memremap.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef _LINUX_MEMREMAP_H_
  3#define _LINUX_MEMREMAP_H_
  4
  5#include <linux/mmzone.h>
  6#include <linux/range.h>
  7#include <linux/ioport.h>
  8#include <linux/percpu-refcount.h>
  9
 10struct resource;
 11struct device;
 12
 13/**
 14 * struct vmem_altmap - pre-allocated storage for vmemmap_populate
 15 * @base_pfn: base of the entire dev_pagemap mapping
 16 * @reserve: pages mapped, but reserved for driver use (relative to @base)
 17 * @free: free pages set aside in the mapping for memmap storage
 18 * @align: pages reserved to meet allocation alignments
 19 * @alloc: track pages consumed, private to vmemmap_populate()
 20 */
 21struct vmem_altmap {
 22	unsigned long base_pfn;
 23	const unsigned long end_pfn;
 24	const unsigned long reserve;
 25	unsigned long free;
 26	unsigned long align;
 27	unsigned long alloc;
 28};
 29
 30/*
 31 * Specialize ZONE_DEVICE memory into multiple types each has a different
 32 * usage.
 33 *
 34 * MEMORY_DEVICE_PRIVATE:
 35 * Device memory that is not directly addressable by the CPU: CPU can neither
 36 * read nor write private memory. In this case, we do still have struct pages
 37 * backing the device memory. Doing so simplifies the implementation, but it is
 38 * important to remember that there are certain points at which the struct page
 39 * must be treated as an opaque object, rather than a "normal" struct page.
 40 *
 41 * A more complete discussion of unaddressable memory may be found in
 42 * include/linux/hmm.h and Documentation/mm/hmm.rst.
 43 *
 44 * MEMORY_DEVICE_COHERENT:
 45 * Device memory that is cache coherent from device and CPU point of view. This
 46 * is used on platforms that have an advanced system bus (like CAPI or CXL). A
 47 * driver can hotplug the device memory using ZONE_DEVICE and with that memory
 48 * type. Any page of a process can be migrated to such memory. However no one
 49 * should be allowed to pin such memory so that it can always be evicted.
 50 *
 51 * MEMORY_DEVICE_FS_DAX:
 52 * Host memory that has similar access semantics as System RAM i.e. DMA
 53 * coherent and supports page pinning. In support of coordinating page
 54 * pinning vs other operations MEMORY_DEVICE_FS_DAX arranges for a
 55 * wakeup event whenever a page is unpinned and becomes idle. This
 56 * wakeup is used to coordinate physical address space management (ex:
 57 * fs truncate/hole punch) vs pinned pages (ex: device dma).
 58 *
 59 * MEMORY_DEVICE_GENERIC:
 60 * Host memory that has similar access semantics as System RAM i.e. DMA
 61 * coherent and supports page pinning. This is for example used by DAX devices
 62 * that expose memory using a character device.
 63 *
 64 * MEMORY_DEVICE_PCI_P2PDMA:
 65 * Device memory residing in a PCI BAR intended for use with Peer-to-Peer
 66 * transactions.
 67 */
 68enum memory_type {
 69	/* 0 is reserved to catch uninitialized type fields */
 70	MEMORY_DEVICE_PRIVATE = 1,
 71	MEMORY_DEVICE_COHERENT,
 72	MEMORY_DEVICE_FS_DAX,
 73	MEMORY_DEVICE_GENERIC,
 74	MEMORY_DEVICE_PCI_P2PDMA,
 75};
 76
 77struct dev_pagemap_ops {
 78	/*
 79	 * Called once the folio refcount reaches 0.  The reference count will be
 80	 * reset to one by the core code after the method is called to prepare
 81	 * for handing out the folio again.
 82	 */
 83	void (*folio_free)(struct folio *folio);
 84
 85	/*
 86	 * Used for private (un-addressable) device memory only.  Must migrate
 87	 * the page back to a CPU accessible page.
 88	 */
 89	vm_fault_t (*migrate_to_ram)(struct vm_fault *vmf);
 90
 91	/*
 92	 * Handle the memory failure happens on a range of pfns.  Notify the
 93	 * processes who are using these pfns, and try to recover the data on
 94	 * them if necessary.  The mf_flags is finally passed to the recover
 95	 * function through the whole notify routine.
 96	 *
 97	 * When this is not implemented, or it returns -EOPNOTSUPP, the caller
 98	 * will fall back to a common handler called mf_generic_kill_procs().
 99	 */
100	int (*memory_failure)(struct dev_pagemap *pgmap, unsigned long pfn,
101			      unsigned long nr_pages, int mf_flags);
102
103	/*
104	 * Used for private (un-addressable) device memory only.
105	 * This callback is used when a folio is split into
106	 * a smaller folio
107	 */
108	void (*folio_split)(struct folio *head, struct folio *tail);
109};
110
111#define PGMAP_ALTMAP_VALID	(1 << 0)
112
113/**
114 * struct dev_pagemap - metadata for ZONE_DEVICE mappings
115 * @altmap: pre-allocated/reserved memory for vmemmap allocations
116 * @ref: reference count that pins the devm_memremap_pages() mapping
117 * @done: completion for @ref
118 * @type: memory type: see MEMORY_* above in memremap.h
119 * @flags: PGMAP_* flags to specify defailed behavior
120 * @vmemmap_shift: structural definition of how the vmemmap page metadata
121 *      is populated, specifically the metadata page order.
122 *	A zero value (default) uses base pages as the vmemmap metadata
123 *	representation. A bigger value will set up compound struct pages
124 *	of the requested order value.
125 * @ops: method table
126 * @owner: an opaque pointer identifying the entity that manages this
127 *	instance.  Used by various helpers to make sure that no
128 *	foreign ZONE_DEVICE memory is accessed.
129 * @nr_range: number of ranges to be mapped
130 * @range: range to be mapped when nr_range == 1
131 * @ranges: array of ranges to be mapped when nr_range > 1
132 */
133struct dev_pagemap {
134	struct vmem_altmap altmap;
135	struct percpu_ref ref;
136	struct completion done;
137	enum memory_type type;
138	unsigned int flags;
139	unsigned long vmemmap_shift;
140	const struct dev_pagemap_ops *ops;
141	void *owner;
142	int nr_range;
143	union {
144		struct range range;
145		DECLARE_FLEX_ARRAY(struct range, ranges);
146	};
147};
148
149static inline bool pgmap_has_memory_failure(struct dev_pagemap *pgmap)
150{
151	return pgmap->ops && pgmap->ops->memory_failure;
152}
153
154static inline struct vmem_altmap *pgmap_altmap(struct dev_pagemap *pgmap)
155{
156	if (pgmap->flags & PGMAP_ALTMAP_VALID)
157		return &pgmap->altmap;
158	return NULL;
159}
160
161static inline unsigned long pgmap_vmemmap_nr(struct dev_pagemap *pgmap)
162{
163	return 1 << pgmap->vmemmap_shift;
164}
165
166static inline bool folio_is_device_private(const struct folio *folio)
167{
168	return IS_ENABLED(CONFIG_DEVICE_PRIVATE) &&
169		folio_is_zone_device(folio) &&
170		folio->pgmap->type == MEMORY_DEVICE_PRIVATE;
171}
172
173static inline bool is_device_private_page(const struct page *page)
174{
175	return IS_ENABLED(CONFIG_DEVICE_PRIVATE) &&
176		folio_is_device_private(page_folio(page));
177}
178
179static inline bool folio_is_pci_p2pdma(const struct folio *folio)
180{
181	return IS_ENABLED(CONFIG_PCI_P2PDMA) &&
182		folio_is_zone_device(folio) &&
183		folio->pgmap->type == MEMORY_DEVICE_PCI_P2PDMA;
184}
185
186static inline void *folio_zone_device_data(const struct folio *folio)
187{
188	VM_WARN_ON_FOLIO(!folio_is_device_private(folio), folio);
189	return folio->page.zone_device_data;
190}
191
192static inline void folio_set_zone_device_data(struct folio *folio, void *data)
193{
194	VM_WARN_ON_FOLIO(!folio_is_device_private(folio), folio);
195	folio->page.zone_device_data = data;
196}
197
198static inline bool is_pci_p2pdma_page(const struct page *page)
199{
200	return IS_ENABLED(CONFIG_PCI_P2PDMA) &&
201		folio_is_pci_p2pdma(page_folio(page));
202}
203
204static inline bool folio_is_device_coherent(const struct folio *folio)
205{
206	return folio_is_zone_device(folio) &&
207		folio->pgmap->type == MEMORY_DEVICE_COHERENT;
208}
209
210static inline bool is_device_coherent_page(const struct page *page)
211{
212	return folio_is_device_coherent(page_folio(page));
213}
214
215static inline bool folio_is_fsdax(const struct folio *folio)
216{
217	return folio_is_zone_device(folio) &&
218		folio->pgmap->type == MEMORY_DEVICE_FS_DAX;
219}
220
221static inline bool is_fsdax_page(const struct page *page)
222{
223	return folio_is_fsdax(page_folio(page));
224}
225
226#ifdef CONFIG_ZONE_DEVICE
227void zone_device_page_init(struct page *page, struct dev_pagemap *pgmap,
228			   unsigned int order);
229void *memremap_pages(struct dev_pagemap *pgmap, int nid);
230void memunmap_pages(struct dev_pagemap *pgmap);
231void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap);
232void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap);
233struct dev_pagemap *get_dev_pagemap(unsigned long pfn);
234bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn);
235
236unsigned long memremap_compat_align(void);
237
238static inline void zone_device_folio_init(struct folio *folio,
239					  struct dev_pagemap *pgmap,
240					  unsigned int order)
241{
242	zone_device_page_init(&folio->page, pgmap, order);
243	if (order)
244		folio_set_large_rmappable(folio);
245}
246
247static inline void zone_device_private_split_cb(struct folio *original_folio,
248						struct folio *new_folio)
249{
250	if (folio_is_device_private(original_folio)) {
251		if (!original_folio->pgmap->ops->folio_split) {
252			if (new_folio) {
253				new_folio->pgmap = original_folio->pgmap;
254				new_folio->page.mapping =
255					original_folio->page.mapping;
256			}
257		} else {
258			original_folio->pgmap->ops->folio_split(original_folio,
259								 new_folio);
260		}
261	}
262}
263
264#else
265static inline void *devm_memremap_pages(struct device *dev,
266		struct dev_pagemap *pgmap)
267{
268	/*
269	 * Fail attempts to call devm_memremap_pages() without
270	 * ZONE_DEVICE support enabled, this requires callers to fall
271	 * back to plain devm_memremap() based on config
272	 */
273	WARN_ON_ONCE(1);
274	return ERR_PTR(-ENXIO);
275}
276
277static inline void devm_memunmap_pages(struct device *dev,
278		struct dev_pagemap *pgmap)
279{
280}
281
282static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn)
283{
284	return NULL;
285}
286
287static inline bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn)
288{
289	return false;
290}
291
292/* when memremap_pages() is disabled all archs can remap a single page */
293static inline unsigned long memremap_compat_align(void)
294{
295	return PAGE_SIZE;
296}
297
298static inline void zone_device_private_split_cb(struct folio *original_folio,
299						struct folio *new_folio)
300{
301}
302#endif /* CONFIG_ZONE_DEVICE */
303
304static inline void put_dev_pagemap(struct dev_pagemap *pgmap)
305{
306	if (pgmap)
307		percpu_ref_put(&pgmap->ref);
308}
309
310#endif /* _LINUX_MEMREMAP_H_ */