]>
Commit | Line | Data |
---|---|---|
5981690d DW |
1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | /* Copyright(c) 2015 Intel Corporation. All rights reserved. */ | |
9476df7d | 3 | #include <linux/radix-tree.h> |
7d3dcf26 | 4 | #include <linux/device.h> |
92281dee | 5 | #include <linux/types.h> |
34c0fd54 | 6 | #include <linux/pfn_t.h> |
92281dee DW |
7 | #include <linux/io.h> |
8 | #include <linux/mm.h> | |
41e94a85 | 9 | #include <linux/memory_hotplug.h> |
5042db43 JG |
10 | #include <linux/swap.h> |
11 | #include <linux/swapops.h> | |
e7638488 | 12 | #include <linux/wait_bit.h> |
92281dee | 13 | |
9476df7d DW |
14 | static DEFINE_MUTEX(pgmap_lock); |
15 | static RADIX_TREE(pgmap_radix, GFP_KERNEL); | |
16 | #define SECTION_MASK ~((1UL << PA_SECTION_SHIFT) - 1) | |
17 | #define SECTION_SIZE (1UL << PA_SECTION_SHIFT) | |
18 | ||
ab1b597e | 19 | static unsigned long order_at(struct resource *res, unsigned long pgoff) |
9476df7d | 20 | { |
ab1b597e DW |
21 | unsigned long phys_pgoff = PHYS_PFN(res->start) + pgoff; |
22 | unsigned long nr_pages, mask; | |
eb7d78c9 | 23 | |
ab1b597e DW |
24 | nr_pages = PHYS_PFN(resource_size(res)); |
25 | if (nr_pages == pgoff) | |
26 | return ULONG_MAX; | |
27 | ||
28 | /* | |
29 | * What is the largest aligned power-of-2 range available from | |
30 | * this resource pgoff to the end of the resource range, | |
31 | * considering the alignment of the current pgoff? | |
32 | */ | |
33 | mask = phys_pgoff | rounddown_pow_of_two(nr_pages - pgoff); | |
34 | if (!mask) | |
35 | return ULONG_MAX; | |
36 | ||
37 | return find_first_bit(&mask, BITS_PER_LONG); | |
38 | } | |
39 | ||
40 | #define foreach_order_pgoff(res, order, pgoff) \ | |
41 | for (pgoff = 0, order = order_at((res), pgoff); order < ULONG_MAX; \ | |
42 | pgoff += 1UL << order, order = order_at((res), pgoff)) | |
43 | ||
5042db43 JG |
44 | #if IS_ENABLED(CONFIG_DEVICE_PRIVATE) |
45 | int device_private_entry_fault(struct vm_area_struct *vma, | |
46 | unsigned long addr, | |
47 | swp_entry_t entry, | |
48 | unsigned int flags, | |
49 | pmd_t *pmdp) | |
50 | { | |
51 | struct page *page = device_private_entry_to_page(entry); | |
52 | ||
53 | /* | |
54 | * The page_fault() callback must migrate page back to system memory | |
55 | * so that CPU can access it. This might fail for various reasons | |
56 | * (device issue, device was unsafely unplugged, ...). When such | |
57 | * error conditions happen, the callback must return VM_FAULT_SIGBUS. | |
58 | * | |
59 | * Note that because memory cgroup charges are accounted to the device | |
60 | * memory, this should never fail because of memory restrictions (but | |
61 | * allocation of regular system page might still fail because we are | |
62 | * out of memory). | |
63 | * | |
64 | * There is a more in-depth description of what that callback can and | |
65 | * cannot do, in include/linux/memremap.h | |
66 | */ | |
67 | return page->pgmap->page_fault(vma, addr, page, flags, pmdp); | |
68 | } | |
69 | EXPORT_SYMBOL(device_private_entry_fault); | |
70 | #endif /* CONFIG_DEVICE_PRIVATE */ | |
71 | ||
77dd66a3 | 72 | static void pgmap_radix_release(struct resource *res, unsigned long end_pgoff) |
ab1b597e DW |
73 | { |
74 | unsigned long pgoff, order; | |
9476df7d DW |
75 | |
76 | mutex_lock(&pgmap_lock); | |
77dd66a3 JS |
77 | foreach_order_pgoff(res, order, pgoff) { |
78 | if (pgoff >= end_pgoff) | |
79 | break; | |
ab1b597e | 80 | radix_tree_delete(&pgmap_radix, PHYS_PFN(res->start) + pgoff); |
77dd66a3 | 81 | } |
9476df7d | 82 | mutex_unlock(&pgmap_lock); |
ab1b597e DW |
83 | |
84 | synchronize_rcu(); | |
9476df7d DW |
85 | } |
86 | ||
e7744aa2 | 87 | static unsigned long pfn_first(struct dev_pagemap *pgmap) |
5c2c2587 | 88 | { |
e7744aa2 LG |
89 | const struct resource *res = &pgmap->res; |
90 | struct vmem_altmap *altmap = &pgmap->altmap; | |
5c2c2587 DW |
91 | unsigned long pfn; |
92 | ||
93 | pfn = res->start >> PAGE_SHIFT; | |
e7744aa2 | 94 | if (pgmap->altmap_valid) |
5c2c2587 DW |
95 | pfn += vmem_altmap_offset(altmap); |
96 | return pfn; | |
97 | } | |
98 | ||
e7744aa2 | 99 | static unsigned long pfn_end(struct dev_pagemap *pgmap) |
5c2c2587 | 100 | { |
e7744aa2 | 101 | const struct resource *res = &pgmap->res; |
5c2c2587 DW |
102 | |
103 | return (res->start + resource_size(res)) >> PAGE_SHIFT; | |
104 | } | |
105 | ||
949b9325 DW |
106 | static unsigned long pfn_next(unsigned long pfn) |
107 | { | |
108 | if (pfn % 1024 == 0) | |
109 | cond_resched(); | |
110 | return pfn + 1; | |
111 | } | |
112 | ||
5c2c2587 | 113 | #define for_each_device_pfn(pfn, map) \ |
949b9325 | 114 | for (pfn = pfn_first(map); pfn < pfn_end(map); pfn = pfn_next(pfn)) |
5c2c2587 | 115 | |
e8d51348 | 116 | static void devm_memremap_pages_release(void *data) |
41e94a85 | 117 | { |
e7744aa2 | 118 | struct dev_pagemap *pgmap = data; |
e8d51348 | 119 | struct device *dev = pgmap->dev; |
e7744aa2 | 120 | struct resource *res = &pgmap->res; |
9476df7d | 121 | resource_size_t align_start, align_size; |
71389703 DW |
122 | unsigned long pfn; |
123 | ||
e7744aa2 | 124 | for_each_device_pfn(pfn, pgmap) |
71389703 | 125 | put_page(pfn_to_page(pfn)); |
9476df7d | 126 | |
5c2c2587 DW |
127 | if (percpu_ref_tryget_live(pgmap->ref)) { |
128 | dev_WARN(dev, "%s: page mapping is still live!\n", __func__); | |
129 | percpu_ref_put(pgmap->ref); | |
130 | } | |
131 | ||
41e94a85 | 132 | /* pages are dead and unused, undo the arch mapping */ |
9476df7d | 133 | align_start = res->start & ~(SECTION_SIZE - 1); |
10a0cd6e JS |
134 | align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE) |
135 | - align_start; | |
b5d24fda | 136 | |
f931ab47 | 137 | mem_hotplug_begin(); |
e7744aa2 LG |
138 | arch_remove_memory(align_start, align_size, pgmap->altmap_valid ? |
139 | &pgmap->altmap : NULL); | |
f931ab47 | 140 | mem_hotplug_done(); |
b5d24fda | 141 | |
9049771f | 142 | untrack_pfn(NULL, PHYS_PFN(align_start), align_size); |
77dd66a3 | 143 | pgmap_radix_release(res, -1); |
e7744aa2 LG |
144 | dev_WARN_ONCE(dev, pgmap->altmap.alloc, |
145 | "%s: failed to free all reserved pages\n", __func__); | |
9476df7d DW |
146 | } |
147 | ||
4b94ffdc DW |
148 | /** |
149 | * devm_memremap_pages - remap and provide memmap backing for the given resource | |
150 | * @dev: hosting device for @res | |
e8d51348 | 151 | * @pgmap: pointer to a struct dev_pgmap |
4b94ffdc | 152 | * |
5c2c2587 | 153 | * Notes: |
e8d51348 CH |
154 | * 1/ At a minimum the res, ref and type members of @pgmap must be initialized |
155 | * by the caller before passing it to this function | |
156 | * | |
157 | * 2/ The altmap field may optionally be initialized, in which case altmap_valid | |
158 | * must be set to true | |
159 | * | |
160 | * 3/ pgmap.ref must be 'live' on entry and 'dead' before devm_memunmap_pages() | |
161 | * time (or devm release event). The expected order of events is that ref has | |
71389703 DW |
162 | * been through percpu_ref_kill() before devm_memremap_pages_release(). The |
163 | * wait for the completion of all references being dropped and | |
164 | * percpu_ref_exit() must occur after devm_memremap_pages_release(). | |
5c2c2587 | 165 | * |
e8d51348 | 166 | * 4/ res is expected to be a host memory range that could feasibly be |
5c2c2587 DW |
167 | * treated as a "System RAM" range, i.e. not a device mmio range, but |
168 | * this is not enforced. | |
4b94ffdc | 169 | */ |
e8d51348 | 170 | void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap) |
41e94a85 | 171 | { |
ab1b597e | 172 | resource_size_t align_start, align_size, align_end; |
e8d51348 CH |
173 | struct vmem_altmap *altmap = pgmap->altmap_valid ? |
174 | &pgmap->altmap : NULL; | |
949b9325 | 175 | struct resource *res = &pgmap->res; |
ab1b597e | 176 | unsigned long pfn, pgoff, order; |
9049771f | 177 | pgprot_t pgprot = PAGE_KERNEL; |
949b9325 | 178 | int error, nid, is_ram; |
15d36fec | 179 | struct dev_pagemap *conflict_pgmap; |
5f29a77c DW |
180 | |
181 | align_start = res->start & ~(SECTION_SIZE - 1); | |
182 | align_size = ALIGN(res->start + resource_size(res), SECTION_SIZE) | |
183 | - align_start; | |
15d36fec DJ |
184 | align_end = align_start + align_size - 1; |
185 | ||
186 | conflict_pgmap = get_dev_pagemap(PHYS_PFN(align_start), NULL); | |
187 | if (conflict_pgmap) { | |
188 | dev_WARN(dev, "Conflicting mapping in same section\n"); | |
189 | put_dev_pagemap(conflict_pgmap); | |
190 | return ERR_PTR(-ENOMEM); | |
191 | } | |
192 | ||
193 | conflict_pgmap = get_dev_pagemap(PHYS_PFN(align_end), NULL); | |
194 | if (conflict_pgmap) { | |
195 | dev_WARN(dev, "Conflicting mapping in same section\n"); | |
196 | put_dev_pagemap(conflict_pgmap); | |
197 | return ERR_PTR(-ENOMEM); | |
198 | } | |
199 | ||
d37a14bb LT |
200 | is_ram = region_intersects(align_start, align_size, |
201 | IORESOURCE_SYSTEM_RAM, IORES_DESC_NONE); | |
41e94a85 CH |
202 | |
203 | if (is_ram == REGION_MIXED) { | |
204 | WARN_ONCE(1, "%s attempted on mixed region %pr\n", | |
205 | __func__, res); | |
206 | return ERR_PTR(-ENXIO); | |
207 | } | |
208 | ||
209 | if (is_ram == REGION_INTERSECTS) | |
210 | return __va(res->start); | |
211 | ||
e8d51348 | 212 | if (!pgmap->ref) |
5c2c2587 DW |
213 | return ERR_PTR(-EINVAL); |
214 | ||
4b94ffdc | 215 | pgmap->dev = dev; |
4b94ffdc | 216 | |
9476df7d DW |
217 | mutex_lock(&pgmap_lock); |
218 | error = 0; | |
ab1b597e DW |
219 | |
220 | foreach_order_pgoff(res, order, pgoff) { | |
ab1b597e | 221 | error = __radix_tree_insert(&pgmap_radix, |
e7744aa2 | 222 | PHYS_PFN(res->start) + pgoff, order, pgmap); |
9476df7d DW |
223 | if (error) { |
224 | dev_err(dev, "%s: failed: %d\n", __func__, error); | |
225 | break; | |
226 | } | |
227 | } | |
228 | mutex_unlock(&pgmap_lock); | |
229 | if (error) | |
230 | goto err_radix; | |
231 | ||
41e94a85 CH |
232 | nid = dev_to_node(dev); |
233 | if (nid < 0) | |
7eff93b7 | 234 | nid = numa_mem_id(); |
41e94a85 | 235 | |
9049771f DW |
236 | error = track_pfn_remap(NULL, &pgprot, PHYS_PFN(align_start), 0, |
237 | align_size); | |
238 | if (error) | |
239 | goto err_pfn_remap; | |
240 | ||
f931ab47 | 241 | mem_hotplug_begin(); |
24e6d5a5 | 242 | error = arch_add_memory(nid, align_start, align_size, altmap, false); |
f1dd2cd1 MH |
243 | if (!error) |
244 | move_pfn_range_to_zone(&NODE_DATA(nid)->node_zones[ZONE_DEVICE], | |
245 | align_start >> PAGE_SHIFT, | |
a99583e7 | 246 | align_size >> PAGE_SHIFT, altmap); |
f931ab47 | 247 | mem_hotplug_done(); |
9476df7d DW |
248 | if (error) |
249 | goto err_add_memory; | |
41e94a85 | 250 | |
e7744aa2 | 251 | for_each_device_pfn(pfn, pgmap) { |
5c2c2587 DW |
252 | struct page *page = pfn_to_page(pfn); |
253 | ||
d77a117e DW |
254 | /* |
255 | * ZONE_DEVICE pages union ->lru with a ->pgmap back | |
256 | * pointer. It is a bug if a ZONE_DEVICE page is ever | |
257 | * freed or placed on a driver-private list. Seed the | |
258 | * storage with LIST_POISON* values. | |
259 | */ | |
260 | list_del(&page->lru); | |
5c2c2587 | 261 | page->pgmap = pgmap; |
e8d51348 | 262 | percpu_ref_get(pgmap->ref); |
5c2c2587 | 263 | } |
e8d51348 CH |
264 | |
265 | devm_add_action(dev, devm_memremap_pages_release, pgmap); | |
266 | ||
41e94a85 | 267 | return __va(res->start); |
9476df7d DW |
268 | |
269 | err_add_memory: | |
9049771f DW |
270 | untrack_pfn(NULL, PHYS_PFN(align_start), align_size); |
271 | err_pfn_remap: | |
9476df7d | 272 | err_radix: |
77dd66a3 | 273 | pgmap_radix_release(res, pgoff); |
9476df7d | 274 | return ERR_PTR(error); |
41e94a85 CH |
275 | } |
276 | EXPORT_SYMBOL(devm_memremap_pages); | |
4b94ffdc DW |
277 | |
278 | unsigned long vmem_altmap_offset(struct vmem_altmap *altmap) | |
279 | { | |
280 | /* number of pfns from base where pfn_to_page() is valid */ | |
281 | return altmap->reserve + altmap->free; | |
282 | } | |
283 | ||
284 | void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns) | |
285 | { | |
286 | altmap->alloc -= nr_pfns; | |
287 | } | |
288 | ||
0822acb8 CH |
289 | /** |
290 | * get_dev_pagemap() - take a new live reference on the dev_pagemap for @pfn | |
291 | * @pfn: page frame number to lookup page_map | |
292 | * @pgmap: optional known pgmap that already has a reference | |
293 | * | |
832d7aa0 CH |
294 | * If @pgmap is non-NULL and covers @pfn it will be returned as-is. If @pgmap |
295 | * is non-NULL but does not cover @pfn the reference to it will be released. | |
0822acb8 CH |
296 | */ |
297 | struct dev_pagemap *get_dev_pagemap(unsigned long pfn, | |
298 | struct dev_pagemap *pgmap) | |
299 | { | |
0822acb8 CH |
300 | resource_size_t phys = PFN_PHYS(pfn); |
301 | ||
302 | /* | |
832d7aa0 | 303 | * In the cached case we're already holding a live reference. |
0822acb8 | 304 | */ |
832d7aa0 | 305 | if (pgmap) { |
e7744aa2 | 306 | if (phys >= pgmap->res.start && phys <= pgmap->res.end) |
832d7aa0 CH |
307 | return pgmap; |
308 | put_dev_pagemap(pgmap); | |
0822acb8 CH |
309 | } |
310 | ||
311 | /* fall back to slow path lookup */ | |
312 | rcu_read_lock(); | |
e697c5b9 | 313 | pgmap = radix_tree_lookup(&pgmap_radix, PHYS_PFN(phys)); |
0822acb8 CH |
314 | if (pgmap && !percpu_ref_tryget_live(pgmap->ref)) |
315 | pgmap = NULL; | |
316 | rcu_read_unlock(); | |
317 | ||
318 | return pgmap; | |
319 | } | |
e7638488 | 320 | EXPORT_SYMBOL_GPL(get_dev_pagemap); |
7b2d55d2 | 321 | |
e7638488 DW |
322 | #ifdef CONFIG_DEV_PAGEMAP_OPS |
323 | DEFINE_STATIC_KEY_FALSE(devmap_managed_key); | |
31c5bda3 | 324 | EXPORT_SYMBOL(devmap_managed_key); |
e7638488 DW |
325 | static atomic_t devmap_enable; |
326 | ||
327 | /* | |
328 | * Toggle the static key for ->page_free() callbacks when dev_pagemap | |
329 | * pages go idle. | |
330 | */ | |
331 | void dev_pagemap_get_ops(void) | |
332 | { | |
333 | if (atomic_inc_return(&devmap_enable) == 1) | |
334 | static_branch_enable(&devmap_managed_key); | |
335 | } | |
336 | EXPORT_SYMBOL_GPL(dev_pagemap_get_ops); | |
337 | ||
338 | void dev_pagemap_put_ops(void) | |
339 | { | |
340 | if (atomic_dec_and_test(&devmap_enable)) | |
341 | static_branch_disable(&devmap_managed_key); | |
342 | } | |
343 | EXPORT_SYMBOL_GPL(dev_pagemap_put_ops); | |
344 | ||
345 | void __put_devmap_managed_page(struct page *page) | |
7b2d55d2 JG |
346 | { |
347 | int count = page_ref_dec_return(page); | |
348 | ||
349 | /* | |
350 | * If refcount is 1 then page is freed and refcount is stable as nobody | |
351 | * holds a reference on the page. | |
352 | */ | |
353 | if (count == 1) { | |
354 | /* Clear Active bit in case of parallel mark_page_accessed */ | |
355 | __ClearPageActive(page); | |
356 | __ClearPageWaiters(page); | |
357 | ||
358 | page->mapping = NULL; | |
c733a828 | 359 | mem_cgroup_uncharge(page); |
7b2d55d2 JG |
360 | |
361 | page->pgmap->page_free(page, page->pgmap->data); | |
362 | } else if (!count) | |
363 | __put_page(page); | |
364 | } | |
31c5bda3 | 365 | EXPORT_SYMBOL(__put_devmap_managed_page); |
e7638488 | 366 | #endif /* CONFIG_DEV_PAGEMAP_OPS */ |