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 #ifndef arch_memremap_can_ram_remap
38 static bool arch_memremap_can_ram_remap(resource_size_t offset, size_t size,
45 static void *try_ram_remap(resource_size_t offset, size_t size,
48 unsigned long pfn = PHYS_PFN(offset);
50 /* In the simple case just return the existing linear address */
51 if (pfn_valid(pfn) && !PageHighMem(pfn_to_page(pfn)) &&
52 arch_memremap_can_ram_remap(offset, size, flags))
55 return NULL; /* fallback to arch_memremap_wb */
59 * memremap() - remap an iomem_resource as cacheable memory
60 * @offset: iomem resource start address
61 * @size: size of remap
62 * @flags: any of MEMREMAP_WB, MEMREMAP_WT, MEMREMAP_WC,
63 * MEMREMAP_ENC, MEMREMAP_DEC
65 * memremap() is "ioremap" for cases where it is known that the resource
66 * being mapped does not have i/o side effects and the __iomem
67 * annotation is not applicable. In the case of multiple flags, the different
68 * mapping types will be attempted in the order listed below until one of
71 * MEMREMAP_WB - matches the default mapping for System RAM on
72 * the architecture. This is usually a read-allocate write-back cache.
73 * Morever, if MEMREMAP_WB is specified and the requested remap region is RAM
74 * memremap() will bypass establishing a new mapping and instead return
75 * a pointer into the direct map.
77 * MEMREMAP_WT - establish a mapping whereby writes either bypass the
78 * cache or are written through to memory and never exist in a
79 * cache-dirty state with respect to program visibility. Attempts to
80 * map System RAM with this mapping type will fail.
82 * MEMREMAP_WC - establish a writecombine mapping, whereby writes may
83 * be coalesced together (e.g. in the CPU's write buffers), but is otherwise
84 * uncached. Attempts to map System RAM with this mapping type will fail.
86 void *memremap(resource_size_t offset, size_t size, unsigned long flags)
88 int is_ram = region_intersects(offset, size,
89 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
95 if (is_ram == REGION_MIXED) {
96 WARN_ONCE(1, "memremap attempted on mixed range %pa size: %#lx\n",
97 &offset, (unsigned long) size);
101 /* Try all mapping types requested until one returns non-NULL */
102 if (flags & MEMREMAP_WB) {
104 * MEMREMAP_WB is special in that it can be satisifed
105 * from the direct map. Some archs depend on the
106 * capability of memremap() to autodetect cases where
107 * the requested range is potentially in System RAM.
109 if (is_ram == REGION_INTERSECTS)
110 addr = try_ram_remap(offset, size, flags);
112 addr = arch_memremap_wb(offset, size);
116 * If we don't have a mapping yet and other request flags are
117 * present then we will be attempting to establish a new virtual
118 * address mapping. Enforce that this mapping is not aliasing
121 if (!addr && is_ram == REGION_INTERSECTS && flags != MEMREMAP_WB) {
122 WARN_ONCE(1, "memremap attempted on ram %pa size: %#lx\n",
123 &offset, (unsigned long) size);
127 if (!addr && (flags & MEMREMAP_WT))
128 addr = ioremap_wt(offset, size);
130 if (!addr && (flags & MEMREMAP_WC))
131 addr = ioremap_wc(offset, size);
135 EXPORT_SYMBOL(memremap);
137 void memunmap(void *addr)
139 if (is_vmalloc_addr(addr))
140 iounmap((void __iomem *) addr);
142 EXPORT_SYMBOL(memunmap);
144 static void devm_memremap_release(struct device *dev, void *res)
146 memunmap(*(void **)res);
149 static int devm_memremap_match(struct device *dev, void *res, void *match_data)
151 return *(void **)res == match_data;
154 void *devm_memremap(struct device *dev, resource_size_t offset,
155 size_t size, unsigned long flags)
159 ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
162 return ERR_PTR(-ENOMEM);
164 addr = memremap(offset, size, flags);
167 devres_add(dev, ptr);
170 return ERR_PTR(-ENXIO);
175 EXPORT_SYMBOL(devm_memremap);
177 void devm_memunmap(struct device *dev, void *addr)
179 WARN_ON(devres_release(dev, devm_memremap_release,
180 devm_memremap_match, addr));
182 EXPORT_SYMBOL(devm_memunmap);
184 #ifdef CONFIG_ZONE_DEVICE
185 static DEFINE_MUTEX(pgmap_lock);
186 static RADIX_TREE(pgmap_radix, GFP_KERNEL);
187 #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
188 #define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
192 struct percpu_ref *ref;
193 struct dev_pagemap pgmap;
194 struct vmem_altmap altmap;
197 static unsigned long order_at(struct resource *res, unsigned long pgoff)
199 unsigned long phys_pgoff = PHYS_PFN(res->start) + pgoff;
200 unsigned long nr_pages, mask;
202 nr_pages = PHYS_PFN(resource_size(res));
203 if (nr_pages == pgoff)
207 * What is the largest aligned power-of-2 range available from
208 * this resource pgoff to the end of the resource range,
209 * considering the alignment of the current pgoff?
211 mask = phys_pgoff | rounddown_pow_of_two(nr_pages - pgoff);
215 return find_first_bit(&mask, BITS_PER_LONG);
218 #define foreach_order_pgoff(res, order, pgoff) \
219 for (pgoff = 0, order = order_at((res), pgoff); order < ULONG_MAX; \
220 pgoff += 1UL << order, order = order_at((res), pgoff))
222 static void pgmap_radix_release(struct resource *res)
224 unsigned long pgoff, order;
226 mutex_lock(&pgmap_lock);
227 foreach_order_pgoff(res, order, pgoff)
228 radix_tree_delete(&pgmap_radix, PHYS_PFN(res->start) + pgoff);
229 mutex_unlock(&pgmap_lock);
234 static unsigned long pfn_first(struct page_map *page_map)
236 struct dev_pagemap *pgmap = &page_map->pgmap;
237 const struct resource *res = &page_map->res;
238 struct vmem_altmap *altmap = pgmap->altmap;
241 pfn = res->start >> PAGE_SHIFT;
243 pfn += vmem_altmap_offset(altmap);
247 static unsigned long pfn_end(struct page_map *page_map)
249 const struct resource *res = &page_map->res;
251 return (res->start + resource_size(res)) >> PAGE_SHIFT;
254 #define for_each_device_pfn(pfn, map) \
255 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn++)
257 static void devm_memremap_pages_release(struct device *dev, void *data)
259 struct page_map *page_map = data;
260 struct resource *res = &page_map->res;
261 resource_size_t align_start, align_size;
262 struct dev_pagemap *pgmap = &page_map->pgmap;
265 for_each_device_pfn(pfn, page_map)
266 put_page(pfn_to_page(pfn));
268 if (percpu_ref_tryget_live(pgmap->ref)) {
269 dev_WARN(dev, "%s: page mapping is still live!\n", __func__);
270 percpu_ref_put(pgmap->ref);
273 /* pages are dead and unused, undo the arch mapping */
274 align_start = res->start & ~(SECTION_SIZE - 1);
275 align_size = ALIGN(resource_size(res), SECTION_SIZE);
278 arch_remove_memory(align_start, align_size);
281 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
282 pgmap_radix_release(res);
283 dev_WARN_ONCE(dev, pgmap->altmap && pgmap->altmap->alloc,
284 "%s: failed to free all reserved pages\n", __func__);
287 /* assumes rcu_read_lock() held at entry */
288 struct dev_pagemap *find_dev_pagemap(resource_size_t phys)
290 struct page_map *page_map;
292 WARN_ON_ONCE(!rcu_read_lock_held());
294 page_map = radix_tree_lookup(&pgmap_radix, PHYS_PFN(phys));
295 return page_map ? &page_map->pgmap : NULL;
299 * devm_memremap_pages - remap and provide memmap backing for the given resource
300 * @dev: hosting device for @res
301 * @res: "host memory" address range
302 * @ref: a live per-cpu reference count
303 * @altmap: optional descriptor for allocating the memmap from @res
306 * 1/ @ref must be 'live' on entry and 'dead' before devm_memunmap_pages() time
307 * (or devm release event). The expected order of events is that @ref has
308 * been through percpu_ref_kill() before devm_memremap_pages_release(). The
309 * wait for the completion of all references being dropped and
310 * percpu_ref_exit() must occur after devm_memremap_pages_release().
312 * 2/ @res is expected to be a host memory range that could feasibly be
313 * treated as a "System RAM" range, i.e. not a device mmio range, but
314 * this is not enforced.
316 void *devm_memremap_pages(struct device *dev, struct resource *res,
317 struct percpu_ref *ref, struct vmem_altmap *altmap)
319 resource_size_t align_start, align_size, align_end;
320 unsigned long pfn, pgoff, order;
321 pgprot_t pgprot = PAGE_KERNEL;
322 struct dev_pagemap *pgmap;
323 struct page_map *page_map;
324 int error, nid, is_ram;
326 align_start = res->start & ~(SECTION_SIZE - 1);
327 align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE)
329 is_ram = region_intersects(align_start, align_size,
330 IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE);
332 if (is_ram == REGION_MIXED) {
333 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
335 return ERR_PTR(-ENXIO);
338 if (is_ram == REGION_INTERSECTS)
339 return __va(res->start);
342 return ERR_PTR(-EINVAL);
344 page_map = devres_alloc_node(devm_memremap_pages_release,
345 sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
347 return ERR_PTR(-ENOMEM);
348 pgmap = &page_map->pgmap;
350 memcpy(&page_map->res, res, sizeof(*res));
354 memcpy(&page_map->altmap, altmap, sizeof(*altmap));
355 pgmap->altmap = &page_map->altmap;
358 pgmap->res = &page_map->res;
360 mutex_lock(&pgmap_lock);
362 align_end = align_start + align_size - 1;
364 foreach_order_pgoff(res, order, pgoff) {
365 struct dev_pagemap *dup;
368 dup = find_dev_pagemap(res->start + PFN_PHYS(pgoff));
371 dev_err(dev, "%s: %pr collides with mapping for %s\n",
372 __func__, res, dev_name(dup->dev));
376 error = __radix_tree_insert(&pgmap_radix,
377 PHYS_PFN(res->start) + pgoff, order, page_map);
379 dev_err(dev, "%s: failed: %d\n", __func__, error);
383 mutex_unlock(&pgmap_lock);
387 nid = dev_to_node(dev);
391 error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(align_start), 0,
397 error = arch_add_memory(nid, align_start, align_size, false);
399 move_pfn_range_to_zone(&NODE_DATA(nid)->node_zones[ZONE_DEVICE],
400 align_start >> PAGE_SHIFT,
401 align_size >> PAGE_SHIFT);
406 for_each_device_pfn(pfn, page_map) {
407 struct page *page = pfn_to_page(pfn);
410 * ZONE_DEVICE pages union ->lru with a ->pgmap back
411 * pointer. It is a bug if a ZONE_DEVICE page is ever
412 * freed or placed on a driver-private list. Seed the
413 * storage with LIST_POISON* values.
415 list_del(&page->lru);
419 devres_add(dev, page_map);
420 return __va(res->start);
423 untrack_pfn(NULL, PHYS_PFN(align_start), align_size);
426 pgmap_radix_release(res);
427 devres_free(page_map);
428 return ERR_PTR(error);
430 EXPORT_SYMBOL(devm_memremap_pages);
432 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
434 /* number of pfns from base where pfn_to_page() is valid */
435 return altmap->reserve + altmap->free;
438 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
440 altmap->alloc -= nr_pfns;
443 struct vmem_altmap *to_vmem_altmap(unsigned long memmap_start)
446 * 'memmap_start' is the virtual address for the first "struct
447 * page" in this range of the vmemmap array. In the case of
448 * CONFIG_SPARSEMEM_VMEMMAP a page_to_pfn conversion is simple
449 * pointer arithmetic, so we can perform this to_vmem_altmap()
450 * conversion without concern for the initialization state of
451 * the struct page fields.
453 struct page *page = (struct page *) memmap_start;
454 struct dev_pagemap *pgmap;
457 * Unconditionally retrieve a dev_pagemap associated with the
458 * given physical address, this is only for use in the
459 * arch_{add|remove}_memory() for setting up and tearing down
463 pgmap = find_dev_pagemap(__pfn_to_phys(page_to_pfn(page)));
466 return pgmap ? pgmap->altmap : NULL;
468 #endif /* CONFIG_ZONE_DEVICE */