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 pfn_t
phys_to_pfn_t(phys_addr_t addr
, u64 flags
)
174 return __pfn_to_pfn_t(addr
>> PAGE_SHIFT
, flags
);
176 EXPORT_SYMBOL(phys_to_pfn_t
);
178 #ifdef CONFIG_ZONE_DEVICE
179 static DEFINE_MUTEX(pgmap_lock
);
180 static RADIX_TREE(pgmap_radix
, GFP_KERNEL
);
181 #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
182 #define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
186 struct percpu_ref
*ref
;
187 struct dev_pagemap pgmap
;
188 struct vmem_altmap altmap
;
191 void get_zone_device_page(struct page
*page
)
193 percpu_ref_get(page
->pgmap
->ref
);
195 EXPORT_SYMBOL(get_zone_device_page
);
197 void put_zone_device_page(struct page
*page
)
199 put_dev_pagemap(page
->pgmap
);
201 EXPORT_SYMBOL(put_zone_device_page
);
203 static void pgmap_radix_release(struct resource
*res
)
205 resource_size_t key
, align_start
, align_size
, align_end
;
207 align_start
= res
->start
& ~(SECTION_SIZE
- 1);
208 align_size
= ALIGN(resource_size(res
), SECTION_SIZE
);
209 align_end
= align_start
+ align_size
- 1;
211 mutex_lock(&pgmap_lock
);
212 for (key
= res
->start
; key
<= res
->end
; key
+= SECTION_SIZE
)
213 radix_tree_delete(&pgmap_radix
, key
>> PA_SECTION_SHIFT
);
214 mutex_unlock(&pgmap_lock
);
217 static unsigned long pfn_first(struct page_map
*page_map
)
219 struct dev_pagemap
*pgmap
= &page_map
->pgmap
;
220 const struct resource
*res
= &page_map
->res
;
221 struct vmem_altmap
*altmap
= pgmap
->altmap
;
224 pfn
= res
->start
>> PAGE_SHIFT
;
226 pfn
+= vmem_altmap_offset(altmap
);
230 static unsigned long pfn_end(struct page_map
*page_map
)
232 const struct resource
*res
= &page_map
->res
;
234 return (res
->start
+ resource_size(res
)) >> PAGE_SHIFT
;
237 #define for_each_device_pfn(pfn, map) \
238 for (pfn = pfn_first(map); pfn < pfn_end(map); pfn++)
240 static void devm_memremap_pages_release(struct device
*dev
, void *data
)
242 struct page_map
*page_map
= data
;
243 struct resource
*res
= &page_map
->res
;
244 resource_size_t align_start
, align_size
;
245 struct dev_pagemap
*pgmap
= &page_map
->pgmap
;
247 if (percpu_ref_tryget_live(pgmap
->ref
)) {
248 dev_WARN(dev
, "%s: page mapping is still live!\n", __func__
);
249 percpu_ref_put(pgmap
->ref
);
252 /* pages are dead and unused, undo the arch mapping */
253 align_start
= res
->start
& ~(SECTION_SIZE
- 1);
254 align_size
= ALIGN(resource_size(res
), SECTION_SIZE
);
255 arch_remove_memory(align_start
, align_size
);
256 pgmap_radix_release(res
);
257 dev_WARN_ONCE(dev
, pgmap
->altmap
&& pgmap
->altmap
->alloc
,
258 "%s: failed to free all reserved pages\n", __func__
);
261 /* assumes rcu_read_lock() held at entry */
262 struct dev_pagemap
*find_dev_pagemap(resource_size_t phys
)
264 struct page_map
*page_map
;
266 WARN_ON_ONCE(!rcu_read_lock_held());
268 page_map
= radix_tree_lookup(&pgmap_radix
, phys
>> PA_SECTION_SHIFT
);
269 return page_map
? &page_map
->pgmap
: NULL
;
273 * devm_memremap_pages - remap and provide memmap backing for the given resource
274 * @dev: hosting device for @res
275 * @res: "host memory" address range
276 * @ref: a live per-cpu reference count
277 * @altmap: optional descriptor for allocating the memmap from @res
280 * 1/ @ref must be 'live' on entry and 'dead' before devm_memunmap_pages() time
281 * (or devm release event).
283 * 2/ @res is expected to be a host memory range that could feasibly be
284 * treated as a "System RAM" range, i.e. not a device mmio range, but
285 * this is not enforced.
287 void *devm_memremap_pages(struct device
*dev
, struct resource
*res
,
288 struct percpu_ref
*ref
, struct vmem_altmap
*altmap
)
290 resource_size_t key
, align_start
, align_size
, align_end
;
291 struct dev_pagemap
*pgmap
;
292 struct page_map
*page_map
;
293 int error
, nid
, is_ram
;
296 align_start
= res
->start
& ~(SECTION_SIZE
- 1);
297 align_size
= ALIGN(res
->start
+ resource_size(res
), SECTION_SIZE
)
299 is_ram
= region_intersects(align_start
, align_size
,
300 IORESOURCE_SYSTEM_RAM
, IORES_DESC_NONE
);
302 if (is_ram
== REGION_MIXED
) {
303 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
305 return ERR_PTR(-ENXIO
);
308 if (is_ram
== REGION_INTERSECTS
)
309 return __va(res
->start
);
311 if (altmap
&& !IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP
)) {
312 dev_err(dev
, "%s: altmap requires CONFIG_SPARSEMEM_VMEMMAP=y\n",
314 return ERR_PTR(-ENXIO
);
318 return ERR_PTR(-EINVAL
);
320 page_map
= devres_alloc_node(devm_memremap_pages_release
,
321 sizeof(*page_map
), GFP_KERNEL
, dev_to_node(dev
));
323 return ERR_PTR(-ENOMEM
);
324 pgmap
= &page_map
->pgmap
;
326 memcpy(&page_map
->res
, res
, sizeof(*res
));
330 memcpy(&page_map
->altmap
, altmap
, sizeof(*altmap
));
331 pgmap
->altmap
= &page_map
->altmap
;
334 pgmap
->res
= &page_map
->res
;
336 mutex_lock(&pgmap_lock
);
338 align_end
= align_start
+ align_size
- 1;
339 for (key
= align_start
; key
<= align_end
; key
+= SECTION_SIZE
) {
340 struct dev_pagemap
*dup
;
343 dup
= find_dev_pagemap(key
);
346 dev_err(dev
, "%s: %pr collides with mapping for %s\n",
347 __func__
, res
, dev_name(dup
->dev
));
351 error
= radix_tree_insert(&pgmap_radix
, key
>> PA_SECTION_SHIFT
,
354 dev_err(dev
, "%s: failed: %d\n", __func__
, error
);
358 mutex_unlock(&pgmap_lock
);
362 nid
= dev_to_node(dev
);
366 error
= arch_add_memory(nid
, align_start
, align_size
, true);
370 for_each_device_pfn(pfn
, page_map
) {
371 struct page
*page
= pfn_to_page(pfn
);
374 * ZONE_DEVICE pages union ->lru with a ->pgmap back
375 * pointer. It is a bug if a ZONE_DEVICE page is ever
376 * freed or placed on a driver-private list. Seed the
377 * storage with LIST_POISON* values.
379 list_del(&page
->lru
);
382 devres_add(dev
, page_map
);
383 return __va(res
->start
);
387 pgmap_radix_release(res
);
388 devres_free(page_map
);
389 return ERR_PTR(error
);
391 EXPORT_SYMBOL(devm_memremap_pages
);
393 unsigned long vmem_altmap_offset(struct vmem_altmap
*altmap
)
395 /* number of pfns from base where pfn_to_page() is valid */
396 return altmap
->reserve
+ altmap
->free
;
399 void vmem_altmap_free(struct vmem_altmap
*altmap
, unsigned long nr_pfns
)
401 altmap
->alloc
-= nr_pfns
;
404 #ifdef CONFIG_SPARSEMEM_VMEMMAP
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_SPARSEMEM_VMEMMAP */
431 #endif /* CONFIG_ZONE_DEVICE */