Commit | Line | Data |
---|---|---|
5981690d DW |
1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | /* Copyright(c) 2015 Intel Corporation. All rights reserved. */ | |
7d3dcf26 | 3 | #include <linux/device.h> |
92281dee | 4 | #include <linux/io.h> |
0207df4f | 5 | #include <linux/kasan.h> |
41e94a85 | 6 | #include <linux/memory_hotplug.h> |
bcfa4b72 MW |
7 | #include <linux/mm.h> |
8 | #include <linux/pfn_t.h> | |
5042db43 JG |
9 | #include <linux/swap.h> |
10 | #include <linux/swapops.h> | |
bcfa4b72 | 11 | #include <linux/types.h> |
e7638488 | 12 | #include <linux/wait_bit.h> |
bcfa4b72 | 13 | #include <linux/xarray.h> |
92281dee | 14 | |
bcfa4b72 | 15 | static DEFINE_XARRAY(pgmap_array); |
9476df7d DW |
16 | #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1) |
17 | #define SECTION_SIZE (1UL << PA_SECTION_SHIFT) | |
18 | ||
f6a55e1a CH |
19 | #ifdef CONFIG_DEV_PAGEMAP_OPS |
20 | DEFINE_STATIC_KEY_FALSE(devmap_managed_key); | |
21 | EXPORT_SYMBOL(devmap_managed_key); | |
22 | static atomic_t devmap_managed_enable; | |
23 | ||
24 | static void devmap_managed_enable_put(void *data) | |
25 | { | |
26 | if (atomic_dec_and_test(&devmap_managed_enable)) | |
27 | static_branch_disable(&devmap_managed_key); | |
28 | } | |
29 | ||
30 | static int devmap_managed_enable_get(struct device *dev, struct dev_pagemap *pgmap) | |
31 | { | |
32 | if (!pgmap->ops->page_free) { | |
33 | WARN(1, "Missing page_free method\n"); | |
34 | return -EINVAL; | |
35 | } | |
36 | ||
37 | if (atomic_inc_return(&devmap_managed_enable) == 1) | |
38 | static_branch_enable(&devmap_managed_key); | |
39 | return devm_add_action_or_reset(dev, devmap_managed_enable_put, NULL); | |
40 | } | |
41 | #else | |
42 | static int devmap_managed_enable_get(struct device *dev, struct dev_pagemap *pgmap) | |
43 | { | |
44 | return -EINVAL; | |
45 | } | |
46 | #endif /* CONFIG_DEV_PAGEMAP_OPS */ | |
47 | ||
bcfa4b72 | 48 | static void pgmap_array_delete(struct resource *res) |
ab1b597e | 49 | { |
bcfa4b72 MW |
50 | xa_store_range(&pgmap_array, PHYS_PFN(res->start), PHYS_PFN(res->end), |
51 | NULL, GFP_KERNEL); | |
ab1b597e | 52 | synchronize_rcu(); |
9476df7d DW |
53 | } |
54 | ||
e7744aa2 | 55 | static unsigned long pfn_first(struct dev_pagemap *pgmap) |
5c2c2587 | 56 | { |
e7744aa2 LG |
57 | const struct resource *res = &pgmap->res; |
58 | struct vmem_altmap *altmap = &pgmap->altmap; | |
5c2c2587 DW |
59 | unsigned long pfn; |
60 | ||
61 | pfn = res->start >> PAGE_SHIFT; | |
e7744aa2 | 62 | if (pgmap->altmap_valid) |
5c2c2587 DW |
63 | pfn += vmem_altmap_offset(altmap); |
64 | return pfn; | |
65 | } | |
66 | ||
e7744aa2 | 67 | static unsigned long pfn_end(struct dev_pagemap *pgmap) |
5c2c2587 | 68 | { |
e7744aa2 | 69 | const struct resource *res = &pgmap->res; |
5c2c2587 DW |
70 | |
71 | return (res->start + resource_size(res)) >> PAGE_SHIFT; | |
72 | } | |
73 | ||
949b9325 DW |
74 | static unsigned long pfn_next(unsigned long pfn) |
75 | { | |
76 | if (pfn % 1024 == 0) | |
77 | cond_resched(); | |
78 | return pfn + 1; | |
79 | } | |
80 | ||
5c2c2587 | 81 | #define for_each_device_pfn(pfn, map) \ |
949b9325 | 82 | for (pfn = pfn_first(map); pfn < pfn_end(map); pfn = pfn_next(pfn)) |
5c2c2587 | 83 | |
e8d51348 | 84 | static void devm_memremap_pages_release(void *data) |
41e94a85 | 85 | { |
e7744aa2 | 86 | struct dev_pagemap *pgmap = data; |
e8d51348 | 87 | struct device *dev = pgmap->dev; |
e7744aa2 | 88 | struct resource *res = &pgmap->res; |
9476df7d | 89 | resource_size_t align_start, align_size; |
71389703 | 90 | unsigned long pfn; |
2c2a5af6 | 91 | int nid; |
71389703 | 92 | |
d8668bb0 | 93 | pgmap->ops->kill(pgmap); |
e7744aa2 | 94 | for_each_device_pfn(pfn, pgmap) |
71389703 | 95 | put_page(pfn_to_page(pfn)); |
d8668bb0 | 96 | pgmap->ops->cleanup(pgmap); |
9476df7d | 97 | |
41e94a85 | 98 | /* pages are dead and unused, undo the arch mapping */ |
9476df7d | 99 | align_start = res->start & ~(SECTION_SIZE - 1); |
10a0cd6e JS |
100 | align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE) |
101 | - align_start; | |
b5d24fda | 102 | |
2c2a5af6 OS |
103 | nid = page_to_nid(pfn_to_page(align_start >> PAGE_SHIFT)); |
104 | ||
f931ab47 | 105 | mem_hotplug_begin(); |
69324b8f DW |
106 | if (pgmap->type == MEMORY_DEVICE_PRIVATE) { |
107 | pfn = align_start >> PAGE_SHIFT; | |
108 | __remove_pages(page_zone(pfn_to_page(pfn)), pfn, | |
109 | align_size >> PAGE_SHIFT, NULL); | |
110 | } else { | |
2c2a5af6 | 111 | arch_remove_memory(nid, align_start, align_size, |
69324b8f DW |
112 | pgmap->altmap_valid ? &pgmap->altmap : NULL); |
113 | kasan_remove_zero_shadow(__va(align_start), align_size); | |
114 | } | |
f931ab47 | 115 | mem_hotplug_done(); |
b5d24fda | 116 | |
9049771f | 117 | untrack_pfn(NULL, PHYS_PFN(align_start), align_size); |
bcfa4b72 | 118 | pgmap_array_delete(res); |
e7744aa2 LG |
119 | dev_WARN_ONCE(dev, pgmap->altmap.alloc, |
120 | "%s: failed to free all reserved pages\n", __func__); | |
9476df7d DW |
121 | } |
122 | ||
4b94ffdc DW |
123 | /** |
124 | * devm_memremap_pages - remap and provide memmap backing for the given resource | |
125 | * @dev: hosting device for @res | |
a95c90f1 | 126 | * @pgmap: pointer to a struct dev_pagemap |
4b94ffdc | 127 | * |
5c2c2587 | 128 | * Notes: |
1e240e8d CH |
129 | * 1/ At a minimum the res, ref and type and ops members of @pgmap must be |
130 | * initialized by the caller before passing it to this function | |
e8d51348 CH |
131 | * |
132 | * 2/ The altmap field may optionally be initialized, in which case altmap_valid | |
133 | * must be set to true | |
134 | * | |
50f44ee7 DW |
135 | * 3/ pgmap->ref must be 'live' on entry and will be killed and reaped |
136 | * at devm_memremap_pages_release() time, or if this routine fails. | |
5c2c2587 | 137 | * |
e8d51348 | 138 | * 4/ res is expected to be a host memory range that could feasibly be |
5c2c2587 DW |
139 | * treated as a "System RAM" range, i.e. not a device mmio range, but |
140 | * this is not enforced. | |
4b94ffdc | 141 | */ |
e8d51348 | 142 | void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap) |
41e94a85 | 143 | { |
ab1b597e | 144 | resource_size_t align_start, align_size, align_end; |
e8d51348 CH |
145 | struct vmem_altmap *altmap = pgmap->altmap_valid ? |
146 | &pgmap->altmap : NULL; | |
949b9325 | 147 | struct resource *res = &pgmap->res; |
966cf44f | 148 | struct dev_pagemap *conflict_pgmap; |
940519f0 MH |
149 | struct mhp_restrictions restrictions = { |
150 | /* | |
151 | * We do not want any optional features only our own memmap | |
152 | */ | |
153 | .altmap = altmap, | |
154 | }; | |
9049771f | 155 | pgprot_t pgprot = PAGE_KERNEL; |
949b9325 | 156 | int error, nid, is_ram; |
f6a55e1a | 157 | bool need_devmap_managed = true; |
5f29a77c | 158 | |
3ed2dcdf CH |
159 | switch (pgmap->type) { |
160 | case MEMORY_DEVICE_PRIVATE: | |
161 | if (!IS_ENABLED(CONFIG_DEVICE_PRIVATE)) { | |
162 | WARN(1, "Device private memory not supported\n"); | |
163 | return ERR_PTR(-EINVAL); | |
164 | } | |
897e6365 CH |
165 | if (!pgmap->ops || !pgmap->ops->migrate_to_ram) { |
166 | WARN(1, "Missing migrate_to_ram method\n"); | |
167 | return ERR_PTR(-EINVAL); | |
168 | } | |
3ed2dcdf CH |
169 | break; |
170 | case MEMORY_DEVICE_FS_DAX: | |
171 | if (!IS_ENABLED(CONFIG_ZONE_DEVICE) || | |
172 | IS_ENABLED(CONFIG_FS_DAX_LIMITED)) { | |
173 | WARN(1, "File system DAX not supported\n"); | |
174 | return ERR_PTR(-EINVAL); | |
175 | } | |
176 | break; | |
177 | case MEMORY_DEVICE_DEVDAX: | |
178 | case MEMORY_DEVICE_PCI_P2PDMA: | |
f6a55e1a | 179 | need_devmap_managed = false; |
3ed2dcdf CH |
180 | break; |
181 | default: | |
182 | WARN(1, "Invalid pgmap type %d\n", pgmap->type); | |
183 | break; | |
184 | } | |
185 | ||
1e240e8d CH |
186 | if (!pgmap->ref || !pgmap->ops || !pgmap->ops->kill || |
187 | !pgmap->ops->cleanup) { | |
50f44ee7 | 188 | WARN(1, "Missing reference count teardown definition\n"); |
a95c90f1 | 189 | return ERR_PTR(-EINVAL); |
50f44ee7 | 190 | } |
a95c90f1 | 191 | |
f6a55e1a CH |
192 | if (need_devmap_managed) { |
193 | error = devmap_managed_enable_get(dev, pgmap); | |
194 | if (error) | |
195 | return ERR_PTR(error); | |
196 | } | |
197 | ||
5f29a77c DW |
198 | align_start = res->start & ~(SECTION_SIZE - 1); |
199 | align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE) | |
200 | - align_start; | |
15d36fec DJ |
201 | align_end = align_start + align_size - 1; |
202 | ||
203 | conflict_pgmap = get_dev_pagemap(PHYS_PFN(align_start), NULL); | |
204 | if (conflict_pgmap) { | |
205 | dev_WARN(dev, "Conflicting mapping in same section\n"); | |
206 | put_dev_pagemap(conflict_pgmap); | |
50f44ee7 DW |
207 | error = -ENOMEM; |
208 | goto err_array; | |
15d36fec DJ |
209 | } |
210 | ||
211 | conflict_pgmap = get_dev_pagemap(PHYS_PFN(align_end), NULL); | |
212 | if (conflict_pgmap) { | |
213 | dev_WARN(dev, "Conflicting mapping in same section\n"); | |
214 | put_dev_pagemap(conflict_pgmap); | |
50f44ee7 DW |
215 | error = -ENOMEM; |
216 | goto err_array; | |
15d36fec DJ |
217 | } |
218 | ||
d37a14bb LT |
219 | is_ram = region_intersects(align_start, align_size, |
220 | IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE); | |
41e94a85 | 221 | |
06489cfb DW |
222 | if (is_ram != REGION_DISJOINT) { |
223 | WARN_ONCE(1, "%s attempted on %s region %pr\n", __func__, | |
224 | is_ram == REGION_MIXED ? "mixed" : "ram", res); | |
a95c90f1 DW |
225 | error = -ENXIO; |
226 | goto err_array; | |
41e94a85 CH |
227 | } |
228 | ||
4b94ffdc | 229 | pgmap->dev = dev; |
4b94ffdc | 230 | |
bcfa4b72 MW |
231 | error = xa_err(xa_store_range(&pgmap_array, PHYS_PFN(res->start), |
232 | PHYS_PFN(res->end), pgmap, GFP_KERNEL)); | |
9476df7d | 233 | if (error) |
bcfa4b72 | 234 | goto err_array; |
9476df7d | 235 | |
41e94a85 CH |
236 | nid = dev_to_node(dev); |
237 | if (nid < 0) | |
7eff93b7 | 238 | nid = numa_mem_id(); |
41e94a85 | 239 | |
9049771f DW |
240 | error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(align_start), 0, |
241 | align_size); | |
242 | if (error) | |
243 | goto err_pfn_remap; | |
244 | ||
f931ab47 | 245 | mem_hotplug_begin(); |
69324b8f DW |
246 | |
247 | /* | |
248 | * For device private memory we call add_pages() as we only need to | |
249 | * allocate and initialize struct page for the device memory. More- | |
250 | * over the device memory is un-accessible thus we do not want to | |
251 | * create a linear mapping for the memory like arch_add_memory() | |
252 | * would do. | |
253 | * | |
254 | * For all other device memory types, which are accessible by | |
255 | * the CPU, we do want the linear mapping and thus use | |
256 | * arch_add_memory(). | |
257 | */ | |
258 | if (pgmap->type == MEMORY_DEVICE_PRIVATE) { | |
259 | error = add_pages(nid, align_start >> PAGE_SHIFT, | |
940519f0 | 260 | align_size >> PAGE_SHIFT, &restrictions); |
69324b8f DW |
261 | } else { |
262 | error = kasan_add_zero_shadow(__va(align_start), align_size); | |
263 | if (error) { | |
264 | mem_hotplug_done(); | |
265 | goto err_kasan; | |
266 | } | |
267 | ||
940519f0 MH |
268 | error = arch_add_memory(nid, align_start, align_size, |
269 | &restrictions); | |
69324b8f DW |
270 | } |
271 | ||
272 | if (!error) { | |
273 | struct zone *zone; | |
274 | ||
275 | zone = &NODE_DATA(nid)->node_zones[ZONE_DEVICE]; | |
276 | move_pfn_range_to_zone(zone, align_start >> PAGE_SHIFT, | |
277 | align_size >> PAGE_SHIFT, altmap); | |
0207df4f AR |
278 | } |
279 | ||
f931ab47 | 280 | mem_hotplug_done(); |
9476df7d DW |
281 | if (error) |
282 | goto err_add_memory; | |
41e94a85 | 283 | |
966cf44f AD |
284 | /* |
285 | * Initialization of the pages has been deferred until now in order | |
286 | * to allow us to do the work while not holding the hotplug lock. | |
287 | */ | |
288 | memmap_init_zone_device(&NODE_DATA(nid)->node_zones[ZONE_DEVICE], | |
289 | align_start >> PAGE_SHIFT, | |
290 | align_size >> PAGE_SHIFT, pgmap); | |
291 | percpu_ref_get_many(pgmap->ref, pfn_end(pgmap) - pfn_first(pgmap)); | |
e8d51348 | 292 | |
a95c90f1 DW |
293 | error = devm_add_action_or_reset(dev, devm_memremap_pages_release, |
294 | pgmap); | |
295 | if (error) | |
296 | return ERR_PTR(error); | |
e8d51348 | 297 | |
41e94a85 | 298 | return __va(res->start); |
9476df7d DW |
299 | |
300 | err_add_memory: | |
0207df4f AR |
301 | kasan_remove_zero_shadow(__va(align_start), align_size); |
302 | err_kasan: | |
9049771f DW |
303 | untrack_pfn(NULL, PHYS_PFN(align_start), align_size); |
304 | err_pfn_remap: | |
bcfa4b72 MW |
305 | pgmap_array_delete(res); |
306 | err_array: | |
d8668bb0 CH |
307 | pgmap->ops->kill(pgmap); |
308 | pgmap->ops->cleanup(pgmap); | |
9476df7d | 309 | return ERR_PTR(error); |
41e94a85 | 310 | } |
808153e1 | 311 | EXPORT_SYMBOL_GPL(devm_memremap_pages); |
4b94ffdc | 312 | |
2e3f139e DW |
313 | void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap) |
314 | { | |
315 | devm_release_action(dev, devm_memremap_pages_release, pgmap); | |
316 | } | |
317 | EXPORT_SYMBOL_GPL(devm_memunmap_pages); | |
318 | ||
4b94ffdc DW |
319 | unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) |
320 | { | |
321 | /* number of pfns from base where pfn_to_page() is valid */ | |
322 | return altmap->reserve + altmap->free; | |
323 | } | |
324 | ||
325 | void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns) | |
326 | { | |
327 | altmap->alloc -= nr_pfns; | |
328 | } | |
329 | ||
0822acb8 CH |
330 | /** |
331 | * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn | |
332 | * @pfn: page frame number to lookup page_map | |
333 | * @pgmap: optional known pgmap that already has a reference | |
334 | * | |
832d7aa0 CH |
335 | * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap |
336 | * is non-NULL but does not cover @pfn the reference to it will be released. | |
0822acb8 CH |
337 | */ |
338 | struct dev_pagemap *get_dev_pagemap(unsigned long pfn, | |
339 | struct dev_pagemap *pgmap) | |
340 | { | |
0822acb8 CH |
341 | resource_size_t phys = PFN_PHYS(pfn); |
342 | ||
343 | /* | |
832d7aa0 | 344 | * In the cached case we're already holding a live reference. |
0822acb8 | 345 | */ |
832d7aa0 | 346 | if (pgmap) { |
e7744aa2 | 347 | if (phys >= pgmap->res.start && phys <= pgmap->res.end) |
832d7aa0 CH |
348 | return pgmap; |
349 | put_dev_pagemap(pgmap); | |
0822acb8 CH |
350 | } |
351 | ||
352 | /* fall back to slow path lookup */ | |
353 | rcu_read_lock(); | |
bcfa4b72 | 354 | pgmap = xa_load(&pgmap_array, PHYS_PFN(phys)); |
0822acb8 CH |
355 | if (pgmap && !percpu_ref_tryget_live(pgmap->ref)) |
356 | pgmap = NULL; | |
357 | rcu_read_unlock(); | |
358 | ||
359 | return pgmap; | |
360 | } | |
e7638488 | 361 | EXPORT_SYMBOL_GPL(get_dev_pagemap); |
7b2d55d2 | 362 | |
e7638488 | 363 | #ifdef CONFIG_DEV_PAGEMAP_OPS |
e7638488 | 364 | void __put_devmap_managed_page(struct page *page) |
7b2d55d2 JG |
365 | { |
366 | int count = page_ref_dec_return(page); | |
367 | ||
368 | /* | |
369 | * If refcount is 1 then page is freed and refcount is stable as nobody | |
370 | * holds a reference on the page. | |
371 | */ | |
372 | if (count == 1) { | |
373 | /* Clear Active bit in case of parallel mark_page_accessed */ | |
374 | __ClearPageActive(page); | |
375 | __ClearPageWaiters(page); | |
376 | ||
c733a828 | 377 | mem_cgroup_uncharge(page); |
7b2d55d2 | 378 | |
80a72d0a | 379 | page->pgmap->ops->page_free(page); |
7b2d55d2 JG |
380 | } else if (!count) |
381 | __put_page(page); | |
382 | } | |
31c5bda3 | 383 | EXPORT_SYMBOL(__put_devmap_managed_page); |
e7638488 | 384 | #endif /* CONFIG_DEV_PAGEMAP_OPS */ |