1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright(c) 2015 Intel Corporation. All rights reserved. */
3 #include <linux/device.h>
5 #include <linux/kasan.h>
6 #include <linux/memory_hotplug.h>
8 #include <linux/pfn_t.h>
9 #include <linux/swap.h>
10 #include <linux/mmzone.h>
11 #include <linux/swapops.h>
12 #include <linux/types.h>
13 #include <linux/wait_bit.h>
14 #include <linux/xarray.h>
16 static DEFINE_XARRAY(pgmap_array);
19 * The memremap() and memremap_pages() interfaces are alternately used
20 * to map persistent memory namespaces. These interfaces place different
21 * constraints on the alignment and size of the mapping (namespace).
22 * memremap() can map individual PAGE_SIZE pages. memremap_pages() can
23 * only map subsections (2MB), and at least one architecture (PowerPC)
24 * the minimum mapping granularity of memremap_pages() is 16MB.
26 * The role of memremap_compat_align() is to communicate the minimum
27 * arch supported alignment of a namespace such that it can freely
28 * switch modes without violating the arch constraint. Namely, do not
29 * allow a namespace to be PAGE_SIZE aligned since that namespace may be
30 * reconfigured into a mode that requires SUBSECTION_SIZE alignment.
32 #ifndef CONFIG_ARCH_HAS_MEMREMAP_COMPAT_ALIGN
33 unsigned long memremap_compat_align(void)
35 return SUBSECTION_SIZE;
37 EXPORT_SYMBOL_GPL(memremap_compat_align);
40 #ifdef CONFIG_DEV_PAGEMAP_OPS
41 DEFINE_STATIC_KEY_FALSE(devmap_managed_key);
42 EXPORT_SYMBOL(devmap_managed_key);
44 static void devmap_managed_enable_put(struct dev_pagemap *pgmap)
46 if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
47 pgmap->type == MEMORY_DEVICE_FS_DAX)
48 static_branch_dec(&devmap_managed_key);
51 static void devmap_managed_enable_get(struct dev_pagemap *pgmap)
53 if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
54 pgmap->type == MEMORY_DEVICE_FS_DAX)
55 static_branch_inc(&devmap_managed_key);
58 static void devmap_managed_enable_get(struct dev_pagemap *pgmap)
61 static void devmap_managed_enable_put(struct dev_pagemap *pgmap)
64 #endif /* CONFIG_DEV_PAGEMAP_OPS */
66 static void pgmap_array_delete(struct range *range)
68 xa_store_range(&pgmap_array, PHYS_PFN(range->start), PHYS_PFN(range->end),
73 static unsigned long pfn_first(struct dev_pagemap *pgmap, int range_id)
75 struct range *range = &pgmap->ranges[range_id];
76 unsigned long pfn = PHYS_PFN(range->start);
80 return pfn + vmem_altmap_offset(pgmap_altmap(pgmap));
83 static unsigned long pfn_end(struct dev_pagemap *pgmap, int range_id)
85 const struct range *range = &pgmap->ranges[range_id];
87 return (range->start + range_len(range)) >> PAGE_SHIFT;
90 static unsigned long pfn_next(unsigned long pfn)
97 #define for_each_device_pfn(pfn, map, i) \
98 for (pfn = pfn_first(map, i); pfn < pfn_end(map, i); pfn = pfn_next(pfn))
100 static void dev_pagemap_kill(struct dev_pagemap *pgmap)
102 if (pgmap->ops && pgmap->ops->kill)
103 pgmap->ops->kill(pgmap);
105 percpu_ref_kill(pgmap->ref);
108 static void dev_pagemap_cleanup(struct dev_pagemap *pgmap)
110 if (pgmap->ops && pgmap->ops->cleanup) {
111 pgmap->ops->cleanup(pgmap);
113 wait_for_completion(&pgmap->done);
114 percpu_ref_exit(pgmap->ref);
117 * Undo the pgmap ref assignment for the internal case as the
118 * caller may re-enable the same pgmap.
120 if (pgmap->ref == &pgmap->internal_ref)
124 static void pageunmap_range(struct dev_pagemap *pgmap, int range_id)
126 struct range *range = &pgmap->ranges[range_id];
127 struct page *first_page;
130 /* make sure to access a memmap that was actually initialized */
131 first_page = pfn_to_page(pfn_first(pgmap, range_id));
133 /* pages are dead and unused, undo the arch mapping */
134 nid = page_to_nid(first_page);
137 remove_pfn_range_from_zone(page_zone(first_page), PHYS_PFN(range->start),
138 PHYS_PFN(range_len(range)));
139 if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
140 __remove_pages(PHYS_PFN(range->start),
141 PHYS_PFN(range_len(range)), NULL);
143 arch_remove_memory(nid, range->start, range_len(range),
144 pgmap_altmap(pgmap));
145 kasan_remove_zero_shadow(__va(range->start), range_len(range));
149 untrack_pfn(NULL, PHYS_PFN(range->start), range_len(range));
150 pgmap_array_delete(range);
153 void memunmap_pages(struct dev_pagemap *pgmap)
158 dev_pagemap_kill(pgmap);
159 for (i = 0; i < pgmap->nr_range; i++)
160 for_each_device_pfn(pfn, pgmap, i)
161 put_page(pfn_to_page(pfn));
162 dev_pagemap_cleanup(pgmap);
164 for (i = 0; i < pgmap->nr_range; i++)
165 pageunmap_range(pgmap, i);
167 WARN_ONCE(pgmap->altmap.alloc, "failed to free all reserved pages\n");
168 devmap_managed_enable_put(pgmap);
170 EXPORT_SYMBOL_GPL(memunmap_pages);
172 static void devm_memremap_pages_release(void *data)
174 memunmap_pages(data);
177 static void dev_pagemap_percpu_release(struct percpu_ref *ref)
179 struct dev_pagemap *pgmap =
180 container_of(ref, struct dev_pagemap, internal_ref);
182 complete(&pgmap->done);
185 static int pagemap_range(struct dev_pagemap *pgmap, struct mhp_params *params,
186 int range_id, int nid)
188 struct range *range = &pgmap->ranges[range_id];
189 struct dev_pagemap *conflict_pgmap;
192 if (WARN_ONCE(pgmap_altmap(pgmap) && range_id > 0,
193 "altmap not supported for multiple ranges\n"))
196 conflict_pgmap = get_dev_pagemap(PHYS_PFN(range->start), NULL);
197 if (conflict_pgmap) {
198 WARN(1, "Conflicting mapping in same section\n");
199 put_dev_pagemap(conflict_pgmap);
203 conflict_pgmap = get_dev_pagemap(PHYS_PFN(range->end), NULL);
204 if (conflict_pgmap) {
205 WARN(1, "Conflicting mapping in same section\n");
206 put_dev_pagemap(conflict_pgmap);
210 is_ram = region_intersects(range->start, range_len(range),
211 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
213 if (is_ram != REGION_DISJOINT) {
214 WARN_ONCE(1, "attempted on %s region %#llx-%#llx\n",
215 is_ram == REGION_MIXED ? "mixed" : "ram",
216 range->start, range->end);
220 error = xa_err(xa_store_range(&pgmap_array, PHYS_PFN(range->start),
221 PHYS_PFN(range->end), pgmap, GFP_KERNEL));
228 error = track_pfn_remap(NULL, ¶ms->pgprot, PHYS_PFN(range->start), 0,
236 * For device private memory we call add_pages() as we only need to
237 * allocate and initialize struct page for the device memory. More-
238 * over the device memory is un-accessible thus we do not want to
239 * create a linear mapping for the memory like arch_add_memory()
242 * For all other device memory types, which are accessible by
243 * the CPU, we do want the linear mapping and thus use
246 if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
247 error = add_pages(nid, PHYS_PFN(range->start),
248 PHYS_PFN(range_len(range)), params);
250 error = kasan_add_zero_shadow(__va(range->start), range_len(range));
256 error = arch_add_memory(nid, range->start, range_len(range),
263 zone = &NODE_DATA(nid)->node_zones[ZONE_DEVICE];
264 move_pfn_range_to_zone(zone, PHYS_PFN(range->start),
265 PHYS_PFN(range_len(range)), params->altmap,
274 * Initialization of the pages has been deferred until now in order
275 * to allow us to do the work while not holding the hotplug lock.
277 memmap_init_zone_device(&NODE_DATA(nid)->node_zones[ZONE_DEVICE],
278 PHYS_PFN(range->start),
279 PHYS_PFN(range_len(range)), pgmap);
280 percpu_ref_get_many(pgmap->ref, pfn_end(pgmap, range_id)
281 - pfn_first(pgmap, range_id));
285 kasan_remove_zero_shadow(__va(range->start), range_len(range));
287 untrack_pfn(NULL, PHYS_PFN(range->start), range_len(range));
289 pgmap_array_delete(range);
295 * Not device managed version of dev_memremap_pages, undone by
296 * memunmap_pages(). Please use dev_memremap_pages if you have a struct
299 void *memremap_pages(struct dev_pagemap *pgmap, int nid)
301 struct mhp_params params = {
302 .altmap = pgmap_altmap(pgmap),
303 .pgprot = PAGE_KERNEL,
305 const int nr_range = pgmap->nr_range;
308 if (WARN_ONCE(!nr_range, "nr_range must be specified\n"))
309 return ERR_PTR(-EINVAL);
311 switch (pgmap->type) {
312 case MEMORY_DEVICE_PRIVATE:
313 if (!IS_ENABLED(CONFIG_DEVICE_PRIVATE)) {
314 WARN(1, "Device private memory not supported\n");
315 return ERR_PTR(-EINVAL);
317 if (!pgmap->ops || !pgmap->ops->migrate_to_ram) {
318 WARN(1, "Missing migrate_to_ram method\n");
319 return ERR_PTR(-EINVAL);
321 if (!pgmap->ops->page_free) {
322 WARN(1, "Missing page_free method\n");
323 return ERR_PTR(-EINVAL);
326 WARN(1, "Missing owner\n");
327 return ERR_PTR(-EINVAL);
330 case MEMORY_DEVICE_FS_DAX:
331 if (!IS_ENABLED(CONFIG_ZONE_DEVICE) ||
332 IS_ENABLED(CONFIG_FS_DAX_LIMITED)) {
333 WARN(1, "File system DAX not supported\n");
334 return ERR_PTR(-EINVAL);
337 case MEMORY_DEVICE_GENERIC:
339 case MEMORY_DEVICE_PCI_P2PDMA:
340 params.pgprot = pgprot_noncached(params.pgprot);
343 WARN(1, "Invalid pgmap type %d\n", pgmap->type);
348 if (pgmap->ops && (pgmap->ops->kill || pgmap->ops->cleanup))
349 return ERR_PTR(-EINVAL);
351 init_completion(&pgmap->done);
352 error = percpu_ref_init(&pgmap->internal_ref,
353 dev_pagemap_percpu_release, 0, GFP_KERNEL);
355 return ERR_PTR(error);
356 pgmap->ref = &pgmap->internal_ref;
358 if (!pgmap->ops || !pgmap->ops->kill || !pgmap->ops->cleanup) {
359 WARN(1, "Missing reference count teardown definition\n");
360 return ERR_PTR(-EINVAL);
364 devmap_managed_enable_get(pgmap);
367 * Clear the pgmap nr_range as it will be incremented for each
368 * successfully processed range. This communicates how many
369 * regions to unwind in the abort case.
373 for (i = 0; i < nr_range; i++) {
374 error = pagemap_range(pgmap, ¶ms, i, nid);
381 memunmap_pages(pgmap);
382 pgmap->nr_range = nr_range;
383 return ERR_PTR(error);
386 return __va(pgmap->ranges[0].start);
388 EXPORT_SYMBOL_GPL(memremap_pages);
391 * devm_memremap_pages - remap and provide memmap backing for the given resource
392 * @dev: hosting device for @res
393 * @pgmap: pointer to a struct dev_pagemap
396 * 1/ At a minimum the res and type members of @pgmap must be initialized
397 * by the caller before passing it to this function
399 * 2/ The altmap field may optionally be initialized, in which case
400 * PGMAP_ALTMAP_VALID must be set in pgmap->flags.
402 * 3/ The ref field may optionally be provided, in which pgmap->ref must be
403 * 'live' on entry and will be killed and reaped at
404 * devm_memremap_pages_release() time, or if this routine fails.
406 * 4/ range is expected to be a host memory range that could feasibly be
407 * treated as a "System RAM" range, i.e. not a device mmio range, but
408 * this is not enforced.
410 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap)
415 ret = memremap_pages(pgmap, dev_to_node(dev));
419 error = devm_add_action_or_reset(dev, devm_memremap_pages_release,
422 return ERR_PTR(error);
425 EXPORT_SYMBOL_GPL(devm_memremap_pages);
427 void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap)
429 devm_release_action(dev, devm_memremap_pages_release, pgmap);
431 EXPORT_SYMBOL_GPL(devm_memunmap_pages);
433 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
435 /* number of pfns from base where pfn_to_page() is valid */
437 return altmap->reserve + altmap->free;
441 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
443 altmap->alloc -= nr_pfns;
447 * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn
448 * @pfn: page frame number to lookup page_map
449 * @pgmap: optional known pgmap that already has a reference
451 * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap
452 * is non-NULL but does not cover @pfn the reference to it will be released.
454 struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
455 struct dev_pagemap *pgmap)
457 resource_size_t phys = PFN_PHYS(pfn);
460 * In the cached case we're already holding a live reference.
463 if (phys >= pgmap->range.start && phys <= pgmap->range.end)
465 put_dev_pagemap(pgmap);
468 /* fall back to slow path lookup */
470 pgmap = xa_load(&pgmap_array, PHYS_PFN(phys));
471 if (pgmap && !percpu_ref_tryget_live(pgmap->ref))
477 EXPORT_SYMBOL_GPL(get_dev_pagemap);
479 #ifdef CONFIG_DEV_PAGEMAP_OPS
480 void free_devmap_managed_page(struct page *page)
482 /* notify page idle for dax */
483 if (!is_device_private_page(page)) {
484 wake_up_var(&page->_refcount);
488 __ClearPageWaiters(page);
490 mem_cgroup_uncharge(page);
493 * When a device_private page is freed, the page->mapping field
494 * may still contain a (stale) mapping value. For example, the
495 * lower bits of page->mapping may still identify the page as an
496 * anonymous page. Ultimately, this entire field is just stale
497 * and wrong, and it will cause errors if not cleared. One
500 * migrate_vma_pages()
501 * migrate_vma_insert_page()
502 * page_add_new_anon_rmap()
503 * __page_set_anon_rmap()
504 * ...checks page->mapping, via PageAnon(page) call,
505 * and incorrectly concludes that the page is an
506 * anonymous page. Therefore, it incorrectly,
507 * silently fails to set up the new anon rmap.
509 * For other types of ZONE_DEVICE pages, migration is either
510 * handled differently or not done at all, so there is no need
511 * to clear page->mapping.
513 page->mapping = NULL;
514 page->pgmap->ops->page_free(page);
516 #endif /* CONFIG_DEV_PAGEMAP_OPS */