]> git.proxmox.com Git - mirror_ubuntu-kernels.git/blame - kernel/memremap.c
libnvdimm, pmem: move request_queue allocation earlier in probe
[mirror_ubuntu-kernels.git] / kernel / memremap.c
CommitLineData
92281dee
DW
1/*
2 * Copyright(c) 2015 Intel Corporation. All rights reserved.
3 *
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.
7 *
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.
12 */
9476df7d
DW
13#include <linux/radix-tree.h>
14#include <linux/memremap.h>
7d3dcf26 15#include <linux/device.h>
92281dee 16#include <linux/types.h>
34c0fd54 17#include <linux/pfn_t.h>
92281dee
DW
18#include <linux/io.h>
19#include <linux/mm.h>
41e94a85 20#include <linux/memory_hotplug.h>
92281dee
DW
21
22#ifndef ioremap_cache
23/* temporary while we convert existing ioremap_cache users to memremap */
24__weak void __iomem *ioremap_cache(resource_size_t offset, unsigned long size)
25{
26 return ioremap(offset, size);
27}
28#endif
29
182475b7
DW
30static void *try_ram_remap(resource_size_t offset, size_t size)
31{
32 struct page *page = pfn_to_page(offset >> PAGE_SHIFT);
33
34 /* In the simple case just return the existing linear address */
35 if (!PageHighMem(page))
36 return __va(offset);
37 return NULL; /* fallback to ioremap_cache */
38}
39
92281dee
DW
40/**
41 * memremap() - remap an iomem_resource as cacheable memory
42 * @offset: iomem resource start address
43 * @size: size of remap
44 * @flags: either MEMREMAP_WB or MEMREMAP_WT
45 *
46 * memremap() is "ioremap" for cases where it is known that the resource
47 * being mapped does not have i/o side effects and the __iomem
48 * annotation is not applicable.
49 *
50 * MEMREMAP_WB - matches the default mapping for "System RAM" on
51 * the architecture. This is usually a read-allocate write-back cache.
52 * Morever, if MEMREMAP_WB is specified and the requested remap region is RAM
53 * memremap() will bypass establishing a new mapping and instead return
54 * a pointer into the direct map.
55 *
56 * MEMREMAP_WT - establish a mapping whereby writes either bypass the
57 * cache or are written through to memory and never exist in a
58 * cache-dirty state with respect to program visibility. Attempts to
59 * map "System RAM" with this mapping type will fail.
60 */
61void *memremap(resource_size_t offset, size_t size, unsigned long flags)
62{
63 int is_ram = region_intersects(offset, size, "System RAM");
64 void *addr = NULL;
65
66 if (is_ram == REGION_MIXED) {
67 WARN_ONCE(1, "memremap attempted on mixed range %pa size: %#lx\n",
68 &offset, (unsigned long) size);
69 return NULL;
70 }
71
72 /* Try all mapping types requested until one returns non-NULL */
73 if (flags & MEMREMAP_WB) {
74 flags &= ~MEMREMAP_WB;
75 /*
76 * MEMREMAP_WB is special in that it can be satisifed
77 * from the direct map. Some archs depend on the
78 * capability of memremap() to autodetect cases where
79 * the requested range is potentially in "System RAM"
80 */
81 if (is_ram == REGION_INTERSECTS)
182475b7
DW
82 addr = try_ram_remap(offset, size);
83 if (!addr)
92281dee
DW
84 addr = ioremap_cache(offset, size);
85 }
86
87 /*
88 * If we don't have a mapping yet and more request flags are
89 * pending then we will be attempting to establish a new virtual
90 * address mapping. Enforce that this mapping is not aliasing
91 * "System RAM"
92 */
93 if (!addr && is_ram == REGION_INTERSECTS && flags) {
94 WARN_ONCE(1, "memremap attempted on ram %pa size: %#lx\n",
95 &offset, (unsigned long) size);
96 return NULL;
97 }
98
99 if (!addr && (flags & MEMREMAP_WT)) {
100 flags &= ~MEMREMAP_WT;
101 addr = ioremap_wt(offset, size);
102 }
103
104 return addr;
105}
106EXPORT_SYMBOL(memremap);
107
108void memunmap(void *addr)
109{
110 if (is_vmalloc_addr(addr))
111 iounmap((void __iomem *) addr);
112}
113EXPORT_SYMBOL(memunmap);
7d3dcf26
CH
114
115static void devm_memremap_release(struct device *dev, void *res)
116{
117 memunmap(res);
118}
119
120static int devm_memremap_match(struct device *dev, void *res, void *match_data)
121{
122 return *(void **)res == match_data;
123}
124
125void *devm_memremap(struct device *dev, resource_size_t offset,
126 size_t size, unsigned long flags)
127{
128 void **ptr, *addr;
129
538ea4aa
DW
130 ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
131 dev_to_node(dev));
7d3dcf26 132 if (!ptr)
b36f4761 133 return ERR_PTR(-ENOMEM);
7d3dcf26
CH
134
135 addr = memremap(offset, size, flags);
136 if (addr) {
137 *ptr = addr;
138 devres_add(dev, ptr);
139 } else
140 devres_free(ptr);
141
142 return addr;
143}
144EXPORT_SYMBOL(devm_memremap);
145
146void devm_memunmap(struct device *dev, void *addr)
147{
d741314f
DW
148 WARN_ON(devres_release(dev, devm_memremap_release,
149 devm_memremap_match, addr));
7d3dcf26
CH
150}
151EXPORT_SYMBOL(devm_memunmap);
41e94a85 152
34c0fd54
DW
153pfn_t phys_to_pfn_t(dma_addr_t addr, unsigned long flags)
154{
155 return __pfn_to_pfn_t(addr >> PAGE_SHIFT, flags);
156}
157EXPORT_SYMBOL(phys_to_pfn_t);
158
41e94a85 159#ifdef CONFIG_ZONE_DEVICE
9476df7d
DW
160static DEFINE_MUTEX(pgmap_lock);
161static RADIX_TREE(pgmap_radix, GFP_KERNEL);
162#define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1)
163#define SECTION_SIZE (1UL << PA_SECTION_SHIFT)
164
41e94a85
CH
165struct page_map {
166 struct resource res;
9476df7d
DW
167 struct percpu_ref *ref;
168 struct dev_pagemap pgmap;
4b94ffdc 169 struct vmem_altmap altmap;
41e94a85
CH
170};
171
9476df7d
DW
172static void pgmap_radix_release(struct resource *res)
173{
174 resource_size_t key;
175
176 mutex_lock(&pgmap_lock);
177 for (key = res->start; key <= res->end; key += SECTION_SIZE)
178 radix_tree_delete(&pgmap_radix, key >> PA_SECTION_SHIFT);
179 mutex_unlock(&pgmap_lock);
180}
181
182static void devm_memremap_pages_release(struct device *dev, void *data)
41e94a85 183{
9476df7d
DW
184 struct page_map *page_map = data;
185 struct resource *res = &page_map->res;
186 resource_size_t align_start, align_size;
4b94ffdc 187 struct dev_pagemap *pgmap = &page_map->pgmap;
9476df7d
DW
188
189 pgmap_radix_release(res);
41e94a85
CH
190
191 /* pages are dead and unused, undo the arch mapping */
9476df7d
DW
192 align_start = res->start & ~(SECTION_SIZE - 1);
193 align_size = ALIGN(resource_size(res), SECTION_SIZE);
194 arch_remove_memory(align_start, align_size);
4b94ffdc
DW
195 dev_WARN_ONCE(dev, pgmap->altmap && pgmap->altmap->alloc,
196 "%s: failed to free all reserved pages\n", __func__);
9476df7d
DW
197}
198
199/* assumes rcu_read_lock() held at entry */
200struct dev_pagemap *find_dev_pagemap(resource_size_t phys)
201{
202 struct page_map *page_map;
203
204 WARN_ON_ONCE(!rcu_read_lock_held());
205
206 page_map = radix_tree_lookup(&pgmap_radix, phys >> PA_SECTION_SHIFT);
207 return page_map ? &page_map->pgmap : NULL;
41e94a85
CH
208}
209
4b94ffdc
DW
210/**
211 * devm_memremap_pages - remap and provide memmap backing for the given resource
212 * @dev: hosting device for @res
213 * @res: "host memory" address range
214 * @altmap: optional descriptor for allocating the memmap from @res
215 *
216 * Note, the expectation is that @res is a host memory range that could
217 * feasibly be treated as a "System RAM" range, i.e. not a device mmio
218 * range, but this is not enforced.
219 */
220void *devm_memremap_pages(struct device *dev, struct resource *res,
221 struct vmem_altmap *altmap)
41e94a85
CH
222{
223 int is_ram = region_intersects(res->start, resource_size(res),
224 "System RAM");
9476df7d 225 resource_size_t key, align_start, align_size;
4b94ffdc 226 struct dev_pagemap *pgmap;
41e94a85
CH
227 struct page_map *page_map;
228 int error, nid;
229
230 if (is_ram == REGION_MIXED) {
231 WARN_ONCE(1, "%s attempted on mixed region %pr\n",
232 __func__, res);
233 return ERR_PTR(-ENXIO);
234 }
235
236 if (is_ram == REGION_INTERSECTS)
237 return __va(res->start);
238
4b94ffdc
DW
239 if (altmap && !IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) {
240 dev_err(dev, "%s: altmap requires CONFIG_SPARSEMEM_VMEMMAP=y\n",
241 __func__);
242 return ERR_PTR(-ENXIO);
243 }
244
538ea4aa
DW
245 page_map = devres_alloc_node(devm_memremap_pages_release,
246 sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
41e94a85
CH
247 if (!page_map)
248 return ERR_PTR(-ENOMEM);
4b94ffdc 249 pgmap = &page_map->pgmap;
41e94a85
CH
250
251 memcpy(&page_map->res, res, sizeof(*res));
252
4b94ffdc
DW
253 pgmap->dev = dev;
254 if (altmap) {
255 memcpy(&page_map->altmap, altmap, sizeof(*altmap));
256 pgmap->altmap = &page_map->altmap;
257 }
258 pgmap->res = &page_map->res;
259
9476df7d
DW
260 mutex_lock(&pgmap_lock);
261 error = 0;
262 for (key = res->start; key <= res->end; key += SECTION_SIZE) {
263 struct dev_pagemap *dup;
264
265 rcu_read_lock();
266 dup = find_dev_pagemap(key);
267 rcu_read_unlock();
268 if (dup) {
269 dev_err(dev, "%s: %pr collides with mapping for %s\n",
270 __func__, res, dev_name(dup->dev));
271 error = -EBUSY;
272 break;
273 }
274 error = radix_tree_insert(&pgmap_radix, key >> PA_SECTION_SHIFT,
275 page_map);
276 if (error) {
277 dev_err(dev, "%s: failed: %d\n", __func__, error);
278 break;
279 }
280 }
281 mutex_unlock(&pgmap_lock);
282 if (error)
283 goto err_radix;
284
41e94a85
CH
285 nid = dev_to_node(dev);
286 if (nid < 0)
7eff93b7 287 nid = numa_mem_id();
41e94a85 288
9476df7d
DW
289 align_start = res->start & ~(SECTION_SIZE - 1);
290 align_size = ALIGN(resource_size(res), SECTION_SIZE);
291 error = arch_add_memory(nid, align_start, align_size, true);
292 if (error)
293 goto err_add_memory;
41e94a85
CH
294
295 devres_add(dev, page_map);
296 return __va(res->start);
9476df7d
DW
297
298 err_add_memory:
299 err_radix:
300 pgmap_radix_release(res);
301 devres_free(page_map);
302 return ERR_PTR(error);
41e94a85
CH
303}
304EXPORT_SYMBOL(devm_memremap_pages);
4b94ffdc
DW
305
306unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
307{
308 /* number of pfns from base where pfn_to_page() is valid */
309 return altmap->reserve + altmap->free;
310}
311
312void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns)
313{
314 altmap->alloc -= nr_pfns;
315}
316
317#ifdef CONFIG_SPARSEMEM_VMEMMAP
318struct vmem_altmap *to_vmem_altmap(unsigned long memmap_start)
319{
320 /*
321 * 'memmap_start' is the virtual address for the first "struct
322 * page" in this range of the vmemmap array. In the case of
323 * CONFIG_SPARSE_VMEMMAP a page_to_pfn conversion is simple
324 * pointer arithmetic, so we can perform this to_vmem_altmap()
325 * conversion without concern for the initialization state of
326 * the struct page fields.
327 */
328 struct page *page = (struct page *) memmap_start;
329 struct dev_pagemap *pgmap;
330
331 /*
332 * Uncoditionally retrieve a dev_pagemap associated with the
333 * given physical address, this is only for use in the
334 * arch_{add|remove}_memory() for setting up and tearing down
335 * the memmap.
336 */
337 rcu_read_lock();
338 pgmap = find_dev_pagemap(__pfn_to_phys(page_to_pfn(page)));
339 rcu_read_unlock();
340
341 return pgmap ? pgmap->altmap : NULL;
342}
343#endif /* CONFIG_SPARSEMEM_VMEMMAP */
41e94a85 344#endif /* CONFIG_ZONE_DEVICE */