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);
17 #ifdef CONFIG_DEV_PAGEMAP_OPS
18 DEFINE_STATIC_KEY_FALSE(devmap_managed_key);
19 EXPORT_SYMBOL(devmap_managed_key);
20 static atomic_t devmap_managed_enable;
22 static void devmap_managed_enable_put(void)
24 if (atomic_dec_and_test(&devmap_managed_enable))
25 static_branch_disable(&devmap_managed_key);
28 static int devmap_managed_enable_get(struct dev_pagemap *pgmap)
30 if (pgmap->type == MEMORY_DEVICE_PRIVATE &&
31 (!pgmap->ops || !pgmap->ops->page_free)) {
32 WARN(1, "Missing page_free method\n");
36 if (atomic_inc_return(&devmap_managed_enable) == 1)
37 static_branch_enable(&devmap_managed_key);
41 static int devmap_managed_enable_get(struct dev_pagemap *pgmap)
45 static void devmap_managed_enable_put(void)
48 #endif /* CONFIG_DEV_PAGEMAP_OPS */
50 static void pgmap_array_delete(struct resource *res)
52 xa_store_range(&pgmap_array, PHYS_PFN(res->start), PHYS_PFN(res->end),
57 static unsigned long pfn_first(struct dev_pagemap *pgmap)
59 return PHYS_PFN(pgmap->res.start) +
60 vmem_altmap_offset(pgmap_altmap(pgmap));
63 static unsigned long pfn_end(struct dev_pagemap *pgmap)
65 const struct resource *res = &pgmap->res;
67 return (res->start + resource_size(res)) >> PAGE_SHIFT;
70 static unsigned long pfn_next(unsigned long pfn)
77 #define for_each_device_pfn(pfn, map) \
78 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn = pfn_next(pfn))
80 static void dev_pagemap_kill(struct dev_pagemap *pgmap)
82 if (pgmap->ops && pgmap->ops->kill)
83 pgmap->ops->kill(pgmap);
85 percpu_ref_kill(pgmap->ref);
88 static void dev_pagemap_cleanup(struct dev_pagemap *pgmap)
90 if (pgmap->ops && pgmap->ops->cleanup) {
91 pgmap->ops->cleanup(pgmap);
93 wait_for_completion(&pgmap->done);
94 percpu_ref_exit(pgmap->ref);
97 * Undo the pgmap ref assignment for the internal case as the
98 * caller may re-enable the same pgmap.
100 if (pgmap->ref == &pgmap->internal_ref)
104 void memunmap_pages(struct dev_pagemap *pgmap)
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(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 WARN_ONCE(pgmap->altmap.alloc, "failed to free all reserved pages\n");
136 devmap_managed_enable_put();
138 EXPORT_SYMBOL_GPL(memunmap_pages);
140 static void devm_memremap_pages_release(void *data)
142 memunmap_pages(data);
145 static void dev_pagemap_percpu_release(struct percpu_ref *ref)
147 struct dev_pagemap *pgmap =
148 container_of(ref, struct dev_pagemap, internal_ref);
150 complete(&pgmap->done);
154 * Not device managed version of dev_memremap_pages, undone by
155 * memunmap_pages(). Please use dev_memremap_pages if you have a struct
158 void *memremap_pages(struct dev_pagemap *pgmap, int nid)
160 struct resource *res = &pgmap->res;
161 struct dev_pagemap *conflict_pgmap;
162 struct mhp_restrictions restrictions = {
164 * We do not want any optional features only our own memmap
166 .altmap = pgmap_altmap(pgmap),
168 pgprot_t pgprot = PAGE_KERNEL;
170 bool need_devmap_managed = true;
172 switch (pgmap->type) {
173 case MEMORY_DEVICE_PRIVATE:
174 if (!IS_ENABLED(CONFIG_DEVICE_PRIVATE)) {
175 WARN(1, "Device private memory not supported\n");
176 return ERR_PTR(-EINVAL);
178 if (!pgmap->ops || !pgmap->ops->migrate_to_ram) {
179 WARN(1, "Missing migrate_to_ram method\n");
180 return ERR_PTR(-EINVAL);
183 case MEMORY_DEVICE_FS_DAX:
184 if (!IS_ENABLED(CONFIG_ZONE_DEVICE) ||
185 IS_ENABLED(CONFIG_FS_DAX_LIMITED)) {
186 WARN(1, "File system DAX not supported\n");
187 return ERR_PTR(-EINVAL);
190 case MEMORY_DEVICE_DEVDAX:
191 case MEMORY_DEVICE_PCI_P2PDMA:
192 need_devmap_managed = false;
195 WARN(1, "Invalid pgmap type %d\n", pgmap->type);
200 if (pgmap->ops && (pgmap->ops->kill || pgmap->ops->cleanup))
201 return ERR_PTR(-EINVAL);
203 init_completion(&pgmap->done);
204 error = percpu_ref_init(&pgmap->internal_ref,
205 dev_pagemap_percpu_release, 0, GFP_KERNEL);
207 return ERR_PTR(error);
208 pgmap->ref = &pgmap->internal_ref;
210 if (!pgmap->ops || !pgmap->ops->kill || !pgmap->ops->cleanup) {
211 WARN(1, "Missing reference count teardown definition\n");
212 return ERR_PTR(-EINVAL);
216 if (need_devmap_managed) {
217 error = devmap_managed_enable_get(pgmap);
219 return ERR_PTR(error);
222 conflict_pgmap = get_dev_pagemap(PHYS_PFN(res->start), NULL);
223 if (conflict_pgmap) {
224 WARN(1, "Conflicting mapping in same section\n");
225 put_dev_pagemap(conflict_pgmap);
230 conflict_pgmap = get_dev_pagemap(PHYS_PFN(res->end), NULL);
231 if (conflict_pgmap) {
232 WARN(1, "Conflicting mapping in same section\n");
233 put_dev_pagemap(conflict_pgmap);
238 is_ram = region_intersects(res->start, resource_size(res),
239 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
241 if (is_ram != REGION_DISJOINT) {
242 WARN_ONCE(1, "%s attempted on %s region %pr\n", __func__,
243 is_ram == REGION_MIXED ? "mixed" : "ram", res);
248 error = xa_err(xa_store_range(&pgmap_array, PHYS_PFN(res->start),
249 PHYS_PFN(res->end), pgmap, GFP_KERNEL));
256 error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(res->start), 0,
264 * For device private memory we call add_pages() as we only need to
265 * allocate and initialize struct page for the device memory. More-
266 * over the device memory is un-accessible thus we do not want to
267 * create a linear mapping for the memory like arch_add_memory()
270 * For all other device memory types, which are accessible by
271 * the CPU, we do want the linear mapping and thus use
274 if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
275 error = add_pages(nid, PHYS_PFN(res->start),
276 PHYS_PFN(resource_size(res)), &restrictions);
278 error = kasan_add_zero_shadow(__va(res->start), resource_size(res));
284 error = arch_add_memory(nid, res->start, resource_size(res),
291 zone = &NODE_DATA(nid)->node_zones[ZONE_DEVICE];
292 move_pfn_range_to_zone(zone, PHYS_PFN(res->start),
293 PHYS_PFN(resource_size(res)), restrictions.altmap);
301 * Initialization of the pages has been deferred until now in order
302 * to allow us to do the work while not holding the hotplug lock.
304 memmap_init_zone_device(&NODE_DATA(nid)->node_zones[ZONE_DEVICE],
305 PHYS_PFN(res->start),
306 PHYS_PFN(resource_size(res)), pgmap);
307 percpu_ref_get_many(pgmap->ref, pfn_end(pgmap) - pfn_first(pgmap));
308 return __va(res->start);
311 kasan_remove_zero_shadow(__va(res->start), resource_size(res));
313 untrack_pfn(NULL, PHYS_PFN(res->start), resource_size(res));
315 pgmap_array_delete(res);
317 dev_pagemap_kill(pgmap);
318 dev_pagemap_cleanup(pgmap);
319 devmap_managed_enable_put();
320 return ERR_PTR(error);
322 EXPORT_SYMBOL_GPL(memremap_pages);
325 * devm_memremap_pages - remap and provide memmap backing for the given resource
326 * @dev: hosting device for @res
327 * @pgmap: pointer to a struct dev_pagemap
330 * 1/ At a minimum the res and type members of @pgmap must be initialized
331 * by the caller before passing it to this function
333 * 2/ The altmap field may optionally be initialized, in which case
334 * PGMAP_ALTMAP_VALID must be set in pgmap->flags.
336 * 3/ The ref field may optionally be provided, in which pgmap->ref must be
337 * 'live' on entry and will be killed and reaped at
338 * devm_memremap_pages_release() time, or if this routine fails.
340 * 4/ res is expected to be a host memory range that could feasibly be
341 * treated as a "System RAM" range, i.e. not a device mmio range, but
342 * this is not enforced.
344 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap)
349 ret = memremap_pages(pgmap, dev_to_node(dev));
353 error = devm_add_action_or_reset(dev, devm_memremap_pages_release,
356 return ERR_PTR(error);
359 EXPORT_SYMBOL_GPL(devm_memremap_pages);
361 void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap)
363 devm_release_action(dev, devm_memremap_pages_release, pgmap);
365 EXPORT_SYMBOL_GPL(devm_memunmap_pages);
367 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
369 /* number of pfns from base where pfn_to_page() is valid */
371 return altmap->reserve + altmap->free;
375 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
377 altmap->alloc -= nr_pfns;
381 * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn
382 * @pfn: page frame number to lookup page_map
383 * @pgmap: optional known pgmap that already has a reference
385 * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap
386 * is non-NULL but does not cover @pfn the reference to it will be released.
388 struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
389 struct dev_pagemap *pgmap)
391 resource_size_t phys = PFN_PHYS(pfn);
394 * In the cached case we're already holding a live reference.
397 if (phys >= pgmap->res.start && phys <= pgmap->res.end)
399 put_dev_pagemap(pgmap);
402 /* fall back to slow path lookup */
404 pgmap = xa_load(&pgmap_array, PHYS_PFN(phys));
405 if (pgmap && !percpu_ref_tryget_live(pgmap->ref))
411 EXPORT_SYMBOL_GPL(get_dev_pagemap);
413 #ifdef CONFIG_DEV_PAGEMAP_OPS
414 void free_devmap_managed_page(struct page *page)
416 /* notify page idle for dax */
417 if (!is_device_private_page(page)) {
418 wake_up_var(&page->_refcount);
422 /* Clear Active bit in case of parallel mark_page_accessed */
423 __ClearPageActive(page);
424 __ClearPageWaiters(page);
426 mem_cgroup_uncharge(page);
429 * When a device_private page is freed, the page->mapping field
430 * may still contain a (stale) mapping value. For example, the
431 * lower bits of page->mapping may still identify the page as an
432 * anonymous page. Ultimately, this entire field is just stale
433 * and wrong, and it will cause errors if not cleared. One
436 * migrate_vma_pages()
437 * migrate_vma_insert_page()
438 * page_add_new_anon_rmap()
439 * __page_set_anon_rmap()
440 * ...checks page->mapping, via PageAnon(page) call,
441 * and incorrectly concludes that the page is an
442 * anonymous page. Therefore, it incorrectly,
443 * silently fails to set up the new anon rmap.
445 * For other types of ZONE_DEVICE pages, migration is either
446 * handled differently or not done at all, so there is no need
447 * to clear page->mapping.
449 page->mapping = NULL;
450 page->pgmap->ops->page_free(page);
452 #endif /* CONFIG_DEV_PAGEMAP_OPS */