swapops.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef _LINUX_SWAPOPS_H
  3#define _LINUX_SWAPOPS_H
  4
  5#include <linux/radix-tree.h>
  6#include <linux/bug.h>
  7#include <linux/mm_types.h>
  8
  9#ifdef CONFIG_MMU
 10
 11#ifdef CONFIG_SWAP
 12#include <linux/swapfile.h>
 13#endif	/* CONFIG_SWAP */
 14
 15/*
 16 * swapcache pages are stored in the swapper_space radix tree.  We want to
 17 * get good packing density in that tree, so the index should be dense in
 18 * the low-order bits.
 19 *
 20 * We arrange the `type' and `offset' fields so that `type' is at the six
 21 * high-order bits of the swp_entry_t and `offset' is right-aligned in the
 22 * remaining bits.  Although `type' itself needs only five bits, we allow for
 23 * shmem/tmpfs to shift it all up a further one bit: see swp_to_radix_entry().
 24 *
 25 * swp_entry_t's are *never* stored anywhere in their arch-dependent format.
 26 */
 27#define SWP_TYPE_SHIFT	(BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
 28#define SWP_OFFSET_MASK	((1UL << SWP_TYPE_SHIFT) - 1)
 29
 30/*
 31 * Definitions only for PFN swap entries (see leafeant_has_pfn()).  To
 32 * store PFN, we only need SWP_PFN_BITS bits.  Each of the pfn swap entries
 33 * can use the extra bits to store other information besides PFN.
 34 */
 35#ifdef MAX_PHYSMEM_BITS
 36#define SWP_PFN_BITS		(MAX_PHYSMEM_BITS - PAGE_SHIFT)
 37#else  /* MAX_PHYSMEM_BITS */
 38#define SWP_PFN_BITS		min_t(int, \
 39				      sizeof(phys_addr_t) * 8 - PAGE_SHIFT, \
 40				      SWP_TYPE_SHIFT)
 41#endif	/* MAX_PHYSMEM_BITS */
 42#define SWP_PFN_MASK		(BIT(SWP_PFN_BITS) - 1)
 43
 44/**
 45 * Migration swap entry specific bitfield definitions.  Layout:
 46 *
 47 *   |----------+--------------------|
 48 *   | swp_type | swp_offset         |
 49 *   |----------+--------+-+-+-------|
 50 *   |          | resv   |D|A|  PFN  |
 51 *   |----------+--------+-+-+-------|
 52 *
 53 * @SWP_MIG_YOUNG_BIT: Whether the page used to have young bit set (bit A)
 54 * @SWP_MIG_DIRTY_BIT: Whether the page used to have dirty bit set (bit D)
 55 *
 56 * Note: A/D bits will be stored in migration entries iff there're enough
 57 * free bits in arch specific swp offset.  By default we'll ignore A/D bits
 58 * when migrating a page.  Please refer to migration_entry_supports_ad()
 59 * for more information.  If there're more bits besides PFN and A/D bits,
 60 * they should be reserved and always be zeros.
 61 */
 62#define SWP_MIG_YOUNG_BIT		(SWP_PFN_BITS)
 63#define SWP_MIG_DIRTY_BIT		(SWP_PFN_BITS + 1)
 64#define SWP_MIG_TOTAL_BITS		(SWP_PFN_BITS + 2)
 65
 66#define SWP_MIG_YOUNG			BIT(SWP_MIG_YOUNG_BIT)
 67#define SWP_MIG_DIRTY			BIT(SWP_MIG_DIRTY_BIT)
 68
 69/* Clear all flags but only keep swp_entry_t related information */
 70static inline pte_t pte_swp_clear_flags(pte_t pte)
 71{
 72	if (pte_swp_exclusive(pte))
 73		pte = pte_swp_clear_exclusive(pte);
 74	if (pte_swp_soft_dirty(pte))
 75		pte = pte_swp_clear_soft_dirty(pte);
 76	if (pte_swp_uffd_wp(pte))
 77		pte = pte_swp_clear_uffd_wp(pte);
 78	return pte;
 79}
 80
 81/*
 82 * Store a type+offset into a swp_entry_t in an arch-independent format
 83 */
 84static inline swp_entry_t swp_entry(unsigned long type, pgoff_t offset)
 85{
 86	swp_entry_t ret;
 87
 88	ret.val = (type << SWP_TYPE_SHIFT) | (offset & SWP_OFFSET_MASK);
 89	return ret;
 90}
 91
 92/*
 93 * Extract the `type' field from a swp_entry_t.  The swp_entry_t is in
 94 * arch-independent format
 95 */
 96static inline unsigned swp_type(swp_entry_t entry)
 97{
 98	return (entry.val >> SWP_TYPE_SHIFT);
 99}
100
101/*
102 * Extract the `offset' field from a swp_entry_t.  The swp_entry_t is in
103 * arch-independent format
104 */
105static inline pgoff_t swp_offset(swp_entry_t entry)
106{
107	return entry.val & SWP_OFFSET_MASK;
108}
109
110/*
111 * Convert the arch-independent representation of a swp_entry_t into the
112 * arch-dependent pte representation.
113 */
114static inline pte_t swp_entry_to_pte(swp_entry_t entry)
115{
116	swp_entry_t arch_entry;
117
118	arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
119	return __swp_entry_to_pte(arch_entry);
120}
121
122static inline swp_entry_t radix_to_swp_entry(void *arg)
123{
124	swp_entry_t entry;
125
126	entry.val = xa_to_value(arg);
127	return entry;
128}
129
130static inline void *swp_to_radix_entry(swp_entry_t entry)
131{
132	return xa_mk_value(entry.val);
133}
134
135#if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
136static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
137{
138	return swp_entry(SWP_DEVICE_READ, offset);
139}
140
141static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
142{
143	return swp_entry(SWP_DEVICE_WRITE, offset);
144}
145
146static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
147{
148	return swp_entry(SWP_DEVICE_EXCLUSIVE, offset);
149}
150
151#else /* CONFIG_DEVICE_PRIVATE */
152static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
153{
154	return swp_entry(0, 0);
155}
156
157static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
158{
159	return swp_entry(0, 0);
160}
161
162static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
163{
164	return swp_entry(0, 0);
165}
166
167#endif /* CONFIG_DEVICE_PRIVATE */
168
169#ifdef CONFIG_MIGRATION
170
171static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
172{
173	return swp_entry(SWP_MIGRATION_READ, offset);
174}
175
176static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
177{
178	return swp_entry(SWP_MIGRATION_READ_EXCLUSIVE, offset);
179}
180
181static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
182{
183	return swp_entry(SWP_MIGRATION_WRITE, offset);
184}
185
186/*
187 * Returns whether the host has large enough swap offset field to support
188 * carrying over pgtable A/D bits for page migrations.  The result is
189 * pretty much arch specific.
190 */
191static inline bool migration_entry_supports_ad(void)
192{
193#ifdef CONFIG_SWAP
194	return swap_migration_ad_supported;
195#else  /* CONFIG_SWAP */
196	return false;
197#endif	/* CONFIG_SWAP */
198}
199
200static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
201{
202	if (migration_entry_supports_ad())
203		return swp_entry(swp_type(entry),
204				 swp_offset(entry) | SWP_MIG_YOUNG);
205	return entry;
206}
207
208static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
209{
210	if (migration_entry_supports_ad())
211		return swp_entry(swp_type(entry),
212				 swp_offset(entry) | SWP_MIG_DIRTY);
213	return entry;
214}
215
216extern void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
217					unsigned long address);
218extern void migration_entry_wait_huge(struct vm_area_struct *vma, unsigned long addr, pte_t *pte);
219#else  /* CONFIG_MIGRATION */
220static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
221{
222	return swp_entry(0, 0);
223}
224
225static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
226{
227	return swp_entry(0, 0);
228}
229
230static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
231{
232	return swp_entry(0, 0);
233}
234
235static inline void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
236					unsigned long address) { }
237static inline void migration_entry_wait_huge(struct vm_area_struct *vma,
238					     unsigned long addr, pte_t *pte) { }
239
240static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
241{
242	return entry;
243}
244
245static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
246{
247	return entry;
248}
249
250#endif	/* CONFIG_MIGRATION */
251
252#ifdef CONFIG_MEMORY_FAILURE
253
254/*
255 * Support for hardware poisoned pages
256 */
257static inline swp_entry_t make_hwpoison_entry(struct page *page)
258{
259	BUG_ON(!PageLocked(page));
260	return swp_entry(SWP_HWPOISON, page_to_pfn(page));
261}
262
263static inline int is_hwpoison_entry(swp_entry_t entry)
264{
265	return swp_type(entry) == SWP_HWPOISON;
266}
267
268#else
269
270static inline swp_entry_t make_hwpoison_entry(struct page *page)
271{
272	return swp_entry(0, 0);
273}
274
275static inline int is_hwpoison_entry(swp_entry_t swp)
276{
277	return 0;
278}
279#endif
280
281typedef unsigned long pte_marker;
282
283#define  PTE_MARKER_UFFD_WP			BIT(0)
284/*
285 * "Poisoned" here is meant in the very general sense of "future accesses are
286 * invalid", instead of referring very specifically to hardware memory errors.
287 * This marker is meant to represent any of various different causes of this.
288 *
289 * Note that, when encountered by the faulting logic, PTEs with this marker will
290 * result in VM_FAULT_HWPOISON and thus regardless trigger hardware memory error
291 * logic.
292 */
293#define  PTE_MARKER_POISONED			BIT(1)
294/*
295 * Indicates that, on fault, this PTE will case a SIGSEGV signal to be
296 * sent. This means guard markers behave in effect as if the region were mapped
297 * PROT_NONE, rather than if they were a memory hole or equivalent.
298 */
299#define  PTE_MARKER_GUARD			BIT(2)
300#define  PTE_MARKER_MASK			(BIT(3) - 1)
301
302static inline swp_entry_t make_pte_marker_entry(pte_marker marker)
303{
304	return swp_entry(SWP_PTE_MARKER, marker);
305}
306
307static inline pte_t make_pte_marker(pte_marker marker)
308{
309	return swp_entry_to_pte(make_pte_marker_entry(marker));
310}
311
312static inline swp_entry_t make_poisoned_swp_entry(void)
313{
314	return make_pte_marker_entry(PTE_MARKER_POISONED);
315}
316
317static inline swp_entry_t make_guard_swp_entry(void)
318{
319	return make_pte_marker_entry(PTE_MARKER_GUARD);
320}
321
322struct page_vma_mapped_walk;
323
324#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
325extern int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
326		struct page *page);
327
328extern void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
329		struct page *new);
330
331extern void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd);
332
333static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
334{
335	swp_entry_t arch_entry;
336
337	arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
338	return __swp_entry_to_pmd(arch_entry);
339}
340
341#else  /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
342static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
343		struct page *page)
344{
345	BUILD_BUG();
346}
347
348static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
349		struct page *new)
350{
351	BUILD_BUG();
352}
353
354static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
355
356static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
357{
358	return __pmd(0);
359}
360
361#endif  /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
362
363#endif /* CONFIG_MMU */
364#endif /* _LINUX_SWAPOPS_H */