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/swapops.h>
11 #include <linux/types.h>
12 #include <linux/wait_bit.h>
13 #include <linux/xarray.h>
15 static DEFINE_XARRAY(pgmap_array);
16 #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
17 #define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
19 #ifdef CONFIG_DEV_PAGEMAP_OPS
20 DEFINE_STATIC_KEY_FALSE(devmap_managed_key);
21 EXPORT_SYMBOL(devmap_managed_key);
22 static atomic_t devmap_managed_enable;
24 static void devmap_managed_enable_put(void *data)
26 if (atomic_dec_and_test(&devmap_managed_enable))
27 static_branch_disable(&devmap_managed_key);
30 static int devmap_managed_enable_get(struct device *dev, struct dev_pagemap *pgmap)
32 if (!pgmap->ops || !pgmap->ops->page_free) {
33 WARN(1, "Missing page_free method\n");
37 if (atomic_inc_return(&devmap_managed_enable) == 1)
38 static_branch_enable(&devmap_managed_key);
39 return devm_add_action_or_reset(dev, devmap_managed_enable_put, NULL);
42 static int devmap_managed_enable_get(struct device *dev, struct dev_pagemap *pgmap)
46 #endif /* CONFIG_DEV_PAGEMAP_OPS */
48 static void pgmap_array_delete(struct resource *res)
50 xa_store_range(&pgmap_array, PHYS_PFN(res->start), PHYS_PFN(res->end),
55 static unsigned long pfn_first(struct dev_pagemap *pgmap)
57 return PHYS_PFN(pgmap->res.start) +
58 vmem_altmap_offset(pgmap_altmap(pgmap));
61 static unsigned long pfn_end(struct dev_pagemap *pgmap)
63 const struct resource *res = &pgmap->res;
65 return (res->start + resource_size(res)) >> PAGE_SHIFT;
68 static unsigned long pfn_next(unsigned long pfn)
75 #define for_each_device_pfn(pfn, map) \
76 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn = pfn_next(pfn))
78 static void dev_pagemap_kill(struct dev_pagemap *pgmap)
80 if (pgmap->ops && pgmap->ops->kill)
81 pgmap->ops->kill(pgmap);
83 percpu_ref_kill(pgmap->ref);
86 static void dev_pagemap_cleanup(struct dev_pagemap *pgmap)
88 if (pgmap->ops && pgmap->ops->cleanup) {
89 pgmap->ops->cleanup(pgmap);
91 wait_for_completion(&pgmap->done);
92 percpu_ref_exit(pgmap->ref);
95 * Undo the pgmap ref assignment for the internal case as the
96 * caller may re-enable the same pgmap.
98 if (pgmap->ref == &pgmap->internal_ref)
102 static void devm_memremap_pages_release(void *data)
104 struct dev_pagemap *pgmap = data;
105 struct device *dev = pgmap->dev;
106 struct resource *res = &pgmap->res;
107 struct page *first_page;
111 dev_pagemap_kill(pgmap);
112 for_each_device_pfn(pfn, pgmap)
113 put_page(pfn_to_page(pfn));
114 dev_pagemap_cleanup(pgmap);
116 /* make sure to access a memmap that was actually initialized */
117 first_page = pfn_to_page(pfn_first(pgmap));
119 /* pages are dead and unused, undo the arch mapping */
120 nid = page_to_nid(first_page);
123 if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
124 __remove_pages(page_zone(first_page), PHYS_PFN(res->start),
125 PHYS_PFN(resource_size(res)), NULL);
127 arch_remove_memory(nid, res->start, resource_size(res),
128 pgmap_altmap(pgmap));
129 kasan_remove_zero_shadow(__va(res->start), resource_size(res));
133 untrack_pfn(NULL, PHYS_PFN(res->start), resource_size(res));
134 pgmap_array_delete(res);
135 dev_WARN_ONCE(dev, pgmap->altmap.alloc,
136 "%s: failed to free all reserved pages\n", __func__);
139 static void dev_pagemap_percpu_release(struct percpu_ref *ref)
141 struct dev_pagemap *pgmap =
142 container_of(ref, struct dev_pagemap, internal_ref);
144 complete(&pgmap->done);
148 * devm_memremap_pages - remap and provide memmap backing for the given resource
149 * @dev: hosting device for @res
150 * @pgmap: pointer to a struct dev_pagemap
153 * 1/ At a minimum the res and type members of @pgmap must be initialized
154 * by the caller before passing it to this function
156 * 2/ The altmap field may optionally be initialized, in which case
157 * PGMAP_ALTMAP_VALID must be set in pgmap->flags.
159 * 3/ The ref field may optionally be provided, in which pgmap->ref must be
160 * 'live' on entry and will be killed and reaped at
161 * devm_memremap_pages_release() time, or if this routine fails.
163 * 4/ res is expected to be a host memory range that could feasibly be
164 * treated as a "System RAM" range, i.e. not a device mmio range, but
165 * this is not enforced.
167 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap)
169 struct resource *res = &pgmap->res;
170 struct dev_pagemap *conflict_pgmap;
171 struct mhp_restrictions restrictions = {
173 * We do not want any optional features only our own memmap
175 .altmap = pgmap_altmap(pgmap),
177 pgprot_t pgprot = PAGE_KERNEL;
178 int error, nid, is_ram;
179 bool need_devmap_managed = true;
181 switch (pgmap->type) {
182 case MEMORY_DEVICE_PRIVATE:
183 if (!IS_ENABLED(CONFIG_DEVICE_PRIVATE)) {
184 WARN(1, "Device private memory not supported\n");
185 return ERR_PTR(-EINVAL);
187 if (!pgmap->ops || !pgmap->ops->migrate_to_ram) {
188 WARN(1, "Missing migrate_to_ram method\n");
189 return ERR_PTR(-EINVAL);
192 case MEMORY_DEVICE_FS_DAX:
193 if (!IS_ENABLED(CONFIG_ZONE_DEVICE) ||
194 IS_ENABLED(CONFIG_FS_DAX_LIMITED)) {
195 WARN(1, "File system DAX not supported\n");
196 return ERR_PTR(-EINVAL);
199 case MEMORY_DEVICE_DEVDAX:
200 case MEMORY_DEVICE_PCI_P2PDMA:
201 need_devmap_managed = false;
204 WARN(1, "Invalid pgmap type %d\n", pgmap->type);
209 if (pgmap->ops && (pgmap->ops->kill || pgmap->ops->cleanup))
210 return ERR_PTR(-EINVAL);
212 init_completion(&pgmap->done);
213 error = percpu_ref_init(&pgmap->internal_ref,
214 dev_pagemap_percpu_release, 0, GFP_KERNEL);
216 return ERR_PTR(error);
217 pgmap->ref = &pgmap->internal_ref;
219 if (!pgmap->ops || !pgmap->ops->kill || !pgmap->ops->cleanup) {
220 WARN(1, "Missing reference count teardown definition\n");
221 return ERR_PTR(-EINVAL);
225 if (need_devmap_managed) {
226 error = devmap_managed_enable_get(dev, pgmap);
228 return ERR_PTR(error);
231 conflict_pgmap = get_dev_pagemap(PHYS_PFN(res->start), NULL);
232 if (conflict_pgmap) {
233 dev_WARN(dev, "Conflicting mapping in same section\n");
234 put_dev_pagemap(conflict_pgmap);
239 conflict_pgmap = get_dev_pagemap(PHYS_PFN(res->end), NULL);
240 if (conflict_pgmap) {
241 dev_WARN(dev, "Conflicting mapping in same section\n");
242 put_dev_pagemap(conflict_pgmap);
247 is_ram = region_intersects(res->start, resource_size(res),
248 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
250 if (is_ram != REGION_DISJOINT) {
251 WARN_ONCE(1, "%s attempted on %s region %pr\n", __func__,
252 is_ram == REGION_MIXED ? "mixed" : "ram", res);
259 error = xa_err(xa_store_range(&pgmap_array, PHYS_PFN(res->start),
260 PHYS_PFN(res->end), pgmap, GFP_KERNEL));
264 nid = dev_to_node(dev);
268 error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(res->start), 0,
276 * For device private memory we call add_pages() as we only need to
277 * allocate and initialize struct page for the device memory. More-
278 * over the device memory is un-accessible thus we do not want to
279 * create a linear mapping for the memory like arch_add_memory()
282 * For all other device memory types, which are accessible by
283 * the CPU, we do want the linear mapping and thus use
286 if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
287 error = add_pages(nid, PHYS_PFN(res->start),
288 PHYS_PFN(resource_size(res)), &restrictions);
290 error = kasan_add_zero_shadow(__va(res->start), resource_size(res));
296 error = arch_add_memory(nid, res->start, resource_size(res),
303 zone = &NODE_DATA(nid)->node_zones[ZONE_DEVICE];
304 move_pfn_range_to_zone(zone, PHYS_PFN(res->start),
305 PHYS_PFN(resource_size(res)), restrictions.altmap);
313 * Initialization of the pages has been deferred until now in order
314 * to allow us to do the work while not holding the hotplug lock.
316 memmap_init_zone_device(&NODE_DATA(nid)->node_zones[ZONE_DEVICE],
317 PHYS_PFN(res->start),
318 PHYS_PFN(resource_size(res)), pgmap);
319 percpu_ref_get_many(pgmap->ref, pfn_end(pgmap) - pfn_first(pgmap));
321 error = devm_add_action_or_reset(dev, devm_memremap_pages_release,
324 return ERR_PTR(error);
326 return __va(res->start);
329 kasan_remove_zero_shadow(__va(res->start), resource_size(res));
331 untrack_pfn(NULL, PHYS_PFN(res->start), resource_size(res));
333 pgmap_array_delete(res);
335 dev_pagemap_kill(pgmap);
336 dev_pagemap_cleanup(pgmap);
337 return ERR_PTR(error);
339 EXPORT_SYMBOL_GPL(devm_memremap_pages);
341 void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap)
343 devm_release_action(dev, devm_memremap_pages_release, pgmap);
345 EXPORT_SYMBOL_GPL(devm_memunmap_pages);
347 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
349 /* number of pfns from base where pfn_to_page() is valid */
351 return altmap->reserve + altmap->free;
355 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
357 altmap->alloc -= nr_pfns;
361 * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn
362 * @pfn: page frame number to lookup page_map
363 * @pgmap: optional known pgmap that already has a reference
365 * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap
366 * is non-NULL but does not cover @pfn the reference to it will be released.
368 struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
369 struct dev_pagemap *pgmap)
371 resource_size_t phys = PFN_PHYS(pfn);
374 * In the cached case we're already holding a live reference.
377 if (phys >= pgmap->res.start && phys <= pgmap->res.end)
379 put_dev_pagemap(pgmap);
382 /* fall back to slow path lookup */
384 pgmap = xa_load(&pgmap_array, PHYS_PFN(phys));
385 if (pgmap && !percpu_ref_tryget_live(pgmap->ref))
391 EXPORT_SYMBOL_GPL(get_dev_pagemap);
393 #ifdef CONFIG_DEV_PAGEMAP_OPS
394 void __put_devmap_managed_page(struct page *page)
396 int count = page_ref_dec_return(page);
399 * If refcount is 1 then page is freed and refcount is stable as nobody
400 * holds a reference on the page.
403 /* Clear Active bit in case of parallel mark_page_accessed */
404 __ClearPageActive(page);
405 __ClearPageWaiters(page);
407 mem_cgroup_uncharge(page);
410 * When a device_private page is freed, the page->mapping field
411 * may still contain a (stale) mapping value. For example, the
412 * lower bits of page->mapping may still identify the page as
413 * an anonymous page. Ultimately, this entire field is just
414 * stale and wrong, and it will cause errors if not cleared.
417 * migrate_vma_pages()
418 * migrate_vma_insert_page()
419 * page_add_new_anon_rmap()
420 * __page_set_anon_rmap()
421 * ...checks page->mapping, via PageAnon(page) call,
422 * and incorrectly concludes that the page is an
423 * anonymous page. Therefore, it incorrectly,
424 * silently fails to set up the new anon rmap.
426 * For other types of ZONE_DEVICE pages, migration is either
427 * handled differently or not done at all, so there is no need
428 * to clear page->mapping.
430 if (is_device_private_page(page))
431 page->mapping = NULL;
433 page->pgmap->ops->page_free(page);
437 EXPORT_SYMBOL(__put_devmap_managed_page);
438 #endif /* CONFIG_DEV_PAGEMAP_OPS */