummy and failure stacks and link them to stack_list */
 132	dummy_stack.stack_record = __stack_depot_get_stack_record(dummy_handle);
 133	failure_stack.stack_record = __stack_depot_get_stack_record(failure_handle);
 134	if (dummy_stack.stack_record)
 135		refcount_set(&dummy_stack.stack_record->count, 1);
 136	if (failure_stack.stack_record)
 137		refcount_set(&failure_stack.stack_record->count, 1);
 138	dummy_stack.next = &failure_stack;
 139	stack_list = &dummy_stack;
 140	static_branch_enable(&page_owner_inited);
 141}
 142
 143struct page_ext_operations page_owner_ops = {
 144	.size = sizeof(struct page_owner),
 145	.need = need_page_owner,
 146	.init = init_page_owner,
 147	.need_shared_flags = true,
 148};
 149
 150static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
 151{
 152	return page_ext_data(page_ext, &page_owner_ops);
 153}
 154
 155static noinline depot_stack_handle_t save_stack(gfp_t flags)
 156{
 157	unsigned long entries[PAGE_OWNER_STACK_DEPTH];
 158	depot_stack_handle_t handle;
 159	unsigned int nr_entries;
 160
 161	if (current->in_page_owner)
 162		return dummy_handle;
 163
 164	set_current_in_page_owner();
 165	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
 166	handle = stack_depot_save(entries, nr_entries, flags);
 167	if (!handle)
 168		handle = failure_handle;
 169	unset_current_in_page_owner();
 170
 171	return handle;
 172}
 173
 174static void add_stack_record_to_list(struct stack_record *stack_record,
 175				     gfp_t gfp_mask)
 176{
 177	unsigned long flags;
 178	struct stack *stack;
 179
 180	if (!gfpflags_allow_spinning(gfp_mask))
 181		return;
 182
 183	set_current_in_page_owner();
 184	stack = kmalloc(sizeof(*stack), gfp_nested_mask(gfp_mask));
 185	if (!stack) {
 186		unset_current_in_page_owner();
 187		return;
 188	}
 189	unset_current_in_page_owner();
 190
 191	stack->stack_record = stack_record;
 192	stack->next = NULL;
 193
 194	spin_lock_irqsave(&stack_list_lock, flags);
 195	stack->next = stack_list;
 196	/*
 197	 * This pairs with smp_load_acquire() from function
 198	 * stack_start(). This guarantees that stack_start()
 199	 * will see an updated stack_list before starting to
 200	 * traverse the list.
 201	 */
 202	smp_store_release(&stack_list, stack);
 203	spin_unlock_irqrestore(&stack_list_lock, flags);
 204}
 205
 206static void inc_stack_record_count(depot_stack_handle_t handle, gfp_t gfp_mask,
 207				   int nr_base_pages)
 208{
 209	struct stack_record *stack_record = __stack_depot_get_stack_record(handle);
 210
 211	if (!stack_record)
 212		return;
 213
 214	/*
 215	 * New stack_record's that do not use STACK_DEPOT_FLAG_GET start
 216	 * with REFCOUNT_SATURATED to catch spurious increments of their
 217	 * refcount.
 218	 * Since we do not use STACK_DEPOT_FLAG_GET API, let us
 219	 * set a refcount of 1 ourselves.
 220	 */
 221	if (refcount_read(&stack_record->count) == REFCOUNT_SATURATED) {
 222		int old = REFCOUNT_SATURATED;
 223
 224		if (atomic_try_cmpxchg_relaxed(&stack_record->count.refs, &old, 1))
 225			/* Add the new stack_record to our list */
 226			add_stack_record_to_list(stack_record, gfp_mask);
 227	}
 228	refcount_add(nr_base_pages, &stack_record->count);
 229}
 230
 231static void dec_stack_record_count(depot_stack_handle_t handle,
 232				   int nr_base_pages)
 233{
 234	struct stack_record *stack_record = __stack_depot_get_stack_record(handle);
 235
 236	if (!stack_record)
 237		return;
 238
 239	if (refcount_sub_and_test(nr_base_pages, &stack_record->count))
 240		pr_warn("%s: refcount went to 0 for %u handle\n", __func__,
 241			handle);
 242}
 243
 244static inline void __update_page_owner_handle(struct page *page,
 245					      depot_stack_handle_t handle,
 246					      unsigned short order,
 247					      gfp_t gfp_mask,
 248					      short last_migrate_reason, u64 ts_nsec,
 249					      pid_t pid, pid_t tgid, char *comm)
 250{
 251	struct page_ext_iter iter;
 252	struct page_ext *page_ext;
 253	struct page_owner *page_owner;
 254
 255	rcu_read_lock();
 256	for_each_page_ext(page, 1 << order, page_ext, iter) {
 257		page_owner = get_page_owner(page_ext);
 258		page_owner->handle = handle;
 259		page_owner->order = order;
 260		page_owner->gfp_mask = gfp_mask;
 261		page_owner->last_migrate_reason = last_migrate_reason;
 262		page_owner->pid = pid;
 263		page_owner->tgid = tgid;
 264		page_owner->ts_nsec = ts_nsec;
 265		strscpy(page_owner->comm, comm,
 266			sizeof(page_owner->comm));
 267		__set_bit(PAGE_EXT_OWNER, &page_ext->flags);
 268		__set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
 269	}
 270	rcu_read_unlock();
 271}
 272
 273static inline void __update_page_owner_free_handle(struct page *page,
 274						   depot_stack_handle_t handle,
 275						   unsigned short order,
 276						   pid_t pid, pid_t tgid,
 277						   u64 free_ts_nsec)
 278{
 279	struct page_ext_iter iter;
 280	struct page_ext *page_ext;
 281	struct page_owner *page_owner;
 282
 283	rcu_read_lock();
 284	for_each_page_ext(page, 1 << order, page_ext, iter) {
 285		page_owner = get_page_owner(page_ext);
 286		/* Only __reset_page_owner() wants to clear the bit */
 287		if (handle) {
 288			__clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
 289			page_owner->free_handle = handle;
 290		}
 291		page_owner->free_ts_nsec = free_ts_nsec;
 292		page_owner->free_pid = current->pid;
 293		page_owner->free_tgid = current->tgid;
 294	}
 295	rcu_read_unlock();
 296}
 297
 298void __reset_page_owner(struct page *page, unsigned short order)
 299{
 300	struct page_ext *page_ext;
 301	depot_stack_handle_t handle;
 302	depot_stack_handle_t alloc_handle;
 303	struct page_owner *page_owner;
 304	u64 free_ts_nsec = local_clock();
 305
 306	page_ext = page_ext_get(page);
 307	if (unlikely(!page_ext))
 308		return;
 309
 310	page_owner = get_page_owner(page_ext);
 311	alloc_handle = page_owner->handle;
 312	page_ext_put(page_ext);
 313
 314	/*
 315	 * Do not specify GFP_NOWAIT to make gfpflags_allow_spinning() == false
 316	 * to prevent issues in stack_depot_save().
 317	 * This is similar to alloc_pages_nolock() gfp flags, but only used
 318	 * to signal stack_depot to avoid spin_locks.
 319	 */
 320	handle = save_stack(__GFP_NOWARN);
 321	__update_page_owner_free_handle(page, handle, order, current->pid,
 322					current->tgid, free_ts_nsec);
 323
 324	if (alloc_handle != early_handle)
 325		/*
 326		 * early_handle is being set as a handle for all those
 327		 * early allocated pages. See init_pages_in_zone().
 328		 * Since their refcount is not being incremented because
 329		 * the machinery is not ready yet, we cannot decrement
 330		 * their refcount either.
 331		 */
 332		dec_stack_record_count(alloc_handle, 1 << order);
 333}
 334
 335noinline void __set_page_owner(struct page *page, unsigned short order,
 336					gfp_t gfp_mask)
 337{
 338	u64 ts_nsec = local_clock();
 339	depot_stack_handle_t handle;
 340
 341	handle = save_stack(gfp_mask);
 342	__update_page_owner_handle(page, handle, order, gfp_mask, -1,
 343				   ts_nsec, current->pid, current->tgid,
 344				   current->comm);
 345	inc_stack_record_count(handle, gfp_mask, 1 << order);
 346}
 347
 348void __folio_set_owner_migrate_reason(struct folio *folio, int reason)
 349{
 350	struct page_ext *page_ext = page_ext_get(&folio->page);
 351	struct page_owner *page_owner;
 352
 353	if (unlikely(!page_ext))
 354		return;
 355
 356	page_owner = get_page_owner(page_ext);
 357	page_owner->last_migrate_reason = reason;
 358	page_ext_put(page_ext);
 359}
 360
 361void __split_page_owner(struct page *page, int old_order, int new_order)
 362{
 363	struct page_ext_iter iter;
 364	struct page_ext *page_ext;
 365	struct page_owner *page_owner;
 366
 367	rcu_read_lock();
 368	for_each_page_ext(page, 1 << old_order, page_ext, iter) {
 369		page_owner = get_page_owner(page_ext);
 370		page_owner->order = new_order;
 371	}
 372	rcu_read_unlock();
 373}
 374
 375void __folio_copy_owner(struct folio *newfolio, struct folio *old)
 376{
 377	struct page_ext *page_ext;
 378	struct page_ext_iter iter;
 379	struct page_owner *old_page_owner;
 380	struct page_owner *new_page_owner;
 381	depot_stack_handle_t migrate_handle;
 382
 383	page_ext = page_ext_get(&old->page);
 384	if (unlikely(!page_ext))
 385		return;
 386
 387	old_page_owner = get_page_owner(page_ext);
 388	page_ext_put(page_ext);
 389
 390	page_ext = page_ext_get(&newfolio->page);
 391	if (unlikely(!page_ext))
 392		return;
 393
 394	new_page_owner = get_page_owner(page_ext);
 395	page_ext_put(page_ext);
 396
 397	migrate_handle = new_page_owner->handle;
 398	__update_page_owner_handle(&newfolio->page, old_page_owner->handle,
 399				   old_page_owner->order, old_page_owner->gfp_mask,
 400				   old_page_owner->last_migrate_reason,
 401				   old_page_owner->ts_nsec, old_page_owner->pid,
 402				   old_page_owner->tgid, old_page_owner->comm);
 403	/*
 404	 * Do not proactively clear PAGE_EXT_OWNER{_ALLOCATED} bits as the folio
 405	 * will be freed after migration. Keep them until then as they may be
 406	 * useful.
 407	 */
 408	__update_page_owner_free_handle(&newfolio->page, 0, old_page_owner->order,
 409					old_page_owner->free_pid,
 410					old_page_owner->free_tgid,
 411					old_page_owner->free_ts_nsec);
 412	/*
 413	 * We linked the original stack to the new folio, we need to do the same
 414	 * for the new one and the old folio otherwise there will be an imbalance
 415	 * when subtracting those pages from the stack.
 416	 */
 417	rcu_read_lock();
 418	for_each_page_ext(&old->page, 1 << new_page_owner->order, page_ext, iter) {
 419		old_page_owner = get_page_owner(page_ext);
 420		old_page_owner->handle = migrate_handle;
 421	}
 422	rcu_read_unlock();
 423}
 424
 425void pagetypeinfo_showmixedcount_print(struct seq_file *m,
 426				       pg_data_t *pgdat, struct zone *zone)
 427{
 428	struct page *page;
 429	struct page_ext *page_ext;
 430	struct page_owner *page_owner;
 431	unsigned long pfn, block_end_pfn;
 432	unsigned long end_pfn = zone_end_pfn(zone);
 433	unsigned long count[MIGRATE_TYPES] = { 0, };
 434	int pageblock_mt, page_mt;
 435	int i;
 436
 437	/* Scan block by block. First and last block may be incomplete */
 438	pfn = zone->zone_start_pfn;
 439
 440	/*
 441	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
 442	 * a zone boundary, it will be double counted between zones. This does
 443	 * not matter as the mixed block count will still be correct
 444	 */
 445	for (; pfn < end_pfn; ) {
 446		page = pfn_to_online_page(pfn);
 447		if (!page) {
 448			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
 449			continue;
 450		}
 451
 452		block_end_pfn = pageblock_end_pfn(pfn);
 453		block_end_pfn = min(block_end_pfn, end_pfn);
 454
 455		pageblock_mt = get_pageblock_migratetype(page);
 456
 457		for (; pfn < block_end_pfn; pfn++) {
 458			/* The pageblock is online, no need to recheck. */
 459			page = pfn_to_page(pfn);
 460
 461			if (page_zone(page) != zone)
 462				continue;
 463
 464			if (PageBuddy(page)) {
 465				unsigned long freepage_order;
 466
 467				freepage_order = buddy_order_unsafe(page);
 468				if (freepage_order <= MAX_PAGE_ORDER)
 469					pfn += (1UL << freepage_order) - 1;
 470				continue;
 471			}
 472
 473			if (PageReserved(page))
 474				continue;
 475
 476			page_ext = page_ext_get(page);
 477			if (unlikely(!page_ext))
 478				continue;
 479
 480			if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
 481				goto ext_put_continue;
 482
 483			page_owner = get_page_owner(page_ext);
 484			page_mt = gfp_migratetype(page_owner->gfp_mask);
 485			if (pageblock_mt != page_mt) {
 486				if (is_migrate_cma(pageblock_mt))
 487					count[MIGRATE_MOVABLE]++;
 488				else
 489					count[pageblock_mt]++;
 490
 491				pfn = block_end_pfn;
 492				page_ext_put(page_ext);
 493				break;
 494			}
 495			pfn += (1UL << page_owner->order) - 1;
 496ext_put_continue:
 497			page_ext_put(page_ext);
 498		}
 499	}
 500
 501	/* Print counts */
 502	seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
 503	for (i = 0; i < MIGRATE_TYPES; i++)
 504		seq_printf(m, "%12lu ", count[i]);
 505	seq_putc(m, '\n');
 506}
 507
 508/*
 509 * Looking for memcg information and print it out
 510 */
 511static inline int print_page_owner_memcg(char *kbuf, size_t count, int ret,
 512					 struct page *page)
 513{
 514#ifdef CONFIG_MEMCG
 515	unsigned long memcg_data;
 516	struct mem_cgroup *memcg;
 517	bool online;
 518	char name[80];
 519
 520	rcu_read_lock();
 521	memcg_data = READ_ONCE(page->memcg_data);
 522	if (!memcg_data || PageTail(page))
 523		goto out_unlock;
 524
 525	if (memcg_data & MEMCG_DATA_OBJEXTS)
 526		ret += scnprintf(kbuf + ret, count - ret,
 527				"Slab cache page\n");
 528
 529	memcg = page_memcg_check(page);
 530	if (!memcg)
 531		goto out_unlock;
 532
 533	online = (memcg->css.flags & CSS_ONLINE);
 534	cgroup_name(memcg->css.cgroup, name, sizeof(name));
 535	ret += scnprintf(kbuf + ret, count - ret,
 536			"Charged %sto %smemcg %s\n",
 537			PageMemcgKmem(page) ? "(via objcg) " : "",
 538			online ? "" : "offline ",
 539			name);
 540out_unlock:
 541	rcu_read_unlock();
 542#endif /* CONFIG_MEMCG */
 543
 544	return ret;
 545}
 546
 547static ssize_t
 548print_page_owner(char __user *buf, size_t count, unsigned long pfn,
 549		struct page *page, struct page_owner *page_owner,
 550		depot_stack_handle_t handle)
 551{
 552	int ret, pageblock_mt, page_mt;
 553	char *kbuf;
 554
 555	count = min_t(size_t, count, PAGE_SIZE);
 556	kbuf = kmalloc(count, GFP_KERNEL);
 557	if (!kbuf)
 558		return -ENOMEM;
 559
 560	ret = scnprintf(kbuf, count,
 561			"Page allocated via order %u, mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu ns\n",
 562			page_owner->order, page_owner->gfp_mask,
 563			&page_owner->gfp_mask, page_owner->pid,
 564			page_owner->tgid, page_owner->comm,
 565			page_owner->ts_nsec);
 566
 567	/* Print information relevant to grouping pages by mobility */
 568	pageblock_mt = get_pageblock_migratetype(page);
 569	page_mt  = gfp_migratetype(page_owner->gfp_mask);
 570	ret += scnprintf(kbuf + ret, count - ret,
 571			"PFN 0x%lx type %s Block %lu type %s Flags %pGp\n",
 572			pfn,
 573			migratetype_names[page_mt],
 574			pfn >> pageblock_order,
 575			migratetype_names[pageblock_mt],
 576			&page->flags);
 577
 578	ret += stack_depot_snprint(handle, kbuf + ret, count - ret, 0);
 579	if (ret >= count)
 580		goto err;
 581
 582	if (page_owner->last_migrate_reason != -1) {
 583		ret += scnprintf(kbuf + ret, count - ret,
 584			"Page has been migrated, last migrate reason: %s\n",
 585			migrate_reason_names[page_owner->last_migrate_reason]);
 586	}
 587
 588	ret = print_page_owner_memcg(kbuf, count, ret, page);
 589
 590	ret += snprintf(kbuf + ret, count - ret, "\n");
 591	if (ret >= count)
 592		goto err;
 593
 594	if (copy_to_user(buf, kbuf, ret))
 595		ret = -EFAULT;
 596
 597	kfree(kbuf);
 598	return ret;
 599
 600err:
 601	kfree(kbuf);
 602	return -ENOMEM;
 603}
 604
 605void __dump_page_owner(const struct page *page)
 606{
 607	struct page_ext *page_ext = page_ext_get((void *)page);
 608	struct page_owner *page_owner;
 609	depot_stack_handle_t handle;
 610	gfp_t gfp_mask;
 611	int mt;
 612
 613	if (unlikely(!page_ext)) {
 614		pr_alert("There is not page extension available.\n");
 615		return;
 616	}
 617
 618	page_owner = get_page_owner(page_ext);
 619	gfp_mask = page_owner->gfp_mask;
 620	mt = gfp_migratetype(gfp_mask);
 621
 622	if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags)) {
 623		pr_alert("page_owner info is not present (never set?)\n");
 624		page_ext_put(page_ext);
 625		return;
 626	}
 627
 628	if (test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
 629		pr_alert("page_owner tracks the page as allocated\n");
 630	else
 631		pr_alert("page_owner tracks the page as freed\n");
 632
 633	pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, tgid %d (%s), ts %llu, free_ts %llu\n",
 634		 page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask,
 635		 page_owner->pid, page_owner->tgid, page_owner->comm,
 636		 page_owner->ts_nsec, page_owner->free_ts_nsec);
 637
 638	handle = READ_ONCE(page_owner->handle);
 639	if (!handle)
 640		pr_alert("page_owner allocation stack trace missing\n");
 641	else
 642		stack_depot_print(handle);
 643
 644	handle = READ_ONCE(page_owner->free_handle);
 645	if (!handle) {
 646		pr_alert("page_owner free stack trace missing\n");
 647	} else {
 648		pr_alert("page last free pid %d tgid %d stack trace:\n",
 649			  page_owner->free_pid, page_owner->free_tgid);
 650		stack_depot_print(handle);
 651	}
 652
 653	if (page_owner->last_migrate_reason != -1)
 654		pr_alert("page has been migrated, last migrate reason: %s\n",
 655			migrate_reason_names[page_owner->last_migrate_reason]);
 656	page_ext_put(page_ext);
 657}
 658
 659static ssize_t
 660read_page_owner(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 661{
 662	unsigned long pfn;
 663	struct page *page;
 664	struct page_ext *page_ext;
 665	struct page_owner *page_owner;
 666	depot_stack_handle_t handle;
 667
 668	if (!static_branch_unlikely(&page_owner_inited))
 669		return -EINVAL;
 670
 671	page = NULL;
 672	if (*ppos == 0)
 673		pfn = min_low_pfn;
 674	else
 675		pfn = *ppos;
 676	/* Find a valid PFN or the start of a MAX_ORDER_NR_PAGES area */
 677	while (!pfn_valid(pfn) && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0)
 678		pfn++;
 679
 680	/* Find an allocated page */
 681	for (; pfn < max_pfn; pfn++) {
 682		/*
 683		 * This temporary page_owner is required so
 684		 * that we can avoid the context switches while holding
 685		 * the rcu lock and copying the page owner information to
 686		 * user through copy_to_user() or GFP_KERNEL allocations.
 687		 */
 688		struct page_owner page_owner_tmp;
 689
 690		/*
 691		 * If the new page is in a new MAX_ORDER_NR_PAGES area,
 692		 * validate the area as existing, skip it if not
 693		 */
 694		if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0 && !pfn_valid(pfn)) {
 695			pfn += MAX_ORDER_NR_PAGES - 1;
 696			continue;
 697		}
 698
 699		page = pfn_to_page(pfn);
 700		if (PageBuddy(page)) {
 701			unsigned long freepage_order = buddy_order_unsafe(page);
 702
 703			if (freepage_order <= MAX_PAGE_ORDER)
 704				pfn += (1UL << freepage_order) - 1;
 705			continue;
 706		}
 707
 708		page_ext = page_ext_get(page);
 709		if (unlikely(!page_ext))
 710			continue;
 711
 712		/*
 713		 * Some pages could be missed by concurrent allocation or free,
 714		 * because we don't hold the zone lock.
 715		 */
 716		if (!test_bit(PAGE_EXT_OWNER, &page_ext->flags))
 717			goto ext_put_continue;
 718
 719		/*
 720		 * Although we do have the info about past allocation of free
 721		 * pages, it's not relevant for current memory usage.
 722		 */
 723		if (!test_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags))
 724			goto ext_put_continue;
 725
 726		page_owner = get_page_owner(page_ext);
 727
 728		/*
 729		 * Don't print "tail" pages of high-order allocations as that
 730		 * would inflate the stats.
 731		 */
 732		if (!IS_ALIGNED(pfn, 1 << page_owner->order))
 733			goto ext_put_continue;
 734
 735		/*
 736		 * Access to page_ext->handle isn't synchronous so we should
 737		 * be careful to access it.
 738		 */
 739		handle = READ_ONCE(page_owner->handle);
 740		if (!handle)
 741			goto ext_put_continue;
 742
 743		/* Record the next PFN to read in the file offset */
 744		*ppos = pfn + 1;
 745
 746		page_owner_tmp = *page_owner;
 747		page_ext_put(page_ext);
 748		return print_page_owner(buf, count, pfn, page,
 749				&page_owner_tmp, handle);
 750ext_put_continue:
 751		page_ext_put(page_ext);
 752	}
 753
 754	return 0;
 755}
 756
 757static loff_t lseek_page_owner(struct file *file, loff_t offset, int orig)
 758{
 759	switch (orig) {
 760	case SEEK_SET:
 761		file->f_pos = offset;
 762		break;
 763	case SEEK_CUR:
 764		file->f_pos += offset;
 765		break;
 766	default:
 767		return -EINVAL;
 768	}
 769	return file->f_pos;
 770}
 771
 772static void init_pages_in_zone(struct zone *zone)
 773{
 774	unsigned long pfn = zone->zone_start_pfn;
 775	unsigned long end_pfn = zone_end_pfn(zone);
 776	unsigned long count = 0;
 777
 778	/*
 779	 * Walk the zone in pageblock_nr_pages steps. If a page block spans
 780	 * a zone boundary, it will be double counted between zones. This does
 781	 * not matter as the mixed block count will still be correct
 782	 */
 783	for (; pfn < end_pfn; ) {
 784		unsigned long block_end_pfn;
 785
 786		if (!pfn_valid(pfn)) {
 787			pfn = ALIGN(pfn + 1, MAX_ORDER_NR_PAGES);
 788			continue;
 789		}
 790
 791		block_end_pfn = pageblock_end_pfn(pfn);
 792		block_end_pfn = min(block_end_pfn, end_pfn);
 793
 794		for (; pfn < block_end_pfn; pfn++) {
 795			struct page *page = pfn_to_page(pfn);
 796			struct page_ext *page_ext;
 797
 798			if (page_zone(page) != zone)
 799				continue;
 800
 801			/*
 802			 * To avoid having to grab zone->lock, be a little
 803			 * careful when reading buddy page order. The only
 804			 * danger is that we skip too much and potentially miss
 805			 * some early allocated pages, which is better than
 806			 * heavy lock contention.
 807			 */
 808			if (PageBuddy(page)) {
 809				unsigned long order = buddy_order_unsafe(page);
 810
 811				if (order > 0 && order <= MAX_PAGE_ORDER)
 812					pfn += (1UL << order) - 1;
 813				continue;
 814			}
 815
 816			if (PageReserved(page))
 817				continue;
 818
 819			page_ext = page_ext_get(page);
 820			if (unlikely(!page_ext))
 821				continue;
 822
 823			/* Maybe overlapping zone */
 824			if (test_bit(PAGE_EXT_OWNER, &page_ext->flags))
 825				goto ext_put_continue;
 826
 827			/* Found early allocated page */
 828			__update_page_owner_handle(page, early_handle, 0, 0,
 829						   -1, local_clock(), current->pid,
 830						   current->tgid, current->comm);
 831			count++;
 832ext_put_continue:
 833			page_ext_put(page_ext);
 834		}
 835		cond_resched();
 836	}
 837
 838	pr_info("Node %d, zone %8s: page owner found early allocated %lu pages\n",
 839		zone->zone_pgdat->node_id, zone->name, count);
 840}
 841
 842static void init_early_allocated_pages(void)
 843{
 844	struct zone *zone;
 845
 846	for_each_populated_zone(zone)
 847		init_pages_in_zone(zone);
 848}
 849
 850static const struct file_operations page_owner_fops = {
 851	.read		= read_page_owner,
 852	.llseek		= lseek_page_owner,
 853};
 854
 855static void *stack_start(struct seq_file *m, loff_t *ppos)
 856{
 857	struct stack *stack;
 858	struct stack_print_ctx *ctx = m->private;
 859
 860	if (*ppos == -1UL)
 861		return NULL;
 862
 863	if (!*ppos) {
 864		/*
 865		 * This pairs with smp_store_release() from function
 866		 * add_stack_record_to_list(), so we get a consistent
 867		 * value of stack_list.
 868		 */
 869		stack = smp_load_acquire(&stack_list);
 870		ctx->stack = stack;
 871	} else {
 872		stack = ctx->stack;
 873	}
 874
 875	return stack;
 876}
 877
 878static void *stack_next(struct seq_file *m, void *v, loff_t *ppos)
 879{
 880	struct stack *stack = v;
 881	struct stack_print_ctx *ctx = m->private;
 882
 883	stack = stack->next;
 884	*ppos = stack ? *ppos + 1 : -1UL;
 885	ctx->stack = stack;
 886
 887	return stack;
 888}
 889
 890static unsigned long page_owner_pages_threshold;
 891
 892static int stack_print(struct seq_file *m, void *v)
 893{
 894	int i, nr_base_pages;
 895	struct stack *stack = v;
 896	unsigned long *entries;
 897	unsigned long nr_entries;
 898	struct stack_record *stack_record = stack->stack_record;
 899	struct stack_print_ctx *ctx = m->private;
 900
 901	if (!stack->stack_record)
 902		return 0;
 903
 904	nr_base_pages = refcount_read(&stack_record->count) - 1;
 905
 906	if (ctx->flags & STACK_PRINT_FLAG_PAGES &&
 907	    (nr_base_pages < 1 || nr_base_pages < page_owner_pages_threshold))
 908		return 0;
 909
 910	if (ctx->flags & STACK_PRINT_FLAG_STACK) {
 911		nr_entries = stack_record->size;
 912		entries = stack_record->entries;
 913		for (i = 0; i < nr_entries; i++)
 914			seq_printf(m, " %pS\n", (void *)entries[i]);
 915	}
 916	if (ctx->flags & STACK_PRINT_FLAG_HANDLE)
 917		seq_printf(m, "handle: %d\n", stack_record->handle.handle);
 918	if (ctx->flags & STACK_PRINT_FLAG_PAGES)
 919		seq_printf(m, "nr_base_pages: %d\n", nr_base_pages);
 920	seq_putc(m, '\n');
 921
 922	return 0;
 923}
 924
 925static void stack_stop(struct seq_file *m, void *v)
 926{
 927}
 928
 929static const struct seq_operations page_owner_stack_op = {
 930	.start	= stack_start,
 931	.next	= stack_next,
 932	.stop	= stack_stop,
 933	.show	= stack_print
 934};
 935
 936static int page_owner_stack_open(struct inode *inode, struct file *file)
 937{
 938	int ret = seq_open_private(file, &page_owner_stack_op,
 939				   sizeof(struct stack_print_ctx));
 940
 941	if (!ret) {
 942		struct seq_file *m = file->private_data;
 943		struct stack_print_ctx *ctx = m->private;
 944
 945		ctx->flags = (uintptr_t) inode->i_private;
 946	}
 947
 948	return ret;
 949}
 950
 951static const struct file_operations page_owner_stack_fops = {
 952	.open		= page_owner_stack_open,
 953	.read		= seq_read,
 954	.llseek		= seq_lseek,
 955	.release	= seq_release_private,
 956};
 957
 958static int page_owner_threshold_get(void *data, u64 *val)
 959{
 960	*val = READ_ONCE(page_owner_pages_threshold);
 961	return 0;
 962}
 963
 964static int page_owner_threshold_set(void *data, u64 val)
 965{
 966	WRITE_ONCE(page_owner_pages_threshold, val);
 967	return 0;
 968}
 969
 970DEFINE_SIMPLE_ATTRIBUTE(page_owner_threshold_fops, &page_owner_threshold_get,
 971			&page_owner_threshold_set, "%llu");
 972
 973
 974static int __init pageowner_init(void)
 975{
 976	struct dentry *dir;
 977
 978	if (!static_branch_unlikely(&page_owner_inited)) {
 979		pr_info("page_owner is disabled\n");
 980		return 0;
 981	}
 982
 983	debugfs_create_file("page_owner", 0400, NULL, NULL, &page_owner_fops);
 984	dir = debugfs_create_dir("page_owner_stacks", NULL);
 985	debugfs_create_file("show_stacks", 0400, dir,
 986			    (void *)(STACK_PRINT_FLAG_STACK |
 987				     STACK_PRINT_FLAG_PAGES),
 988			     &page_owner_stack_fops);
 989	debugfs_create_file("show_handles", 0400, dir,
 990			    (void *)(STACK_PRINT_FLAG_HANDLE |
 991				     STACK_PRINT_FLAG_PAGES),
 992			    &page_owner_stack_fops);
 993	debugfs_create_file("show_stacks_handles", 0400, dir,
 994			    (void *)(STACK_PRINT_FLAG_STACK |
 995				     STACK_PRINT_FLAG_HANDLE),
 996			    &page_owner_stack_fops);
 997	debugfs_create_file("count_threshold", 0600, dir, NULL,
 998			    &page_owner_threshold_fops);
 999	return 0;
1000}
1001late_initcall(pageowner_init)