internal.h

   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/* internal.h: mm/ internal definitions
   3 *
   4 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
   5 * Written by David Howells (dhowells@redhat.com)
   6 */
   7#ifndef __MM_INTERNAL_H
   8#define __MM_INTERNAL_H
   9
  10#include <linux/fs.h>
  11#include <linux/khugepaged.h>
  12#include <linux/mm.h>
  13#include <linux/mm_inline.h>
  14#include <linux/pagemap.h>
  15#include <linux/pagewalk.h>
  16#include <linux/rmap.h>
  17#include <linux/swap.h>
  18#include <linux/leafops.h>
  19#include <linux/swap_cgroup.h>
  20#include <linux/tracepoint-defs.h>
  21
  22/* Internal core VMA manipulation functions. */
  23#include "vma.h"
  24
  25struct folio_batch;
  26
  27/*
  28 * Maintains state across a page table move. The operation assumes both source
  29 * and destination VMAs already exist and are specified by the user.
  30 *
  31 * Partial moves are permitted, but the old and new ranges must both reside
  32 * within a VMA.
  33 *
  34 * mmap lock must be held in write and VMA write locks must be held on any VMA
  35 * that is visible.
  36 *
  37 * Use the PAGETABLE_MOVE() macro to initialise this struct.
  38 *
  39 * The old_addr and new_addr fields are updated as the page table move is
  40 * executed.
  41 *
  42 * NOTE: The page table move is affected by reading from [old_addr, old_end),
  43 * and old_addr may be updated for better page table alignment, so len_in
  44 * represents the length of the range being copied as specified by the user.
  45 */
  46struct pagetable_move_control {
  47	struct vm_area_struct *old; /* Source VMA. */
  48	struct vm_area_struct *new; /* Destination VMA. */
  49	unsigned long old_addr; /* Address from which the move begins. */
  50	unsigned long old_end; /* Exclusive address at which old range ends. */
  51	unsigned long new_addr; /* Address to move page tables to. */
  52	unsigned long len_in; /* Bytes to remap specified by user. */
  53
  54	bool need_rmap_locks; /* Do rmap locks need to be taken? */
  55	bool for_stack; /* Is this an early temp stack being moved? */
  56};
  57
  58#define PAGETABLE_MOVE(name, old_, new_, old_addr_, new_addr_, len_)	\
  59	struct pagetable_move_control name = {				\
  60		.old = old_,						\
  61		.new = new_,						\
  62		.old_addr = old_addr_,					\
  63		.old_end = (old_addr_) + (len_),			\
  64		.new_addr = new_addr_,					\
  65		.len_in = len_,						\
  66	}
  67
  68/*
  69 * The set of flags that only affect watermark checking and reclaim
  70 * behaviour. This is used by the MM to obey the caller constraints
  71 * about IO, FS and watermark checking while ignoring placement
  72 * hints such as HIGHMEM usage.
  73 */
  74#define GFP_RECLAIM_MASK (__GFP_RECLAIM|__GFP_HIGH|__GFP_IO|__GFP_FS|\
  75			__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_NOFAIL|\
  76			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC|\
  77			__GFP_NOLOCKDEP)
  78
  79/* The GFP flags allowed during early boot */
  80#define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_RECLAIM|__GFP_IO|__GFP_FS))
  81
  82/* Control allocation cpuset and node placement constraints */
  83#define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
  84
  85/* Do not use these with a slab allocator */
  86#define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
  87
  88/*
  89 * Different from WARN_ON_ONCE(), no warning will be issued
  90 * when we specify __GFP_NOWARN.
  91 */
  92#define WARN_ON_ONCE_GFP(cond, gfp)	({				\
  93	static bool __section(".data..once") __warned;			\
  94	int __ret_warn_once = !!(cond);					\
  95									\
  96	if (unlikely(!(gfp & __GFP_NOWARN) && __ret_warn_once && !__warned)) { \
  97		__warned = true;					\
  98		WARN_ON(1);						\
  99	}								\
 100	unlikely(__ret_warn_once);					\
 101})
 102
 103void page_writeback_init(void);
 104
 105/*
 106 * If a 16GB hugetlb folio were mapped by PTEs of all of its 4kB pages,
 107 * its nr_pages_mapped would be 0x400000: choose the ENTIRELY_MAPPED bit
 108 * above that range, instead of 2*(PMD_SIZE/PAGE_SIZE).  Hugetlb currently
 109 * leaves nr_pages_mapped at 0, but avoid surprise if it participates later.
 110 */
 111#define ENTIRELY_MAPPED		0x800000
 112#define FOLIO_PAGES_MAPPED	(ENTIRELY_MAPPED - 1)
 113
 114/*
 115 * Flags passed to __show_mem() and show_free_areas() to suppress output in
 116 * various contexts.
 117 */
 118#define SHOW_MEM_FILTER_NODES		(0x0001u)	/* disallowed nodes */
 119
 120/*
 121 * How many individual pages have an elevated _mapcount.  Excludes
 122 * the folio's entire_mapcount.
 123 *
 124 * Don't use this function outside of debugging code.
 125 */
 126static inline int folio_nr_pages_mapped(const struct folio *folio)
 127{
 128	if (IS_ENABLED(CONFIG_NO_PAGE_MAPCOUNT))
 129		return -1;
 130	return atomic_read(&folio->_nr_pages_mapped) & FOLIO_PAGES_MAPPED;
 131}
 132
 133/*
 134 * Retrieve the first entry of a folio based on a provided entry within the
 135 * folio. We cannot rely on folio->swap as there is no guarantee that it has
 136 * been initialized. Used for calling arch_swap_restore()
 137 */
 138static inline swp_entry_t folio_swap(swp_entry_t entry,
 139		const struct folio *folio)
 140{
 141	swp_entry_t swap = {
 142		.val = ALIGN_DOWN(entry.val, folio_nr_pages(folio)),
 143	};
 144
 145	return swap;
 146}
 147
 148static inline void *folio_raw_mapping(const struct folio *folio)
 149{
 150	unsigned long mapping = (unsigned long)folio->mapping;
 151
 152	return (void *)(mapping & ~FOLIO_MAPPING_FLAGS);
 153}
 154
 155/*
 156 * This is a file-backed mapping, and is about to be memory mapped - invoke its
 157 * mmap hook and safely handle error conditions. On error, VMA hooks will be
 158 * mutated.
 159 *
 160 * @file: File which backs the mapping.
 161 * @vma:  VMA which we are mapping.
 162 *
 163 * Returns: 0 if success, error otherwise.
 164 */
 165static inline int mmap_file(struct file *file, struct vm_area_struct *vma)
 166{
 167	int err = vfs_mmap(file, vma);
 168
 169	if (likely(!err))
 170		return 0;
 171
 172	/*
 173	 * OK, we tried to call the file hook for mmap(), but an error
 174	 * arose. The mapping is in an inconsistent state and we most not invoke
 175	 * any further hooks on it.
 176	 */
 177	vma->vm_ops = &vma_dummy_vm_ops;
 178
 179	return err;
 180}
 181
 182/*
 183 * If the VMA has a close hook then close it, and since closing it might leave
 184 * it in an inconsistent state which makes the use of any hooks suspect, clear
 185 * them down by installing dummy empty hooks.
 186 */
 187static inline void vma_close(struct vm_area_struct *vma)
 188{
 189	if (vma->vm_ops && vma->vm_ops->close) {
 190		vma->vm_ops->close(vma);
 191
 192		/*
 193		 * The mapping is in an inconsistent state, and no further hooks
 194		 * may be invoked upon it.
 195		 */
 196		vma->vm_ops = &vma_dummy_vm_ops;
 197	}
 198}
 199
 200#ifdef CONFIG_MMU
 201
 202/* Flags for folio_pte_batch(). */
 203typedef int __bitwise fpb_t;
 204
 205/* Compare PTEs respecting the dirty bit. */
 206#define FPB_RESPECT_DIRTY		((__force fpb_t)BIT(0))
 207
 208/* Compare PTEs respecting the soft-dirty bit. */
 209#define FPB_RESPECT_SOFT_DIRTY		((__force fpb_t)BIT(1))
 210
 211/* Compare PTEs respecting the writable bit. */
 212#define FPB_RESPECT_WRITE		((__force fpb_t)BIT(2))
 213
 214/*
 215 * Merge PTE write bits: if any PTE in the batch is writable, modify the
 216 * PTE at @ptentp to be writable.
 217 */
 218#define FPB_MERGE_WRITE			((__force fpb_t)BIT(3))
 219
 220/*
 221 * Merge PTE young and dirty bits: if any PTE in the batch is young or dirty,
 222 * modify the PTE at @ptentp to be young or dirty, respectively.
 223 */
 224#define FPB_MERGE_YOUNG_DIRTY		((__force fpb_t)BIT(4))
 225
 226static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
 227{
 228	if (!(flags & FPB_RESPECT_DIRTY))
 229		pte = pte_mkclean(pte);
 230	if (likely(!(flags & FPB_RESPECT_SOFT_DIRTY)))
 231		pte = pte_clear_soft_dirty(pte);
 232	if (likely(!(flags & FPB_RESPECT_WRITE)))
 233		pte = pte_wrprotect(pte);
 234	return pte_mkold(pte);
 235}
 236
 237/**
 238 * folio_pte_batch_flags - detect a PTE batch for a large folio
 239 * @folio: The large folio to detect a PTE batch for.
 240 * @vma: The VMA. Only relevant with FPB_MERGE_WRITE, otherwise can be NULL.
 241 * @ptep: Page table pointer for the first entry.
 242 * @ptentp: Pointer to a COPY of the first page table entry whose flags this
 243 *	    function updates based on @flags if appropriate.
 244 * @max_nr: The maximum number of table entries to consider.
 245 * @flags: Flags to modify the PTE batch semantics.
 246 *
 247 * Detect a PTE batch: consecutive (present) PTEs that map consecutive
 248 * pages of the same large folio in a single VMA and a single page table.
 249 *
 250 * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
 251 * the accessed bit, writable bit, dirty bit (unless FPB_RESPECT_DIRTY is set)
 252 * and soft-dirty bit (unless FPB_RESPECT_SOFT_DIRTY is set).
 253 *
 254 * @ptep must map any page of the folio. max_nr must be at least one and
 255 * must be limited by the caller so scanning cannot exceed a single VMA and
 256 * a single page table.
 257 *
 258 * Depending on the FPB_MERGE_* flags, the pte stored at @ptentp will
 259 * be updated: it's crucial that a pointer to a COPY of the first
 260 * page table entry, obtained through ptep_get(), is provided as @ptentp.
 261 *
 262 * This function will be inlined to optimize based on the input parameters;
 263 * consider using folio_pte_batch() instead if applicable.
 264 *
 265 * Return: the number of table entries in the batch.
 266 */
 267static inline unsigned int folio_pte_batch_flags(struct folio *folio,
 268		struct vm_area_struct *vma, pte_t *ptep, pte_t *ptentp,
 269		unsigned int max_nr, fpb_t flags)
 270{
 271	bool any_writable = false, any_young = false, any_dirty = false;
 272	pte_t expected_pte, pte = *ptentp;
 273	unsigned int nr, cur_nr;
 274
 275	VM_WARN_ON_FOLIO(!pte_present(pte), folio);
 276	VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
 277	VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
 278	/*
 279	 * Ensure this is a pointer to a copy not a pointer into a page table.
 280	 * If this is a stack value, it won't be a valid virtual address, but
 281	 * that's fine because it also cannot be pointing into the page table.
 282	 */
 283	VM_WARN_ON(virt_addr_valid(ptentp) && PageTable(virt_to_page(ptentp)));
 284
 285	/* Limit max_nr to the actual remaining PFNs in the folio we could batch. */
 286	max_nr = min_t(unsigned long, max_nr,
 287		       folio_pfn(folio) + folio_nr_pages(folio) - pte_pfn(pte));
 288
 289	nr = pte_batch_hint(ptep, pte);
 290	expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
 291	ptep = ptep + nr;
 292
 293	while (nr < max_nr) {
 294		pte = ptep_get(ptep);
 295
 296		if (!pte_same(__pte_batch_clear_ignored(pte, flags), expected_pte))
 297			break;
 298
 299		if (flags & FPB_MERGE_WRITE)
 300			any_writable |= pte_write(pte);
 301		if (flags & FPB_MERGE_YOUNG_DIRTY) {
 302			any_young |= pte_young(pte);
 303			any_dirty |= pte_dirty(pte);
 304		}
 305
 306		cur_nr = pte_batch_hint(ptep, pte);
 307		expected_pte = pte_advance_pfn(expected_pte, cur_nr);
 308		ptep += cur_nr;
 309		nr += cur_nr;
 310	}
 311
 312	if (any_writable)
 313		*ptentp = pte_mkwrite(*ptentp, vma);
 314	if (any_young)
 315		*ptentp = pte_mkyoung(*ptentp);
 316	if (any_dirty)
 317		*ptentp = pte_mkdirty(*ptentp);
 318
 319	return min(nr, max_nr);
 320}
 321
 322unsigned int folio_pte_batch(struct folio *folio, pte_t *ptep, pte_t pte,
 323		unsigned int max_nr);
 324
 325/**
 326 * pte_move_swp_offset - Move the swap entry offset field of a swap pte
 327 *	 forward or backward by delta
 328 * @pte: The initial pte state; must be a swap entry
 329 * @delta: The direction and the offset we are moving; forward if delta
 330 *	 is positive; backward if delta is negative
 331 *
 332 * Moves the swap offset, while maintaining all other fields, including
 333 * swap type, and any swp pte bits. The resulting pte is returned.
 334 */
 335static inline pte_t pte_move_swp_offset(pte_t pte, long delta)
 336{
 337	const softleaf_t entry = softleaf_from_pte(pte);
 338	pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
 339						   (swp_offset(entry) + delta)));
 340
 341	if (pte_swp_soft_dirty(pte))
 342		new = pte_swp_mksoft_dirty(new);
 343	if (pte_swp_exclusive(pte))
 344		new = pte_swp_mkexclusive(new);
 345	if (pte_swp_uffd_wp(pte))
 346		new = pte_swp_mkuffd_wp(new);
 347
 348	return new;
 349}
 350
 351
 352/**
 353 * pte_next_swp_offset - Increment the swap entry offset field of a swap pte.
 354 * @pte: The initial pte state; must be a swap entry.
 355 *
 356 * Increments the swap offset, while maintaining all other fields, including
 357 * swap type, and any swp pte bits. The resulting pte is returned.
 358 */
 359static inline pte_t pte_next_swp_offset(pte_t pte)
 360{
 361	return pte_move_swp_offset(pte, 1);
 362}
 363
 364/**
 365 * swap_pte_batch - detect a PTE batch for a set of contiguous swap entries
 366 * @start_ptep: Page table pointer for the first entry.
 367 * @max_nr: The maximum number of table entries to consider.
 368 * @pte: Page table entry for the first entry.
 369 *
 370 * Detect a batch of contiguous swap entries: consecutive (non-present) PTEs
 371 * containing swap entries all with consecutive offsets and targeting the same
 372 * swap type, all with matching swp pte bits.
 373 *
 374 * max_nr must be at least one and must be limited by the caller so scanning
 375 * cannot exceed a single page table.
 376 *
 377 * Return: the number of table entries in the batch.
 378 */
 379static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
 380{
 381	pte_t expected_pte = pte_next_swp_offset(pte);
 382	const pte_t *end_ptep = start_ptep + max_nr;
 383	const softleaf_t entry = softleaf_from_pte(pte);
 384	pte_t *ptep = start_ptep + 1;
 385	unsigned short cgroup_id;
 386
 387	VM_WARN_ON(max_nr < 1);
 388	VM_WARN_ON(!softleaf_is_swap(entry));
 389
 390	cgroup_id = lookup_swap_cgroup_id(entry);
 391	while (ptep < end_ptep) {
 392		softleaf_t entry;
 393
 394		pte = ptep_get(ptep);
 395
 396		if (!pte_same(pte, expected_pte))
 397			break;
 398		entry = softleaf_from_pte(pte);
 399		if (lookup_swap_cgroup_id(entry) != cgroup_id)
 400			break;
 401		expected_pte = pte_next_swp_offset(expected_pte);
 402		ptep++;
 403	}
 404
 405	return ptep - start_ptep;
 406}
 407#endif /* CONFIG_MMU */
 408
 409void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
 410						int nr_throttled);
 411static inline void acct_reclaim_writeback(struct folio *folio)
 412{
 413	pg_data_t *pgdat = folio_pgdat(folio);
 414	int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);
 415
 416	if (nr_throttled)
 417		__acct_reclaim_writeback(pgdat, folio, nr_throttled);
 418}
 419
 420static inline void wake_throttle_isolated(pg_data_t *pgdat)
 421{
 422	wait_queue_head_t *wqh;
 423
 424	wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
 425	if (waitqueue_active(wqh))
 426		wake_up(wqh);
 427}
 428
 429vm_fault_t __vmf_anon_prepare(struct vm_fault *vmf);
 430static inline vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
 431{
 432	vm_fault_t ret = __vmf_anon_prepare(vmf);
 433
 434	if (unlikely(ret & VM_FAULT_RETRY))
 435		vma_end_read(vmf->vma);
 436	return ret;
 437}
 438
 439vm_fault_t do_swap_page(struct vm_fault *vmf);
 440void folio_rotate_reclaimable(struct folio *folio);
 441bool __folio_end_writeback(struct folio *folio);
 442void deactivate_file_folio(struct folio *folio);
 443void folio_activate(struct folio *folio);
 444
 445void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
 446		   struct vm_area_struct *start_vma, unsigned long floor,
 447		   unsigned long ceiling, bool mm_wr_locked);
 448void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte);
 449
 450struct zap_details;
 451void unmap_page_range(struct mmu_gather *tlb,
 452			     struct vm_area_struct *vma,
 453			     unsigned long addr, unsigned long end,
 454			     struct zap_details *details);
 455void zap_page_range_single_batched(struct mmu_gather *tlb,
 456		struct vm_area_struct *vma, unsigned long addr,
 457		unsigned long size, struct zap_details *details);
 458int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio,
 459			   gfp_t gfp);
 460
 461void page_cache_ra_order(struct readahead_control *, struct file_ra_state *);
 462void force_page_cache_ra(struct readahead_control *, unsigned long nr);
 463static inline void force_page_cache_readahead(struct address_space *mapping,
 464		struct file *file, pgoff_t index, unsigned long nr_to_read)
 465{
 466	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, index);
 467	force_page_cache_ra(&ractl, nr_to_read);
 468}
 469
 470unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
 471		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
 472unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
 473		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices);
 474int truncate_inode_folio(struct address_space *mapping, struct folio *folio);
 475bool truncate_inode_partial_folio(struct folio *folio, loff_t start,
 476		loff_t end);
 477long mapping_evict_folio(struct address_space *mapping, struct folio *folio);
 478unsigned long mapping_try_invalidate(struct address_space *mapping,
 479		pgoff_t start, pgoff_t end, unsigned long *nr_failed);
 480
 481/**
 482 * folio_evictable - Test whether a folio is evictable.
 483 * @folio: The folio to test.
 484 *
 485 * Test whether @folio is evictable -- i.e., should be placed on
 486 * active/inactive lists vs unevictable list.
 487 *
 488 * Reasons folio might not be evictable:
 489 * 1. folio's mapping marked unevictable
 490 * 2. One of the pages in the folio is part of an mlocked VMA
 491 */
 492static inline bool folio_evictable(struct folio *folio)
 493{
 494	bool ret;
 495
 496	/* Prevent address_space of inode and swap cache from being freed */
 497	rcu_read_lock();
 498	ret = !mapping_unevictable(folio_mapping(folio)) &&
 499			!folio_test_mlocked(folio);
 500	rcu_read_unlock();
 501	return ret;
 502}
 503
 504/*
 505 * Turn a non-refcounted page (->_refcount == 0) into refcounted with
 506 * a count of one.
 507 */
 508static inline void set_page_refcounted(struct page *page)
 509{
 510	VM_BUG_ON_PAGE(PageTail(page), page);
 511	VM_BUG_ON_PAGE(page_ref_count(page), page);
 512	set_page_count(page, 1);
 513}
 514
 515/*
 516 * Return true if a folio needs ->release_folio() calling upon it.
 517 */
 518static inline bool folio_needs_release(struct folio *folio)
 519{
 520	struct address_space *mapping = folio_mapping(folio);
 521
 522	return folio_has_private(folio) ||
 523		(mapping && mapping_release_always(mapping));
 524}
 525
 526extern unsigned long highest_memmap_pfn;
 527
 528/*
 529 * Maximum number of reclaim retries without progress before the OOM
 530 * killer is consider the only way forward.
 531 */
 532#define MAX_RECLAIM_RETRIES 16
 533
 534/*
 535 * in mm/vmscan.c:
 536 */
 537bool folio_isolate_lru(struct folio *folio);
 538void folio_putback_lru(struct folio *folio);
 539extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason);
 540int user_proactive_reclaim(char *buf,
 541			   struct mem_cgroup *memcg, pg_data_t *pgdat);
 542
 543/*
 544 * in mm/rmap.c:
 545 */
 546pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
 547
 548/*
 549 * in mm/page_alloc.c
 550 */
 551#define K(x) ((x) << (PAGE_SHIFT-10))
 552
 553extern char * const zone_names[MAX_NR_ZONES];
 554
 555/* perform sanity checks on struct pages being allocated or freed */
 556DECLARE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
 557
 558extern int min_free_kbytes;
 559extern int defrag_mode;
 560
 561void setup_per_zone_wmarks(void);
 562void calculate_min_free_kbytes(void);
 563int __meminit init_per_zone_wmark_min(void);
 564void page_alloc_sysctl_init(void);
 565
 566/*
 567 * Structure for holding the mostly immutable allocation parameters passed
 568 * between functions involved in allocations, including the alloc_pages*
 569 * family of functions.
 570 *
 571 * nodemask, migratetype and highest_zoneidx are initialized only once in
 572 * __alloc_pages() and then never change.
 573 *
 574 * zonelist, preferred_zone and highest_zoneidx are set first in
 575 * __alloc_pages() for the fast path, and might be later changed
 576 * in __alloc_pages_slowpath(). All other functions pass the whole structure
 577 * by a const pointer.
 578 */
 579struct alloc_context {
 580	struct zonelist *zonelist;
 581	nodemask_t *nodemask;
 582	struct zoneref *preferred_zoneref;
 583	int migratetype;
 584
 585	/*
 586	 * highest_zoneidx represents highest usable zone index of
 587	 * the allocation request. Due to the nature of the zone,
 588	 * memory on lower zone than the highest_zoneidx will be
 589	 * protected by lowmem_reserve[highest_zoneidx].
 590	 *
 591	 * highest_zoneidx is also used by reclaim/compaction to limit
 592	 * the target zone since higher zone than this index cannot be
 593	 * usable for this allocation request.
 594	 */
 595	enum zone_type highest_zoneidx;
 596	bool spread_dirty_pages;
 597};
 598
 599/*
 600 * This function returns the order of a free page in the buddy system. In
 601 * general, page_zone(page)->lock must be held by the caller to prevent the
 602 * page from being allocated in parallel and returning garbage as the order.
 603 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
 604 * page cannot be allocated or merged in parallel. Alternatively, it must
 605 * handle invalid values gracefully, and use buddy_order_unsafe() below.
 606 */
 607static inline unsigned int buddy_order(struct page *page)
 608{
 609	/* PageBuddy() must be checked by the caller */
 610	return page_private(page);
 611}
 612
 613/*
 614 * Like buddy_order(), but for callers who cannot afford to hold the zone lock.
 615 * PageBuddy() should be checked first by the caller to minimize race window,
 616 * and invalid values must be handled gracefully.
 617 *
 618 * READ_ONCE is used so that if the caller assigns the result into a local
 619 * variable and e.g. tests it for valid range before using, the compiler cannot
 620 * decide to remove the variable and inline the page_private(page) multiple
 621 * times, potentially observing different values in the tests and the actual
 622 * use of the result.
 623 */
 624#define buddy_order_unsafe(page)	READ_ONCE(page_private(page))
 625
 626/*
 627 * This function checks whether a page is free && is the buddy
 628 * we can coalesce a page and its buddy if
 629 * (a) the buddy is not in a hole (check before calling!) &&
 630 * (b) the buddy is in the buddy system &&
 631 * (c) a page and its buddy have the same order &&
 632 * (d) a page and its buddy are in the same zone.
 633 *
 634 * For recording whether a page is in the buddy system, we set PageBuddy.
 635 * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
 636 *
 637 * For recording page's order, we use page_private(page).
 638 */
 639static inline bool page_is_buddy(struct page *page, struct page *buddy,
 640				 unsigned int order)
 641{
 642	if (!page_is_guard(buddy) && !PageBuddy(buddy))
 643		return false;
 644
 645	if (buddy_order(buddy) != order)
 646		return false;
 647
 648	/*
 649	 * zone check is done late to avoid uselessly calculating
 650	 * zone/node ids for pages that could never merge.
 651	 */
 652	if (page_zone_id(page) != page_zone_id(buddy))
 653		return false;
 654
 655	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
 656
 657	return true;
 658}
 659
 660/*
 661 * Locate the struct page for both the matching buddy in our
 662 * pair (buddy1) and the combined O(n+1) page they form (page).
 663 *
 664 * 1) Any buddy B1 will have an order O twin B2 which satisfies
 665 * the following equation:
 666 *     B2 = B1 ^ (1 << O)
 667 * For example, if the starting buddy (buddy2) is #8 its order
 668 * 1 buddy is #10:
 669 *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 670 *
 671 * 2) Any buddy B will have an order O+1 parent P which
 672 * satisfies the following equation:
 673 *     P = B & ~(1 << O)
 674 *
 675 * Assumption: *_mem_map is contiguous at least up to MAX_PAGE_ORDER
 676 */
 677static inline unsigned long
 678__find_buddy_pfn(unsigned long page_pfn, unsigned int order)
 679{
 680	return page_pfn ^ (1 << order);
 681}
 682
 683/*
 684 * Find the buddy of @page and validate it.
 685 * @page: The input page
 686 * @pfn: The pfn of the page, it saves a call to page_to_pfn() when the
 687 *       function is used in the performance-critical __free_one_page().
 688 * @order: The order of the page
 689 * @buddy_pfn: The output pointer to the buddy pfn, it also saves a call to
 690 *             page_to_pfn().
 691 *
 692 * The found buddy can be a non PageBuddy, out of @page's zone, or its order is
 693 * not the same as @page. The validation is necessary before use it.
 694 *
 695 * Return: the found buddy page or NULL if not found.
 696 */
 697static inline struct page *find_buddy_page_pfn(struct page *page,
 698			unsigned long pfn, unsigned int order, unsigned long *buddy_pfn)
 699{
 700	unsigned long __buddy_pfn = __find_buddy_pfn(pfn, order);
 701	struct page *buddy;
 702
 703	buddy = page + (__buddy_pfn - pfn);
 704	if (buddy_pfn)
 705		*buddy_pfn = __buddy_pfn;
 706
 707	if (page_is_buddy(page, buddy, order))
 708		return buddy;
 709	return NULL;
 710}
 711
 712extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
 713				unsigned long end_pfn, struct zone *zone);
 714
 715static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
 716				unsigned long end_pfn, struct zone *zone)
 717{
 718	if (zone->contiguous)
 719		return pfn_to_page(start_pfn);
 720
 721	return __pageblock_pfn_to_page(start_pfn, end_pfn, zone);
 722}
 723
 724void set_zone_contiguous(struct zone *zone);
 725bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
 726			   unsigned long nr_pages);
 727
 728static inline void clear_zone_contiguous(struct zone *zone)
 729{
 730	zone->contiguous = false;
 731}
 732
 733extern int __isolate_free_page(struct page *page, unsigned int order);
 734extern void __putback_isolated_page(struct page *page, unsigned int order,
 735				    int mt);
 736extern void memblock_free_pages(struct page *page, unsigned long pfn,
 737					unsigned int order);
 738extern void __free_pages_core(struct page *page, unsigned int order,
 739		enum meminit_context context);
 740
 741/*
 742 * This will have no effect, other than possibly generating a warning, if the
 743 * caller passes in a non-large folio.
 744 */
 745static inline void folio_set_order(struct folio *folio, unsigned int order)
 746{
 747	if (WARN_ON_ONCE(!order || !folio_test_large(folio)))
 748		return;
 749	VM_WARN_ON_ONCE(order > MAX_FOLIO_ORDER);
 750
 751	folio->_flags_1 = (folio->_flags_1 & ~0xffUL) | order;
 752#ifdef NR_PAGES_IN_LARGE_FOLIO
 753	folio->_nr_pages = 1U << order;
 754#endif
 755}
 756
 757bool __folio_unqueue_deferred_split(struct folio *folio);
 758static inline bool folio_unqueue_deferred_split(struct folio *folio)
 759{
 760	if (folio_order(folio) <= 1 || !folio_test_large_rmappable(folio))
 761		return false;
 762
 763	/*
 764	 * At this point, there is no one trying to add the folio to
 765	 * deferred_list. If folio is not in deferred_list, it's safe
 766	 * to check without acquiring the split_queue_lock.
 767	 */
 768	if (data_race(list_empty(&folio->_deferred_list)))
 769		return false;
 770
 771	return __folio_unqueue_deferred_split(folio);
 772}
 773
 774static inline struct folio *page_rmappable_folio(struct page *page)
 775{
 776	struct folio *folio = (struct folio *)page;
 777
 778	if (folio && folio_test_large(folio))
 779		folio_set_large_rmappable(folio);
 780	return folio;
 781}
 782
 783static inline void prep_compound_head(struct page *page, unsigned int order)
 784{
 785	struct folio *folio = (struct folio *)page;
 786
 787	folio_set_order(folio, order);
 788	atomic_set(&folio->_large_mapcount, -1);
 789	if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
 790		atomic_set(&folio->_nr_pages_mapped, 0);
 791	if (IS_ENABLED(CONFIG_MM_ID)) {
 792		folio->_mm_ids = 0;
 793		folio->_mm_id_mapcount[0] = -1;
 794		folio->_mm_id_mapcount[1] = -1;
 795	}
 796	if (IS_ENABLED(CONFIG_64BIT) || order > 1) {
 797		atomic_set(&folio->_pincount, 0);
 798		atomic_set(&folio->_entire_mapcount, -1);
 799	}
 800	if (order > 1)
 801		INIT_LIST_HEAD(&folio->_deferred_list);
 802}
 803
 804static inline void prep_compound_tail(struct page *head, int tail_idx)
 805{
 806	struct page *p = head + tail_idx;
 807
 808	p->mapping = TAIL_MAPPING;
 809	set_compound_head(p, head);
 810	set_page_private(p, 0);
 811}
 812
 813void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags);
 814extern bool free_pages_prepare(struct page *page, unsigned int order);
 815
 816extern int user_min_free_kbytes;
 817
 818struct page *__alloc_frozen_pages_noprof(gfp_t, unsigned int order, int nid,
 819		nodemask_t *);
 820#define __alloc_frozen_pages(...) \
 821	alloc_hooks(__alloc_frozen_pages_noprof(__VA_ARGS__))
 822void free_frozen_pages(struct page *page, unsigned int order);
 823void free_unref_folios(struct folio_batch *fbatch);
 824
 825#ifdef CONFIG_NUMA
 826struct page *alloc_frozen_pages_noprof(gfp_t, unsigned int order);
 827#else
 828static inline struct page *alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order)
 829{
 830	return __alloc_frozen_pages_noprof(gfp, order, numa_node_id(), NULL);
 831}
 832#endif
 833
 834#define alloc_frozen_pages(...) \
 835	alloc_hooks(alloc_frozen_pages_noprof(__VA_ARGS__))
 836
 837struct page *alloc_frozen_pages_nolock_noprof(gfp_t gfp_flags, int nid, unsigned int order);
 838#define alloc_frozen_pages_nolock(...) \
 839	alloc_hooks(alloc_frozen_pages_nolock_noprof(__VA_ARGS__))
 840
 841extern void zone_pcp_reset(struct zone *zone);
 842extern void zone_pcp_disable(struct zone *zone);
 843extern void zone_pcp_enable(struct zone *zone);
 844extern void zone_pcp_init(struct zone *zone);
 845
 846extern void *memmap_alloc(phys_addr_t size, phys_addr_t align,
 847			  phys_addr_t min_addr,
 848			  int nid, bool exact_nid);
 849
 850void memmap_init_range(unsigned long, int, unsigned long, unsigned long,
 851		unsigned long, enum meminit_context, struct vmem_altmap *, int,
 852		bool);
 853
 854#if defined CONFIG_COMPACTION || defined CONFIG_CMA
 855
 856/*
 857 * in mm/compaction.c
 858 */
 859/*
 860 * compact_control is used to track pages being migrated and the free pages
 861 * they are being migrated to during memory compaction. The free_pfn starts
 862 * at the end of a zone and migrate_pfn begins at the start. Movable pages
 863 * are moved to the end of a zone during a compaction run and the run
 864 * completes when free_pfn <= migrate_pfn
 865 */
 866struct compact_control {
 867	struct list_head freepages[NR_PAGE_ORDERS];	/* List of free pages to migrate to */
 868	struct list_head migratepages;	/* List of pages being migrated */
 869	unsigned int nr_freepages;	/* Number of isolated free pages */
 870	unsigned int nr_migratepages;	/* Number of pages to migrate */
 871	unsigned long free_pfn;		/* isolate_freepages search base */
 872	/*
 873	 * Acts as an in/out parameter to page isolation for migration.
 874	 * isolate_migratepages uses it as a search base.
 875	 * isolate_migratepages_block will update the value to the next pfn
 876	 * after the last isolated one.
 877	 */
 878	unsigned long migrate_pfn;
 879	unsigned long fast_start_pfn;	/* a pfn to start linear scan from */
 880	struct zone *zone;
 881	unsigned long total_migrate_scanned;
 882	unsigned long total_free_scanned;
 883	unsigned short fast_search_fail;/* failures to use free list searches */
 884	short search_order;		/* order to start a fast search at */
 885	const gfp_t gfp_mask;		/* gfp mask of a direct compactor */
 886	int order;			/* order a direct compactor needs */
 887	int migratetype;		/* migratetype of direct compactor */
 888	const unsigned int alloc_flags;	/* alloc flags of a direct compactor */
 889	const int highest_zoneidx;	/* zone index of a direct compactor */
 890	enum migrate_mode mode;		/* Async or sync migration mode */
 891	bool ignore_skip_hint;		/* Scan blocks even if marked skip */
 892	bool no_set_skip_hint;		/* Don't mark blocks for skipping */
 893	bool ignore_block_suitable;	/* Scan blocks considered unsuitable */
 894	bool direct_compaction;		/* False from kcompactd or /proc/... */
 895	bool proactive_compaction;	/* kcompactd proactive compaction */
 896	bool whole_zone;		/* Whole zone should/has been scanned */
 897	bool contended;			/* Signal lock contention */
 898	bool finish_pageblock;		/* Scan the remainder of a pageblock. Used
 899					 * when there are potentially transient
 900					 * isolation or migration failures to
 901					 * ensure forward progress.
 902					 */
 903	bool alloc_contig;		/* alloc_contig_range allocation */
 904};
 905
 906/*
 907 * Used in direct compaction when a page should be taken from the freelists
 908 * immediately when one is created during the free path.
 909 */
 910struct capture_control {
 911	struct compact_control *cc;
 912	struct page *page;
 913};
 914
 915unsigned long
 916isolate_freepages_range(struct compact_control *cc,
 917			unsigned long start_pfn, unsigned long end_pfn);
 918int
 919isolate_migratepages_range(struct compact_control *cc,
 920			   unsigned long low_pfn, unsigned long end_pfn);
 921
 922/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
 923void init_cma_reserved_pageblock(struct page *page);
 924
 925#endif /* CONFIG_COMPACTION || CONFIG_CMA */
 926
 927struct cma;
 928
 929#ifdef CONFIG_CMA
 930void *cma_reserve_early(struct cma *cma, unsigned long size);
 931void init_cma_pageblock(struct page *page);
 932#else
 933static inline void *cma_reserve_early(struct cma *cma, unsigned long size)
 934{
 935	return NULL;
 936}
 937static inline void init_cma_pageblock(struct page *page)
 938{
 939}
 940#endif
 941
 942
 943int find_suitable_fallback(struct free_area *area, unsigned int order,
 944			   int migratetype, bool claimable);
 945
 946static inline bool free_area_empty(struct free_area *area, int migratetype)
 947{
 948	return list_empty(&area->free_list[migratetype]);
 949}
 950
 951/* mm/util.c */
 952struct anon_vma *folio_anon_vma(const struct folio *folio);
 953
 954#ifdef CONFIG_MMU
 955void unmap_mapping_folio(struct folio *folio);
 956extern long populate_vma_page_range(struct vm_area_struct *vma,
 957		unsigned long start, unsigned long end, int *locked);
 958extern long faultin_page_range(struct mm_struct *mm, unsigned long start,
 959		unsigned long end, bool write, int *locked);
 960bool mlock_future_ok(const struct mm_struct *mm, vm_flags_t vm_flags,
 961		unsigned long bytes);
 962
 963/*
 964 * NOTE: This function can't tell whether the folio is "fully mapped" in the
 965 * range.
 966 * "fully mapped" means all the pages of folio is associated with the page
 967 * table of range while this function just check whether the folio range is
 968 * within the range [start, end). Function caller needs to do page table
 969 * check if it cares about the page table association.
 970 *
 971 * Typical usage (like mlock or madvise) is:
 972 * Caller knows at least 1 page of folio is associated with page table of VMA
 973 * and the range [start, end) is intersect with the VMA range. Caller wants
 974 * to know whether the folio is fully associated with the range. It calls
 975 * this function to check whether the folio is in the range first. Then checks
 976 * the page table to know whether the folio is fully mapped to the range.
 977 */
 978static inline bool
 979folio_within_range(struct folio *folio, struct vm_area_struct *vma,
 980		unsigned long start, unsigned long end)
 981{
 982	pgoff_t pgoff, addr;
 983	unsigned long vma_pglen = vma_pages(vma);
 984
 985	VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio);
 986	if (start > end)
 987		return false;
 988
 989	if (start < vma->vm_start)
 990		start = vma->vm_start;
 991
 992	if (end > vma->vm_end)
 993		end = vma->vm_end;
 994
 995	pgoff = folio_pgoff(folio);
 996
 997	/* if folio start address is not in vma range */
 998	if (!in_range(pgoff, vma->vm_pgoff, vma_pglen))
 999		return false;
1000
1001	addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1002
1003	return !(addr < start || end - addr < folio_size(folio));
1004}
1005
1006static inline bool
1007folio_within_vma(struct folio *folio, struct vm_area_struct *vma)
1008{
1009	return folio_within_range(folio, vma, vma->vm_start, vma->vm_end);
1010}
1011
1012/*
1013 * mlock_vma_folio() and munlock_vma_folio():
1014 * should be called with vma's mmap_lock held for read or write,
1015 * under page table lock for the pte/pmd being added or removed.
1016 *
1017 * mlock is usually called at the end of folio_add_*_rmap_*(), munlock at
1018 * the end of folio_remove_rmap_*(); but new anon folios are managed by
1019 * folio_add_lru_vma() calling mlock_new_folio().
1020 */
1021void mlock_folio(struct folio *folio);
1022static inline void mlock_vma_folio(struct folio *folio,
1023				struct vm_area_struct *vma)
1024{
1025	/*
1026	 * The VM_SPECIAL check here serves two purposes.
1027	 * 1) VM_IO check prevents migration from double-counting during mlock.
1028	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
1029	 *    is never left set on a VM_SPECIAL vma, there is an interval while
1030	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
1031	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
1032	 */
1033	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED))
1034		mlock_folio(folio);
1035}
1036
1037void munlock_folio(struct folio *folio);
1038static inline void munlock_vma_folio(struct folio *folio,
1039					struct vm_area_struct *vma)
1040{
1041	/*
1042	 * munlock if the function is called. Ideally, we should only
1043	 * do munlock if any page of folio is unmapped from VMA and
1044	 * cause folio not fully mapped to VMA.
1045	 *
1046	 * But it's not easy to confirm that's the situation. So we
1047	 * always munlock the folio and page reclaim will correct it
1048	 * if it's wrong.
1049	 */
1050	if (unlikely(vma->vm_flags & VM_LOCKED))
1051		munlock_folio(folio);
1052}
1053
1054void mlock_new_folio(struct folio *folio);
1055bool need_mlock_drain(int cpu);
1056void mlock_drain_local(void);
1057void mlock_drain_remote(int cpu);
1058
1059extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
1060
1061/**
1062 * vma_address - Find the virtual address a page range is mapped at
1063 * @vma: The vma which maps this object.
1064 * @pgoff: The page offset within its object.
1065 * @nr_pages: The number of pages to consider.
1066 *
1067 * If any page in this range is mapped by this VMA, return the first address
1068 * where any of these pages appear.  Otherwise, return -EFAULT.
1069 */
1070static inline unsigned long vma_address(const struct vm_area_struct *vma,
1071		pgoff_t pgoff, unsigned long nr_pages)
1072{
1073	unsigned long address;
1074
1075	if (pgoff >= vma->vm_pgoff) {
1076		address = vma->vm_start +
1077			((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1078		/* Check for address beyond vma (or wrapped through 0?) */
1079		if (address < vma->vm_start || address >= vma->vm_end)
1080			address = -EFAULT;
1081	} else if (pgoff + nr_pages - 1 >= vma->vm_pgoff) {
1082		/* Test above avoids possibility of wrap to 0 on 32-bit */
1083		address = vma->vm_start;
1084	} else {
1085		address = -EFAULT;
1086	}
1087	return address;
1088}
1089
1090/*
1091 * Then at what user virtual address will none of the range be found in vma?
1092 * Assumes that vma_address() already returned a good starting address.
1093 */
1094static inline unsigned long vma_address_end(struct page_vma_mapped_walk *pvmw)
1095{
1096	struct vm_area_struct *vma = pvmw->vma;
1097	pgoff_t pgoff;
1098	unsigned long address;
1099
1100	/* Common case, plus ->pgoff is invalid for KSM */
1101	if (pvmw->nr_pages == 1)
1102		return pvmw->address + PAGE_SIZE;
1103
1104	pgoff = pvmw->pgoff + pvmw->nr_pages;
1105	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
1106	/* Check for address beyond vma (or wrapped through 0?) */
1107	if (address < vma->vm_start || address > vma->vm_end)
1108		address = vma->vm_end;
1109	return address;
1110}
1111
1112static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
1113						    struct file *fpin)
1114{
1115	int flags = vmf->flags;
1116
1117	if (fpin)
1118		return fpin;
1119
1120	/*
1121	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
1122	 * anything, so we only pin the file and drop the mmap_lock if only
1123	 * FAULT_FLAG_ALLOW_RETRY is set, while this is the first attempt.
1124	 */
1125	if (fault_flag_allow_retry_first(flags) &&
1126	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
1127		fpin = get_file(vmf->vma->vm_file);
1128		release_fault_lock(vmf);
1129	}
1130	return fpin;
1131}
1132#else /* !CONFIG_MMU */
1133static inline void unmap_mapping_folio(struct folio *folio) { }
1134static inline void mlock_new_folio(struct folio *folio) { }
1135static inline bool need_mlock_drain(int cpu) { return false; }
1136static inline void mlock_drain_local(void) { }
1137static inline void mlock_drain_remote(int cpu) { }
1138static inline void vunmap_range_noflush(unsigned long start, unsigned long end)
1139{
1140}
1141#endif /* !CONFIG_MMU */
1142
1143/* Memory initialisation debug and verification */
1144#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1145DECLARE_STATIC_KEY_TRUE(deferred_pages);
1146
1147bool __init deferred_grow_zone(struct zone *zone, unsigned int order);
1148#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1149
1150void init_deferred_page(unsigned long pfn, int nid);
1151
1152enum mminit_level {
1153	MMINIT_WARNING,
1154	MMINIT_VERIFY,
1155	MMINIT_TRACE
1156};
1157
1158#ifdef CONFIG_DEBUG_MEMORY_INIT
1159
1160extern int mminit_loglevel;
1161
1162#define mminit_dprintk(level, prefix, fmt, arg...) \
1163do { \
1164	if (level < mminit_loglevel) { \
1165		if (level <= MMINIT_WARNING) \
1166			pr_warn("mminit::" prefix " " fmt, ##arg);	\
1167		else \
1168			printk(KERN_DEBUG "mminit::" prefix " " fmt, ##arg); \
1169	} \
1170} while (0)
1171
1172extern void mminit_verify_pageflags_layout(void);
1173extern void mminit_verify_zonelist(void);
1174#else
1175
1176static inline void mminit_dprintk(enum mminit_level level,
1177				const char *prefix, const char *fmt, ...)
1178{
1179}
1180
1181static inline void mminit_verify_pageflags_layout(void)
1182{
1183}
1184
1185static inline void mminit_verify_zonelist(void)
1186{
1187}
1188#endif /* CONFIG_DEBUG_MEMORY_INIT */
1189
1190#define NODE_RECLAIM_NOSCAN	-2
1191#define NODE_RECLAIM_FULL	-1
1192#define NODE_RECLAIM_SOME	0
1193#define NODE_RECLAIM_SUCCESS	1
1194
1195#ifdef CONFIG_NUMA
1196extern int node_reclaim_mode;
1197
1198extern int node_reclaim(struct pglist_data *, gfp_t, unsigned int);
1199extern int find_next_best_node(int node, nodemask_t *used_node_mask);
1200#else
1201#define node_reclaim_mode 0
1202
1203static inline int node_reclaim(struct pglist_data *pgdat, gfp_t mask,
1204				unsigned int order)
1205{
1206	return NODE_RECLAIM_NOSCAN;
1207}
1208static inline int find_next_best_node(int node, nodemask_t *used_node_mask)
1209{
1210	return NUMA_NO_NODE;
1211}
1212#endif
1213
1214static inline bool node_reclaim_enabled(void)
1215{
1216	/* Is any node_reclaim_mode bit set? */
1217	return node_reclaim_mode & (RECLAIM_ZONE|RECLAIM_WRITE|RECLAIM_UNMAP);
1218}
1219
1220/*
1221 * mm/memory-failure.c
1222 */
1223#ifdef CONFIG_MEMORY_FAILURE
1224int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill);
1225void shake_folio(struct folio *folio);
1226typedef int hwpoison_filter_func_t(struct page *p);
1227void hwpoison_filter_register(hwpoison_filter_func_t *filter);
1228void hwpoison_filter_unregister(void);
1229
1230#define MAGIC_HWPOISON	0x48575053U	/* HWPS */
1231void SetPageHWPoisonTakenOff(struct page *page);
1232void ClearPageHWPoisonTakenOff(struct page *page);
1233bool take_page_off_buddy(struct page *page);
1234bool put_page_back_buddy(struct page *page);
1235struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
1236void add_to_kill_ksm(struct task_struct *tsk, const struct page *p,
1237		     struct vm_area_struct *vma, struct list_head *to_kill,
1238		     unsigned long ksm_addr);
1239unsigned long page_mapped_in_vma(const struct page *page,
1240		struct vm_area_struct *vma);
1241
1242#else
1243static inline int unmap_poisoned_folio(struct folio *folio, unsigned long pfn, bool must_kill)
1244{
1245	return -EBUSY;
1246}
1247#endif
1248
1249extern unsigned long  __must_check vm_mmap_pgoff(struct file *, unsigned long,
1250        unsigned long, unsigned long,
1251        unsigned long, unsigned long);
1252
1253extern void set_pageblock_order(void);
1254unsigned long reclaim_pages(struct list_head *folio_list);
1255unsigned int reclaim_clean_pages_from_list(struct zone *zone,
1256					    struct list_head *folio_list);
1257/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1258#define ALLOC_WMARK_MIN		WMARK_MIN
1259#define ALLOC_WMARK_LOW		WMARK_LOW
1260#define ALLOC_WMARK_HIGH	WMARK_HIGH
1261#define ALLOC_NO_WATERMARKS	0x04 /* don't check watermarks at all */
1262
1263/* Mask to get the watermark bits */
1264#define ALLOC_WMARK_MASK	(ALLOC_NO_WATERMARKS-1)
1265
1266/*
1267 * Only MMU archs have async oom victim reclaim - aka oom_reaper so we
1268 * cannot assume a reduced access to memory reserves is sufficient for
1269 * !MMU
1270 */
1271#ifdef CONFIG_MMU
1272#define ALLOC_OOM		0x08
1273#else
1274#define ALLOC_OOM		ALLOC_NO_WATERMARKS
1275#endif
1276
1277#define ALLOC_NON_BLOCK		 0x10 /* Caller cannot block. Allow access
1278				       * to 25% of the min watermark or
1279				       * 62.5% if __GFP_HIGH is set.
1280				       */
1281#define ALLOC_MIN_RESERVE	 0x20 /* __GFP_HIGH set. Allow access to 50%
1282				       * of the min watermark.
1283				       */
1284#define ALLOC_CPUSET		 0x40 /* check for correct cpuset */
1285#define ALLOC_CMA		 0x80 /* allow allocations from CMA areas */
1286#ifdef CONFIG_ZONE_DMA32
1287#define ALLOC_NOFRAGMENT	0x100 /* avoid mixing pageblock types */
1288#else
1289#define ALLOC_NOFRAGMENT	  0x0
1290#endif
1291#define ALLOC_HIGHATOMIC	0x200 /* Allows access to MIGRATE_HIGHATOMIC */
1292#define ALLOC_TRYLOCK		0x400 /* Only use spin_trylock in allocation path */
1293#define ALLOC_KSWAPD		0x800 /* allow waking of kswapd, __GFP_KSWAPD_RECLAIM set */
1294
1295/* Flags that allow allocations below the min watermark. */
1296#define ALLOC_RESERVES (ALLOC_NON_BLOCK|ALLOC_MIN_RESERVE|ALLOC_HIGHATOMIC|ALLOC_OOM)
1297
1298enum ttu_flags;
1299struct tlbflush_unmap_batch;
1300
1301
1302/*
1303 * only for MM internal work items which do not depend on
1304 * any allocations or locks which might depend on allocations
1305 */
1306extern struct workqueue_struct *mm_percpu_wq;
1307
1308#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
1309void try_to_unmap_flush(void);
1310void try_to_unmap_flush_dirty(void);
1311void flush_tlb_batched_pending(struct mm_struct *mm);
1312#else
1313static inline void try_to_unmap_flush(void)
1314{
1315}
1316static inline void try_to_unmap_flush_dirty(void)
1317{
1318}
1319static inline void flush_tlb_batched_pending(struct mm_struct *mm)
1320{
1321}
1322#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
1323
1324extern const struct trace_print_flags pageflag_names[];
1325extern const struct trace_print_flags vmaflag_names[];
1326extern const struct trace_print_flags gfpflag_names[];
1327
1328void setup_zone_pageset(struct zone *zone);
1329
1330struct migration_target_control {
1331	int nid;		/* preferred node id */
1332	nodemask_t *nmask;
1333	gfp_t gfp_mask;
1334	enum migrate_reason reason;
1335};
1336
1337/*
1338 * mm/filemap.c
1339 */
1340size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
1341			      struct folio *folio, loff_t fpos, size_t size);
1342
1343/*
1344 * mm/vmalloc.c
1345 */
1346#ifdef CONFIG_MMU
1347void __init vmalloc_init(void);
1348int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1349	pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask);
1350unsigned int get_vm_area_page_order(struct vm_struct *vm);
1351#else
1352static inline void vmalloc_init(void)
1353{
1354}
1355
1356static inline
1357int __must_check vmap_pages_range_noflush(unsigned long addr, unsigned long end,
1358	pgprot_t prot, struct page **pages, unsigned int page_shift, gfp_t gfp_mask)
1359{
1360	return -EINVAL;
1361}
1362#endif
1363
1364int __must_check __vmap_pages_range_noflush(unsigned long addr,
1365			       unsigned long end, pgprot_t prot,
1366			       struct page **pages, unsigned int page_shift);
1367
1368void vunmap_range_noflush(unsigned long start, unsigned long end);
1369
1370void __vunmap_range_noflush(unsigned long start, unsigned long end);
1371
1372static inline bool vma_is_single_threaded_private(struct vm_area_struct *vma)
1373{
1374	if (vma->vm_flags & VM_SHARED)
1375		return false;
1376
1377	return atomic_read(&vma->vm_mm->mm_users) == 1;
1378}
1379
1380#ifdef CONFIG_NUMA_BALANCING
1381bool folio_can_map_prot_numa(struct folio *folio, struct vm_area_struct *vma,
1382		bool is_private_single_threaded);
1383
1384#else
1385static inline bool folio_can_map_prot_numa(struct folio *folio,
1386		struct vm_area_struct *vma, bool is_private_single_threaded)
1387{
1388	return false;
1389}
1390#endif
1391
1392int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
1393		      unsigned long addr, int *flags, bool writable,
1394		      int *last_cpupid);
1395
1396void free_zone_device_folio(struct folio *folio);
1397int migrate_device_coherent_folio(struct folio *folio);
1398
1399struct vm_struct *__get_vm_area_node(unsigned long size,
1400				     unsigned long align, unsigned long shift,
1401				     unsigned long vm_flags, unsigned long start,
1402				     unsigned long end, int node, gfp_t gfp_mask,
1403				     const void *caller);
1404
1405/*
1406 * mm/gup.c
1407 */
1408int __must_check try_grab_folio(struct folio *folio, int refs,
1409				unsigned int flags);
1410
1411/*
1412 * mm/huge_memory.c
1413 */
1414void touch_pud(struct vm_area_struct *vma, unsigned long addr,
1415	       pud_t *pud, bool write);
1416bool touch_pmd(struct vm_area_struct *vma, unsigned long addr,
1417	       pmd_t *pmd, bool write);
1418
1419/*
1420 * Parses a string with mem suffixes into its order. Useful to parse kernel
1421 * parameters.
1422 */
1423static inline int get_order_from_str(const char *size_str,
1424				     unsigned long valid_orders)
1425{
1426	unsigned long size;
1427	char *endptr;
1428	int order;
1429
1430	size = memparse(size_str, &endptr);
1431
1432	if (!is_power_of_2(size))
1433		return -EINVAL;
1434	order = get_order(size);
1435	if (BIT(order) & ~valid_orders)
1436		return -EINVAL;
1437
1438	return order;
1439}
1440
1441enum {
1442	/* mark page accessed */
1443	FOLL_TOUCH = 1 << 16,
1444	/* a retry, previous pass started an IO */
1445	FOLL_TRIED = 1 << 17,
1446	/* we are working on non-current tsk/mm */
1447	FOLL_REMOTE = 1 << 18,
1448	/* pages must be released via unpin_user_page */
1449	FOLL_PIN = 1 << 19,
1450	/* gup_fast: prevent fall-back to slow gup */
1451	FOLL_FAST_ONLY = 1 << 20,
1452	/* allow unlocking the mmap lock */
1453	FOLL_UNLOCKABLE = 1 << 21,
1454	/* VMA lookup+checks compatible with MADV_POPULATE_(READ|WRITE) */
1455	FOLL_MADV_POPULATE = 1 << 22,
1456};
1457
1458#define INTERNAL_GUP_FLAGS (FOLL_TOUCH | FOLL_TRIED | FOLL_REMOTE | FOLL_PIN | \
1459			    FOLL_FAST_ONLY | FOLL_UNLOCKABLE | \
1460			    FOLL_MADV_POPULATE)
1461
1462/*
1463 * Indicates for which pages that are write-protected in the page table,
1464 * whether GUP has to trigger unsharing via FAULT_FLAG_UNSHARE such that the
1465 * GUP pin will remain consistent with the pages mapped into the page tables
1466 * of the MM.
1467 *
1468 * Temporary unmapping of PageAnonExclusive() pages or clearing of
1469 * PageAnonExclusive() has to protect against concurrent GUP:
1470 * * Ordinary GUP: Using the PT lock
1471 * * GUP-fast and fork(): mm->write_protect_seq
1472 * * GUP-fast and KSM or temporary unmapping (swap, migration): see
1473 *    folio_try_share_anon_rmap_*()
1474 *
1475 * Must be called with the (sub)page that's actually referenced via the
1476 * page table entry, which might not necessarily be the head page for a
1477 * PTE-mapped THP.
1478 *
1479 * If the vma is NULL, we're coming from the GUP-fast path and might have
1480 * to fallback to the slow path just to lookup the vma.
1481 */
1482static inline bool gup_must_unshare(struct vm_area_struct *vma,
1483				    unsigned int flags, struct page *page)
1484{
1485	/*
1486	 * FOLL_WRITE is implicitly handled correctly as the page table entry
1487	 * has to be writable -- and if it references (part of) an anonymous
1488	 * folio, that part is required to be marked exclusive.
1489	 */
1490	if ((flags & (FOLL_WRITE | FOLL_PIN)) != FOLL_PIN)
1491		return false;
1492	/*
1493	 * Note: PageAnon(page) is stable until the page is actually getting
1494	 * freed.
1495	 */
1496	if (!PageAnon(page)) {
1497		/*
1498		 * We only care about R/O long-term pining: R/O short-term
1499		 * pinning does not have the semantics to observe successive
1500		 * changes through the process page tables.
1501		 */
1502		if (!(flags & FOLL_LONGTERM))
1503			return false;
1504
1505		/* We really need the vma ... */
1506		if (!vma)
1507			return true;
1508
1509		/*
1510		 * ... because we only care about writable private ("COW")
1511		 * mappings where we have to break COW early.
1512		 */
1513		return is_cow_mapping(vma->vm_flags);
1514	}
1515
1516	/* Paired with a memory barrier in folio_try_share_anon_rmap_*(). */
1517	if (IS_ENABLED(CONFIG_HAVE_GUP_FAST))
1518		smp_rmb();
1519
1520	/*
1521	 * Note that KSM pages cannot be exclusive, and consequently,
1522	 * cannot get pinned.
1523	 */
1524	return !PageAnonExclusive(page);
1525}
1526
1527extern bool mirrored_kernelcore;
1528bool memblock_has_mirror(void);
1529void memblock_free_all(void);
1530
1531static __always_inline void vma_set_range(struct vm_area_struct *vma,
1532					  unsigned long start, unsigned long end,
1533					  pgoff_t pgoff)
1534{
1535	vma->vm_start = start;
1536	vma->vm_end = end;
1537	vma->vm_pgoff = pgoff;
1538}
1539
1540static inline bool vma_soft_dirty_enabled(struct vm_area_struct *vma)
1541{
1542	/*
1543	 * NOTE: we must check this before VM_SOFTDIRTY on soft-dirty
1544	 * enablements, because when without soft-dirty being compiled in,
1545	 * VM_SOFTDIRTY is defined as 0x0, then !(vm_flags & VM_SOFTDIRTY)
1546	 * will be constantly true.
1547	 */
1548	if (!pgtable_supports_soft_dirty())
1549		return false;
1550
1551	/*
1552	 * Soft-dirty is kind of special: its tracking is enabled when the
1553	 * vma flags not set.
1554	 */
1555	return !(vma->vm_flags & VM_SOFTDIRTY);
1556}
1557
1558static inline bool pmd_needs_soft_dirty_wp(struct vm_area_struct *vma, pmd_t pmd)
1559{
1560	return vma_soft_dirty_enabled(vma) && !pmd_soft_dirty(pmd);
1561}
1562
1563static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte)
1564{
1565	return vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte);
1566}
1567
1568void __meminit __init_single_page(struct page *page, unsigned long pfn,
1569				unsigned long zone, int nid);
1570void __meminit __init_page_from_nid(unsigned long pfn, int nid);
1571
1572/* shrinker related functions */
1573unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
1574			  int priority);
1575
1576int shmem_add_to_page_cache(struct folio *folio,
1577			    struct address_space *mapping,
1578			    pgoff_t index, void *expected, gfp_t gfp);
1579int shmem_inode_acct_blocks(struct inode *inode, long pages);
1580bool shmem_recalc_inode(struct inode *inode, long alloced, long swapped);
1581
1582#ifdef CONFIG_SHRINKER_DEBUG
1583static inline __printf(2, 0) int shrinker_debugfs_name_alloc(
1584			struct shrinker *shrinker, const char *fmt, va_list ap)
1585{
1586	shrinker->name = kvasprintf_const(GFP_KERNEL, fmt, ap);
1587
1588	return shrinker->name ? 0 : -ENOMEM;
1589}
1590
1591static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1592{
1593	kfree_const(shrinker->name);
1594	shrinker->name = NULL;
1595}
1596
1597extern int shrinker_debugfs_add(struct shrinker *shrinker);
1598extern struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1599					      int *debugfs_id);
1600extern void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1601				    int debugfs_id);
1602#else /* CONFIG_SHRINKER_DEBUG */
1603static inline int shrinker_debugfs_add(struct shrinker *shrinker)
1604{
1605	return 0;
1606}
1607static inline int shrinker_debugfs_name_alloc(struct shrinker *shrinker,
1608					      const char *fmt, va_list ap)
1609{
1610	return 0;
1611}
1612static inline void shrinker_debugfs_name_free(struct shrinker *shrinker)
1613{
1614}
1615static inline struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
1616						     int *debugfs_id)
1617{
1618	*debugfs_id = -1;
1619	return NULL;
1620}
1621static inline void shrinker_debugfs_remove(struct dentry *debugfs_entry,
1622					   int debugfs_id)
1623{
1624}
1625#endif /* CONFIG_SHRINKER_DEBUG */
1626
1627/* Only track the nodes of mappings with shadow entries */
1628void workingset_update_node(struct xa_node *node);
1629extern struct list_lru shadow_nodes;
1630#define mapping_set_update(xas, mapping) do {			\
1631	if (!dax_mapping(mapping) && !shmem_mapping(mapping)) {	\
1632		xas_set_update(xas, workingset_update_node);	\
1633		xas_set_lru(xas, &shadow_nodes);		\
1634	}							\
1635} while (0)
1636
1637/* mremap.c */
1638unsigned long move_page_tables(struct pagetable_move_control *pmc);
1639
1640#ifdef CONFIG_UNACCEPTED_MEMORY
1641void accept_page(struct page *page);
1642#else /* CONFIG_UNACCEPTED_MEMORY */
1643static inline void accept_page(struct page *page)
1644{
1645}
1646#endif /* CONFIG_UNACCEPTED_MEMORY */
1647
1648/* pagewalk.c */
1649int walk_page_range_mm_unsafe(struct mm_struct *mm, unsigned long start,
1650		unsigned long end, const struct mm_walk_ops *ops,
1651		void *private);
1652int walk_page_range_vma_unsafe(struct vm_area_struct *vma, unsigned long start,
1653		unsigned long end, const struct mm_walk_ops *ops,
1654		void *private);
1655int walk_page_range_debug(struct mm_struct *mm, unsigned long start,
1656			  unsigned long end, const struct mm_walk_ops *ops,
1657			  pgd_t *pgd, void *private);
1658
1659/* pt_reclaim.c */
1660bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval);
1661void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb,
1662	      pmd_t pmdval);
1663void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
1664		     struct mmu_gather *tlb);
1665
1666#ifdef CONFIG_PT_RECLAIM
1667bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1668			   struct zap_details *details);
1669#else
1670static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
1671					 struct zap_details *details)
1672{
1673	return false;
1674}
1675#endif /* CONFIG_PT_RECLAIM */
1676
1677void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
1678int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
1679
1680void remap_pfn_range_prepare(struct vm_area_desc *desc, unsigned long pfn);
1681int remap_pfn_range_complete(struct vm_area_struct *vma, unsigned long addr,
1682		unsigned long pfn, unsigned long size, pgprot_t pgprot);
1683
1684static inline void io_remap_pfn_range_prepare(struct vm_area_desc *desc,
1685		unsigned long orig_pfn, unsigned long size)
1686{
1687	const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size);
1688
1689	return remap_pfn_range_prepare(desc, pfn);
1690}
1691
1692static inline int io_remap_pfn_range_complete(struct vm_area_struct *vma,
1693		unsigned long addr, unsigned long orig_pfn, unsigned long size,
1694		pgprot_t orig_prot)
1695{
1696	const unsigned long pfn = io_remap_pfn_range_pfn(orig_pfn, size);
1697	const pgprot_t prot = pgprot_decrypted(orig_prot);
1698
1699	return remap_pfn_range_complete(vma, addr, pfn, size, prot);
1700}
1701
1702#endif	/* __MM_INTERNAL_H */