2 * Copyright(c) 2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/radix-tree.h>
14 #include <linux/memremap.h>
15 #include <linux/device.h>
16 #include <linux/types.h>
17 #include <linux/pfn_t.h>
20 #include <linux/memory_hotplug.h>
23 /* temporary while we convert existing ioremap_cache users to memremap */
24 __weak void __iomem *ioremap_cache(resource_size_t offset, unsigned long size)
26 return ioremap(offset, size);
30 #ifndef arch_memremap_wb
31 static void *arch_memremap_wb(resource_size_t offset, unsigned long size)
33 return (__force void *)ioremap_cache(offset, size);
37 static void *try_ram_remap(resource_size_t offset, size_t size)
39 unsigned long pfn = PHYS_PFN(offset);
41 /* In the simple case just return the existing linear address */
42 if (pfn_valid(pfn) && !PageHighMem(pfn_to_page(pfn)))
44 return NULL; /* fallback to arch_memremap_wb */
48 * memremap() - remap an iomem_resource as cacheable memory
49 * @offset: iomem resource start address
50 * @size: size of remap
51 * @flags: any of MEMREMAP_WB, MEMREMAP_WT and MEMREMAP_WC
53 * memremap() is "ioremap" for cases where it is known that the resource
54 * being mapped does not have i/o side effects and the __iomem
55 * annotation is not applicable. In the case of multiple flags, the different
56 * mapping types will be attempted in the order listed below until one of
59 * MEMREMAP_WB - matches the default mapping for System RAM on
60 * the architecture. This is usually a read-allocate write-back cache.
61 * Morever, if MEMREMAP_WB is specified and the requested remap region is RAM
62 * memremap() will bypass establishing a new mapping and instead return
63 * a pointer into the direct map.
65 * MEMREMAP_WT - establish a mapping whereby writes either bypass the
66 * cache or are written through to memory and never exist in a
67 * cache-dirty state with respect to program visibility. Attempts to
68 * map System RAM with this mapping type will fail.
70 * MEMREMAP_WC - establish a writecombine mapping, whereby writes may
71 * be coalesced together (e.g. in the CPU's write buffers), but is otherwise
72 * uncached. Attempts to map System RAM with this mapping type will fail.
74 void *memremap(resource_size_t offset, size_t size, unsigned long flags)
76 int is_ram = region_intersects(offset, size,
77 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
83 if (is_ram == REGION_MIXED) {
84 WARN_ONCE(1, "memremap attempted on mixed range %pa size: %#lx\n",
85 &offset, (unsigned long) size);
89 /* Try all mapping types requested until one returns non-NULL */
90 if (flags & MEMREMAP_WB) {
92 * MEMREMAP_WB is special in that it can be satisifed
93 * from the direct map. Some archs depend on the
94 * capability of memremap() to autodetect cases where
95 * the requested range is potentially in System RAM.
97 if (is_ram == REGION_INTERSECTS)
98 addr = try_ram_remap(offset, size);
100 addr = arch_memremap_wb(offset, size);
104 * If we don't have a mapping yet and other request flags are
105 * present then we will be attempting to establish a new virtual
106 * address mapping. Enforce that this mapping is not aliasing
109 if (!addr && is_ram == REGION_INTERSECTS && flags != MEMREMAP_WB) {
110 WARN_ONCE(1, "memremap attempted on ram %pa size: %#lx\n",
111 &offset, (unsigned long) size);
115 if (!addr && (flags & MEMREMAP_WT))
116 addr = ioremap_wt(offset, size);
118 if (!addr && (flags & MEMREMAP_WC))
119 addr = ioremap_wc(offset, size);
123 EXPORT_SYMBOL(memremap);
125 void memunmap(void *addr)
127 if (is_vmalloc_addr(addr))
128 iounmap((void __iomem *) addr);
130 EXPORT_SYMBOL(memunmap);
132 static void devm_memremap_release(struct device *dev, void *res)
134 memunmap(*(void **)res);
137 static int devm_memremap_match(struct device *dev, void *res, void *match_data)
139 return *(void **)res == match_data;
142 void *devm_memremap(struct device *dev, resource_size_t offset,
143 size_t size, unsigned long flags)
147 ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
150 return ERR_PTR(-ENOMEM);
152 addr = memremap(offset, size, flags);
155 devres_add(dev, ptr);
158 return ERR_PTR(-ENXIO);
163 EXPORT_SYMBOL(devm_memremap);
165 void devm_memunmap(struct device *dev, void *addr)
167 WARN_ON(devres_release(dev, devm_memremap_release,
168 devm_memremap_match, addr));
170 EXPORT_SYMBOL(devm_memunmap);
172 #ifdef CONFIG_ZONE_DEVICE
173 static DEFINE_MUTEX(pgmap_lock);
174 static RADIX_TREE(pgmap_radix, GFP_KERNEL);
175 #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
176 #define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
180 struct percpu_ref *ref;
181 struct dev_pagemap pgmap;
182 struct vmem_altmap altmap;
185 void get_zone_device_page(struct page *page)
187 percpu_ref_get(page->pgmap->ref);
189 EXPORT_SYMBOL(get_zone_device_page);
191 void put_zone_device_page(struct page *page)
193 put_dev_pagemap(page->pgmap);
195 EXPORT_SYMBOL(put_zone_device_page);
197 static void pgmap_radix_release(struct resource *res)
199 resource_size_t key, align_start, align_size, align_end;
201 align_start = res->start & ~(SECTION_SIZE - 1);
202 align_size = ALIGN(resource_size(res), SECTION_SIZE);
203 align_end = align_start + align_size - 1;
205 mutex_lock(&pgmap_lock);
206 for (key = res->start; key <= res->end; key += SECTION_SIZE)
207 radix_tree_delete(&pgmap_radix, key >> PA_SECTION_SHIFT);
208 mutex_unlock(&pgmap_lock);
211 static unsigned long pfn_first(struct page_map *page_map)
213 struct dev_pagemap *pgmap = &page_map->pgmap;
214 const struct resource *res = &page_map->res;
215 struct vmem_altmap *altmap = pgmap->altmap;
218 pfn = res->start >> PAGE_SHIFT;
220 pfn += vmem_altmap_offset(altmap);
224 static unsigned long pfn_end(struct page_map *page_map)
226 const struct resource *res = &page_map->res;
228 return (res->start + resource_size(res)) >> PAGE_SHIFT;
231 #define for_each_device_pfn(pfn, map) \
232 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn++)
234 static void devm_memremap_pages_release(struct device *dev, void *data)
236 struct page_map *page_map = data;
237 struct resource *res = &page_map->res;
238 resource_size_t align_start, align_size;
239 struct dev_pagemap *pgmap = &page_map->pgmap;
241 if (percpu_ref_tryget_live(pgmap->ref)) {
242 dev_WARN(dev, "%s: page mapping is still live!\n", __func__);
243 percpu_ref_put(pgmap->ref);
246 /* pages are dead and unused, undo the arch mapping */
247 align_start = res->start & ~(SECTION_SIZE - 1);
248 align_size = ALIGN(resource_size(res), SECTION_SIZE);
250 arch_remove_memory(align_start, align_size);
252 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
253 pgmap_radix_release(res);
254 dev_WARN_ONCE(dev, pgmap->altmap && pgmap->altmap->alloc,
255 "%s: failed to free all reserved pages\n", __func__);
258 /* assumes rcu_read_lock() held at entry */
259 struct dev_pagemap *find_dev_pagemap(resource_size_t phys)
261 struct page_map *page_map;
263 WARN_ON_ONCE(!rcu_read_lock_held());
265 page_map = radix_tree_lookup(&pgmap_radix, phys >> PA_SECTION_SHIFT);
266 return page_map ? &page_map->pgmap : NULL;
270 * devm_memremap_pages - remap and provide memmap backing for the given resource
271 * @dev: hosting device for @res
272 * @res: "host memory" address range
273 * @ref: a live per-cpu reference count
274 * @altmap: optional descriptor for allocating the memmap from @res
277 * 1/ @ref must be 'live' on entry and 'dead' before devm_memunmap_pages() time
278 * (or devm release event).
280 * 2/ @res is expected to be a host memory range that could feasibly be
281 * treated as a "System RAM" range, i.e. not a device mmio range, but
282 * this is not enforced.
284 void *devm_memremap_pages(struct device *dev, struct resource *res,
285 struct percpu_ref *ref, struct vmem_altmap *altmap)
287 resource_size_t key, align_start, align_size, align_end;
288 pgprot_t pgprot = PAGE_KERNEL;
289 struct dev_pagemap *pgmap;
290 struct page_map *page_map;
291 int error, nid, is_ram;
294 align_start = res->start & ~(SECTION_SIZE - 1);
295 align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE)
297 is_ram = region_intersects(align_start, align_size,
298 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
300 if (is_ram == REGION_MIXED) {
301 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
303 return ERR_PTR(-ENXIO);
306 if (is_ram == REGION_INTERSECTS)
307 return __va(res->start);
310 return ERR_PTR(-EINVAL);
312 page_map = devres_alloc_node(devm_memremap_pages_release,
313 sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
315 return ERR_PTR(-ENOMEM);
316 pgmap = &page_map->pgmap;
318 memcpy(&page_map->res, res, sizeof(*res));
322 memcpy(&page_map->altmap, altmap, sizeof(*altmap));
323 pgmap->altmap = &page_map->altmap;
326 pgmap->res = &page_map->res;
328 mutex_lock(&pgmap_lock);
330 align_end = align_start + align_size - 1;
331 for (key = align_start; key <= align_end; key += SECTION_SIZE) {
332 struct dev_pagemap *dup;
335 dup = find_dev_pagemap(key);
338 dev_err(dev, "%s: %pr collides with mapping for %s\n",
339 __func__, res, dev_name(dup->dev));
343 error = radix_tree_insert(&pgmap_radix, key >> PA_SECTION_SHIFT,
346 dev_err(dev, "%s: failed: %d\n", __func__, error);
350 mutex_unlock(&pgmap_lock);
354 nid = dev_to_node(dev);
358 error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(align_start), 0,
364 error = arch_add_memory(nid, align_start, align_size, true);
369 for_each_device_pfn(pfn, page_map) {
370 struct page *page = pfn_to_page(pfn);
373 * ZONE_DEVICE pages union ->lru with a ->pgmap back
374 * pointer. It is a bug if a ZONE_DEVICE page is ever
375 * freed or placed on a driver-private list. Seed the
376 * storage with LIST_POISON* values.
378 list_del(&page->lru);
381 devres_add(dev, page_map);
382 return __va(res->start);
385 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
388 pgmap_radix_release(res);
389 devres_free(page_map);
390 return ERR_PTR(error);
392 EXPORT_SYMBOL(devm_memremap_pages);
394 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
396 /* number of pfns from base where pfn_to_page() is valid */
397 return altmap->reserve + altmap->free;
400 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
402 altmap->alloc -= nr_pfns;
405 struct vmem_altmap *to_vmem_altmap(unsigned long memmap_start)
408 * 'memmap_start' is the virtual address for the first "struct
409 * page" in this range of the vmemmap array. In the case of
410 * CONFIG_SPARSEMEM_VMEMMAP a page_to_pfn conversion is simple
411 * pointer arithmetic, so we can perform this to_vmem_altmap()
412 * conversion without concern for the initialization state of
413 * the struct page fields.
415 struct page *page = (struct page *) memmap_start;
416 struct dev_pagemap *pgmap;
419 * Unconditionally retrieve a dev_pagemap associated with the
420 * given physical address, this is only for use in the
421 * arch_{add|remove}_memory() for setting up and tearing down
425 pgmap = find_dev_pagemap(__pfn_to_phys(page_to_pfn(page)));
428 return pgmap ? pgmap->altmap : NULL;
430 #endif /* CONFIG_ZONE_DEVICE */